Tom Perkins
Co-founded Kleiner & Perkins; backed and chaired Tandem and Genentech, with firm and founder attribution preserved.
A technical operator becomes an investor, 1932–1971
Thomas James Perkins was born in Oak Park, Illinois, in 1932 and grew up in White Plains, New York. He earned an electrical-engineering degree from MIT and an MBA from Harvard Business School, and entered technology through Hewlett-Packard rather than securities underwriting. The record supports that sequence; it does not support a family-background explanation for his later investment method. [researcher inference · strong] Computer History Museum biography Bancroft/NVCA oral history, printed pp. 5–10
At HBS, Perkins took Georges Doriot's Manufacturing course and later remembered its emphasis on making a product, selling it, satisfying customers, and entrusting money to leaders. That is evidence of remembered influence, not a contemporaneous syllabus-to-deal causal chain. [investor-stated · moderate] Bancroft/NVCA, printed pp. 5–7
HP supplied the more observable apprenticeship. Perkins began in the machine shop, moved through sales and marketing, briefly left for Booz Allen and an HP-backed optics company, returned, helped administer HP Laboratories, and eventually became the first general manager of HP's computer division. His recollection places product, manufacturing, sales, organization, and capital allocation in one operating career; its later effect is plausible but retrospective. [investor-stated · moderate] Bancroft/NVCA, printed pp. 10–27
University Laboratories is the strongest pre-venture causal bridge. With David Packard's permission, Perkins used household savings to develop a practical laser while remaining at HP, supervised the company, and later combined it with Spectra-Physics. This was direct experience with invention, scarce capital, recruiting, oversight, and a corporate outcome. Capitalization, equity split, consideration, taxes, and personal proceeds remain unavailable. [investor-stated · moderate] CHM oral history, pp. 5–7
The defensible formative claim is narrow: operating experience gave Perkins a vocabulary for interrogating technical and organizational risk and credibility with technical founders. It did not prove selection skill, confer expertise in every sector, or guarantee sound judgment outside company formation. [researcher inference · moderate] operator-to-investor account 2014 domain-transfer control
Building Kleiner & Perkins: partnership, incentives, and scarce capital, 1972–1974
Sandy Robertson introduced Perkins to Fairchild co-founder Eugene Kleiner; they formed Kleiner & Perkins in 1972. Perkins later reported an $8 million first partnership, about $4 million deployed, and the remainder returned. The public record conflicts between 17 and 18 first-fund investments, and Perkins later questioned the higher-sounding count. No executed agreement, audited schedule, capital account, or complete company roster is public. [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–35 count reconciliation
Perkins said the original plan contained an “Investor's Bill of Rights”: no personal investment the fund could make, independent audits, immediate profit distribution, and separate follow-on vehicles. The artifact was missing in his 2009 account and reportedly recovered by 2011, but the public corpus still does not expose it. The rules are evidence of stated conflict design, not inspected contract terms or proof that K&P caused industry adoption. [investor-stated · moderate] Bancroft/NVCA, printed pp. 33–35 and 40 CHM, pp. 17–18
The vehicle plausibly reinforced an involved practice: a small private partnership could make concentrated commitments, use boards and follow-on signaling as practical control, and distribute rather than warehouse gains. Yet fund duration, reserves, fees, liquidity policy, geographic authority, voting rights, and Fund I carry are not public. Perkins's later recollection of a move from 20% to 30% carry cannot be projected backward onto Fund I. [researcher inference · strong] vehicle reconstruction
Tandem, Snow-Job, and the first test of the method, 1973–1977
Tandem shows the method in its best decision-near form. Jim Treybig supplied the customer problem and company drive; Jack Loustaunou, Jim Katzman, and Mike Green supplied operating and technical work. Perkins sponsored the effort and chaired it; Kleiner consented; Brook Byers performed diligence; Pitch Johnson supplied outside capital and board work. The division supports substantial Perkins credit without converting sponsorship into sole founding. [retrospective witness · strong] Byers oral history, printed pp. 9–11 Johnson oral history, printed pp. 39–45
The financing sequence is more important than the celebrated outcome. A decision-near account reports roughly $50,000 for recruiting and architecture work, then $1 million after the design persuaded investors, followed by a reported $500,000 K&P participation in a larger syndicate while the machine remained a prototype. Capital bought information before it bought scale. Tandem went public in 1977 and later merged into Compaq, but the public record does not provide a complete K&P ownership-to-distribution chain. [documented behavior · strong] BusinessWeek, article pp. 64–65 Tandem decision lab
Snow-Job/Advanced Recreation Equipment prevents a clean success narrative. The company sold a motorcycle-to-snowmobile conversion; Perkins later said the fuel shock damaged distribution and conceded that even success would not have produced a venture-scale company. A later secondary table reports cost and zero residual value, but no audited loss chain is public. The case exposes two errors: exogenous shock and a maximum-upside test that should have been asked before the shock. [researcher inference · moderate] WBUR, 00:12:35–00:15:43 Snow-Job decision lab
These paired cases suggest a process: name the decisive uncertainty, buy a small discriminating proof, and require the opportunity to matter if proof succeeds. Tandem supports the first two steps; Snow-Job challenges the third. [researcher inference · strong] Tandem reconstruction Snow-Job reconstruction
Genentech: staging science without absorbing founder credit, 1976–1980
Bob Swanson and Herbert Boyer founded Genentech around recombinant-DNA commercialization; university scientists, external laboratories, City of Hope, licensing institutions, regulators, Eli Lilly, employees, and later executives were part of the production system. K&P supplied early capital and governance, and Perkins became chair. Calling Perkins the founder would erase commercial, scientific, institutional, and operating authors. [researcher inference · strong] Genentech founders Perkins Genentech oral history, printed pp. 3–10
The strongest supported investment contribution was to reject a fully integrated-company budget as the first step and finance contracted molecular proof. Somatostatin evidence in 1977 preceded human-insulin work in 1978, external partnership, and a broader organization. This is a falsifiable allocation mechanism: the first check should answer a named scientific question whose result changes whether and how the next check is made. [researcher inference · strong] Genentech oral history, printed pp. 4–9 Genentech decision lab
The record also limits the story. Recalled initial amounts of $100,000, $200,000, and $250,000 may refer to different moments, but no public security-and-date chain resolves them. The 1980 prospectus places a 14.5% pre-offering holding in the Kleiner & Perkins entity, not Perkins personally. Fred Middleton's account disputes some later financial-structure credit. K&P cash flows, dilution, distributions, carry, and Perkins's personal proceeds remain unavailable. [researcher inference · strong] Genentech prospectus, certain-shareholders table oral-history attribution boundary
Genentech's 1980 IPO established a major company outcome, not an audited partner return. It supports the proposition that staged scientific proof could open capital and partnership options; it cannot establish the size of Perkins's alpha or separate selection from access, founders, science, market regime, and a powerful outlier. [researcher inference · strong] 1980 prospectus performance boundary
From partnership craft to institution, 1978–1995
Tandem's 1991 proxy identifies Perkins as general partner of the original K&P partnership and KPCB I–IV, formed in 1978, 1980, 1982, and 1986. By 1984, three partnerships had reportedly raised about $220 million and invested in 86 companies. The institution had outgrown the founding vehicle; these are firm and vehicle facts, not evidence that Perkins selected or governed every company. [contemporaneous record · strong] Tandem 1991 proxy 1984 JEC hearing, p. 166
Frank Caufield and Brook Byers widened capital, sourcing, diligence, and healthcare capability. Hybritech is principally a Byers-sponsored company; Perkins's supported roles are approval, mentorship, and board service. Collagen has a clearer Perkins association but incomplete terms. LSI Logic and VLSI Technology expose collective syndication and contested governance rather than a single KPCB picker. [retrospective witness · strong] Byers on Hybritech formation, printed pp. 15–21 Byers on Perkins's mentorship and shared board, printed p. 28 attribution controls
Acuson is the useful ordinary-success test. Sam Maslak supplied the technical insight and operating work; Perkins interrogated acoustic diffraction, supported financing, helped recruit a team, and later disputed product sequencing. A secondary reconstruction reports a $2.5 million 1982 round split with KPCB and a later bridge, while SEC records establish the 2000 Siemens tender. KPCB ownership and proceeds are missing, so the case supports technical diligence and governance more strongly than economics. [researcher inference · moderate] CHM, pp. 20–21 Acuson decision lab
Perkins described board work—recruiting, financing, crisis handling, and follow-on signaling—as venture capital's non-cash product. By 1993 he said he had reduced a peak load of seventeen boards to four. The workload is retrospective, but issuer records and cases show board intensity was part of the model. They do not show that every intervention helped or that founders experienced the relationship similarly. [researcher inference · moderate] Red Herring, board-role and workload answers
Later governance: where the model traveled—and where it did not, 1996–2016
Perkins joined Compaq's board after Tandem's acquisition and supported the Compaq–HP transaction. KPCB's history attributes original Compaq sourcing to Sevin Rosen and sponsorship to John Doerr, so Perkins's defensible contribution is later governance and M&A, not the original investment. His later “we started Compaq” formulation warns against allowing firm identity or board service to overwrite deal attribution. [researcher inference · strong] KPCB Compaq history WBUR, 00:37:22–00:38:53
At HP in 2005–06, Perkins participated in the process that removed Carly Fiorina, later revised his judgment, and resigned during the board-leak dispute. The SEC chronology says his immediate objection favored private handling and opposed forcing George Keyworth's resignation; his later inquiry uncovered pretexting. HP, not Perkins, was respondent in the SEC disclosure order. The episode supports costly dissent and governance attention, but neither resignation nor vindication narrative proves that his preferred process was correct. [researcher inference · strong] SEC order, ¶¶4–15 HP corrective 8-K, Item 8.01
The 2014 wealth controversy is a harder domain-boundary test. Perkins compared criticism of wealthy Americans with Nazi persecution of Jews, later called the historical term a poor choice, and separately suggested tax-weighted voting in remarks later described as provocative or unserious. KPCB publicly distanced itself and stated he had not been involved in years. These records bear on rhetoric, historical proportionality, and public judgment; they do not alter an investment return or justify a general inference of misconduct. [researcher inference · strong] New Yorker, January 27, 2014 TIME, February 14, 2014
Perkins died in June 2016 at 84. Obituaries commonly attach Amazon, Google, Netscape, and other later KPCB successes to his name because it remained in the brand. Those are legacy associations unless person-, vehicle-, and decision-level evidence establishes his role. [researcher inference · strong] Fortune obituary firm/person boundary
People, capital, and influence map
| Layer | Actors | Supported contribution | Attribution boundary |
|---|---|---|---|
| Formation | Doriot; Packard; Hewlett; HP teams | Remembered manufacturing/customer frame plus technical, sales, operating, and governance apprenticeship. [researcher inference · moderate] Bancroft/NVCA, printed pp. 5–27 | Mentorship and institutional access are not Perkins-created assets; no course file proves a rule-to-deal chain. |
| Personal capital | Perkins household; University Laboratories; Spectra-Physics | Household capital and supervision preceded the fund and supplied operating experience. [investor-stated · moderate] CHM, pp. 5–7 | Not K&P performance; capitalization and personal economics unknown. |
| First partnership | Kleiner; Perkins; LPs; Robertson; Hillman interests | Robertson introduced partners and helped fundraising; Kleiner and Perkins formed and governed; LPs bore vehicle risk. [retrospective witness · moderate] Bancroft/NVCA, printed pp. 28–37 | Exact LPs, votes, GP economics, and company allocations incomplete. |
| Tandem | Treybig; Loustaunou; Green; Katzman; Kleiner; Perkins; Byers; Johnson; employees | Founders/operators built; Perkins sponsored/chair; Byers diligenced; Kleiner approved; Johnson co-invested/governed. [researcher inference · strong] role map | No sole-founder credit; ownership, dilution, exit influence, and distributions incomplete. |
| Genentech | Swanson; Boyer; Cohen; scientists; labs; K&P; Lilly; Middleton; employees | Founders/scientists created; K&P financed; Perkins chaired; partners and executives licensed, manufactured, commercialized, and financed. [researcher inference · strong] role map | No lone-founder or personal-return claim; some financial-structure credit disputed. |
| Later KPCB | Caufield; Byers; Doerr; later partners; LPs | Expanded sourcing, specialist judgment, capital, governance, and succession. [researcher inference · strong] 1984 JEC hearing, p. 166 | Firm-logo wins and later funds are not Perkins-personal decisions or economics. |
| Public boards | HP, Compaq, News Corp directors, executives, shareholders, regulators | Evidence of intervention, dissent, and judgment under institutional constraints. [researcher inference · strong] HP/SEC note | A board seat is not sourcing, investment ownership, or fund return. |
A falsifiable thesis of the edge
The favorable thesis is deliberately specific: from the early 1970s through the early 1990s, in technically uncertain company formation, Perkins's plausible repeatable advantage was to translate operator knowledge into a named-risk sequence—identify a potentially important product, finance the smallest decisive technical or commercial proof, keep burn below the unresolved risk, and then use a close board and referral network to recruit missing capabilities and release more capital. [researcher inference · moderate] Tandem and Genentech provide the strongest support; Acuson provides a less famous partial replication. Tandem Genentech Acuson
The thesis predicts that unseen Perkins-led files should contain a specific pre-investment bottleneck, a cheap milestone whose result changes financing, low burn before proof, explicit role allocation, and identifiable recruiting, financing, or board interventions after proof. Across a complete portfolio, cases following that sequence should outperform matched opportunities after controlling for sector, stage, vehicle, access, syndicate, and founder quality. [researcher inference · moderate] risk-decomposition account, WBUR 00:23:23–00:31:38
The thesis would weaken or fail if contemporaneous memos show no stable named-risk/staging practice; other partners supplied analysis and interventions attributed to Perkins; ordinary staged cases performed no better than unstaged peers; board work was ceremonial or harmful; results disappear after removing Tandem and Genentech; or complete cash flows show weak net performance. Apple is already a mechanism-level falsifier: Perkins rejected the intake meeting because he saw little technical risk and treated marketing risk as difficult to stage, so the screen prevented information collection that might have revealed an exception. [researcher inference · strong] Bancroft/NVCA, printed p. 61 Apple decision lab
Causal chain audit: experience to outcome
| Formative input | Proposed belief | Observable process | Action | Outcome and boundary |
|---|---|---|---|---|
| HP product, manufacturing, sales, and management work | A technical proposition must become a manufacturable, sellable organization. [researcher inference · moderate] Bancroft/NVCA, pp. 16–27 | Interrogate product mechanism and team gaps. | Tandem architecture review; Acuson physical-mechanism inquiry and recruiting. | Both became public companies; causality and fund economics incomplete. ledger |
| University Laboratories built with household capital | Early capital should buy evidence before overhead. [researcher inference · moderate] CHM, pp. 5–7 | Stage proof, then recruit and scale. | Tandem recruiting/design step; Genentech contracted proof. | Strong behavior evidence in two outliers, weak denominator evidence elsewhere. portfolio forensics |
| Doriot and Packard mentorship, as remembered | Importance and customer value matter more than novelty alone. [investor-stated · moderate] Wharton, opening and “Getting the Right Answers” | Ask whether success can be large and decisive risk financeable. | Tandem/Genentech selected; Snow-Job later judged too small. | Snow-Job supports the lesson retrospectively, not proof it was applied ex ante. failure lab |
| Private LP with close boards and follow-on signaling | Governance and aligned allocation can be more valuable than cash alone. [investor-stated · moderate] Bancroft/NVCA, pp. 33–35 and 62–63 | Conflict rules, audits, boards, recruiting, financing intervention. | Tandem/Genentech chairs; later HP dissent. | Conduct supported; original conflict document and comparative board value missing. writings |
| Dense founder/executive network | Prior operators can improve referrals and diligence. [investor-stated · moderate] WBUR, 00:25:51–00:26:58 | Source through trusted portfolio executives. | KPCB company network expanded. | Access visible; it may exclude outsiders and is not selection skill by itself. talks |
The favorable record and strongest skeptical case
The favorable reading is that Perkins helped convert venture capital from passive risk money into technical company formation: small proofs, deliberately recruited teams, active boards, and explicit conflict controls. Tandem and Genentech are unusually clear examples in computing and biotechnology, and Acuson shows the method outside the two canonical winners. His willingness to name Snow-Job, Apple, and satellite-insurance luck makes the later teaching record more useful than a victory list. [researcher inference · moderate] three ranked views spoken synthesis
The skeptical reading is at least as important. The visible record may be dominated by two outliers and outcome-selected memory; Perkins recalled seven or eight other Fund I positions near break-even plus failures. Only 15 public candidate paths were researched, the 17-versus-18 first-fund denominator remains unresolved, and only five cases support useful decision reconstruction. No complete fund or person cash-flow series exists. The record cannot establish personal hit rate, net return, or repeatable alpha after excluding Tandem and Genentech. [researcher inference · strong] searched denominator return boundaries
Access, team, vehicle, era, and luck are not residual footnotes. HP and Fairchild networks improved entry and credibility; Kleiner, Byers, Caufield, Johnson, founders, scientists, operators, LPs, and co-investors supplied judgment and labor; scarce early capital and publicly supported science created opportunities; the LP form enabled concentrated private governance; and the satellite rescue was explicitly called luck. These forces can coexist with skill, but the evidence cannot allocate their marginal contribution. [researcher inference · strong] attribution matrix historical context
Apple shows the edge's shadow: a screen optimized for technical formation rejected an opportunity before a founder/product meeting because apparent risk was commercial rather than technical. Snow-Job shows that product function is insufficient when distribution, macro sensitivity, and maximum market size are wrong. HP and 2014 show technical and board confidence does not automatically transfer to every governance, historical, or political judgment. [researcher inference · strong] Apple lab Snow-Job lab HP control
Evidence that would decide between the theses is concrete: the complete Fund I roster and cash flows; original memos and partnership votes; securities, ownership, dilution, follow-ons, and distributions; board minutes showing intervention timing; ordinary and failed files; and matched cases led by other partners. Until those surface, “great technical company builder with major shared wins” is better supported than “audited top-tier personal investor.” [researcher inference · strong] retrieval ledger
Vehicle, market, and regulatory context
| Era / structure | Supported context | Behavioral implication and limit |
|---|---|---|
| University Laboratories, mid-1960s | Household capital financed a technical operating company while Perkins remained at HP. [investor-stated · moderate] CHM, pp. 5–7 | Personal exposure may sharpen burn discipline; no counterfactual tests it. |
| K&P first partnership, 1972 | Small private LP, reported $8 million commitment and about $4 million deployment; documents absent. [investor-stated · moderate] Bancroft/NVCA, pp. 31–35 | Scarce capital and close boards plausibly favored staging; duration, reserves, carry, authority, and liquidity unknown. |
| 1970s market and policy | Startup capital was scarce; 1974 shock preceded later capital-gains and pension-rule changes. [researcher inference · moderate] 1978 Congressional Record, p. 15697 | Later abundance must not be projected backward; Snow-Job's shock is outcome evidence. |
| Biotechnology system | University science, licensing, external labs, FDA pathways, and pharmaceutical partners mattered alongside capital. [researcher inference · strong] Genentech founders 1980 prospectus | Investor skill cannot absorb science, founder invention, regulation, or partner capabilities. |
| KPCB I–IV and scale | Separate successor partnerships and expanding partners replaced the founding vehicle. [contemporaneous record · strong] Tandem 1991 proxy | Firm brand, network size, and later returns cannot be allocated to Perkins without deal evidence. |
Comparison with completed peers
Only three completed profiles are close enough to clarify the boundary; Eugene Kleiner and Don Valentine are excluded because their packages are unfinished. [researcher inference · strong] source-map comparison boundary
| Peer | Why valid | What it clarifies |
|---|---|---|
| Arthur Rock | Same early technology ecosystem, overlapping founders and boards, but a nontechnical financier emphasizing people, structure, and governance. [researcher inference · strong] Rock biography | Perkins's proposed difference is technical risk decomposition. Rock's eventual Apple investment after a negative first impression also shows how reversible intake could outperform refusal to meet. |
| Pitch Johnson | Same-era company formation and direct Tandem co-investor whose testimony separates lead, diligence, capital, and board roles. [retrospective witness · strong] Johnson on Tandem | Tandem was a syndicate/team achievement; Johnson also provides a people-plus-market contrast to technology-risk-first screening. |
| Peter Crisp | Same 1932 cohort and early technology/healthcare investing, but inside Rockefeller family capital and Venrock. [researcher inference · strong] Crisp vehicle history | The contrast isolates vehicle/access: Perkins's LP faced different allocation and succession questions; both depended on specialists and founders. |
The comparison is qualitative. These profiles lack matched audited cash flows and complete personal denominators, so no relative return ranking is supportable. [researcher inference · strong] Perkins denominator
Sourced chronology
| Date | Turning point | Evidence boundary |
|---|---|---|
| 1932 | Born in Oak Park; grew up in White Plains | Birthplace and upbringing supported; family-to-method causation unavailable. [contemporaneous record · moderate] CHM |
| 1950s | MIT BSEE; HBS MBA; Doriot course | Degrees supported; influence retrospective. [investor-stated · moderate] Bancroft/NVCA, pp. 5–7 |
| 1957–1971 | HP machine shop, marketing, labs, computer division | Career sequence mainly participant testimony. [investor-stated · moderate] Bancroft/NVCA, pp. 10–27 |
| mid-1960s | University Laboratories; later Spectra-Physics combination | Operating role supported; economics unavailable. [investor-stated · moderate] CHM, pp. 5–7 |
| 1972 | Kleiner & Perkins; reported $8 million partnership | Documents and 17/18 roster missing. [investor-stated · moderate] Bancroft/NVCA, pp. 31–35 |
| 1973–1977 | Snow-Job; Tandem formation and 1977 IPO | Tandem decision evidence stronger than Snow-Job memory. [researcher inference · moderate] Tandem lab |
| 1976–1980 | Genentech formation, proofs, Lilly, IPO | Chronology supported; seed/fund cash flows unresolved. [documented behavior · strong] Genentech lab |
| 1978–1986 | KPCB I–IV successor partnerships | Legal roles supported; economics absent. [contemporaneous record · strong] Tandem proxy |
| early 1980s–2000 | Acuson investment through Siemens tender | Governance partly retrospective; transaction public; proceeds unknown. [researcher inference · moderate] Acuson |
| 1993 | Peak seventeen-board load reportedly reduced to four | Self-report; exact denominator absent. [investor-stated · moderate] Red Herring |
| 1997–2002 | Tandem–Compaq and Compaq–HP governance | Later board role supported; original Compaq selection elsewhere. [researcher inference · strong] KPCB |
| 2005–2007 | Fiorina removal, HP leak dispute, resignation, SEC order | Party narratives conflict; order controls and is against HP. [contemporaneous record · strong] SEC |
| 2007–2013 | Memoir and retrospective teaching | Rich direct voice, decades after key decisions. [researcher inference · strong] spoken corpus |
| 2014 | Wealth analogy and tax-weighted-vote remarks | Authenticated fragments; seriousness partly ambiguous. [investor-stated · moderate] source note |
| 2016 | Death at age 84 | Date/age supported; obituary attributions bounded. [contemporaneous record · moderate] Fortune |
Learning path through the package
Read this biography first for chronology, causality, attribution, and competing edge theses. Then use:
- Writings for conflict rules, HP's written record, memoir boundary, and the 2014 contradiction.
- Talks for the 15-family direct-voice corpus, staging doctrine, forecasts, and retrospective contradictions.
- Investments for the denominator, three Pareto views, uniform ledger, return boundaries, and five no-hindsight labs.
- Investment philosophy for the integrated operating system and modern translation.
- Mental models for reusable tests, falsifiers, and misuse warnings.
- Sources for the 68-route source map, access boundaries, and unresolved archives.
For active study, decide before revealing each outcome: Tandem, Genentech, Acuson, Apple, and Snow-Job. The last two test whether the proposed edge gathers disconfirming information and prices nontechnical risk before hindsight supplies the answer.
1. Snapshot and testable edge thesis
Snapshot. Tom Perkins was an MIT-trained electrical engineer, an HBS graduate who studied Georges Doriot's manufacturing course, an HP operator, and a founder-operator before he became a venture capitalist. The public record supports a career built around technically difficult, board-intensive company formation; it does not support a complete, audited personal or fund return series. [retrospective witness · moderate] (Bancroft/NVCA oral history, printed pp. 5–27; return boundary and searched denominator)
Testable edge thesis. From the formation of Kleiner & Perkins in 1972 through his active company-building years, Perkins's plausible repeatable edge was operator-technologist pattern recognition converted into staged uncertainty reduction and unusually active governance. The proposed mechanism was: identify a discontinuity large enough to matter; name the decisive technical, market, financing, or regulatory risk; fund the smallest proof that discriminates success from failure; recruit and scale after evidence; then use board position, follow-on signaling, and operating work to change the outcome. [researcher inference · moderate] (talk synthesis; decision labs)
Scope. The thesis is strongest for early-stage U.S. computing, biotechnology, and medical-device businesses in an era of scarce private capital, technically testable risks, concentrated boards, and long holding periods. It should not be generalized without fresh evidence to passive growth investing, liquid securities, geographically dispersed portfolios, or businesses whose decisive uncertainties cannot be cheaply tested. [researcher inference · moderate] (historical context)
Predictions on unseen cases. If the thesis is real, decision-near records from ordinary as well as famous investments should show (1) an explicitly named decisive uncertainty, (2) modest capital before its discriminating proof, (3) hiring and follow-on capital gated by evidence, (4) material board interventions attributable to Perkins, and (5) better outcomes than matched same-vintage opportunities available through comparable networks. [researcher inference · provisional] (searched denominator)
Falsifiers. The thesis weakens materially if original memoranda show these steps were reconstructed after success; ordinary and failed cases do not exhibit the same process; Tandem and Genentech explain most economic value; founder or partner work explains the interventions attributed to Perkins; access and era effects dominate selection; or complete fund cash flows show no matched-cohort advantage. [researcher inference · provisional] (searched denominator)
2. Core philosophy
Own words first
Across public appearances, Perkins described venture investing as beginning with an important opportunity rather than a preassembled team, treating risk as the source of opportunity, and requiring the individual who sponsors a deal to remain accountable for the decision. He also described board service and crisis work—not capital alone—as the venture capitalist's consequential product. These are his stated principles, not audited evidence that he applied them consistently. [investor-stated · moderate] (Red Herring 1993 interview mirror; Wharton 2012 edited report; WBUR, 06:58–07:57 and 23:23–31:38)
His strongest compact rule was that an opportunity already stripped of meaningful risk was probably too late for venture returns. Read literally, that is incomplete: risk creates no value by itself. The investable condition is a large possible payoff plus uncertainty that can be named, tested, financed in stages, and survived. [investor-stated · moderate] (Wharton 2012 edited report)
Synthesized doctrine
The operating doctrine that best fits the record is:
- Importance before polish. Seek an order-of-magnitude improvement or a market-opening discontinuity rather than a merely competent product.
- Opportunity before complete team. Establish that the opportunity is real, then recruit missing capabilities around it.
- Convert ambiguity into tests. Separate technical, market, financing, regulatory, and organizational risks; buy evidence against the decisive one first.
- Stage irreversible commitments. Scale burn, team, and manufacturing only when evidence changes the probability distribution.
- Treat governance as production. Recruiting, conflict resolution, financing, and crisis decisions are part of the investment product.
- Preserve attribution and the denominator. Separate founder, partner, firm, vehicle, and luck contributions; retain passes, losses, and ordinary outcomes.
The first five propositions synthesize Perkins's accounts and decision records; the sixth is an analyst control required because the public corpus is winner-heavy and attribution is frequently blurred. [researcher inference · moderate] (talk synthesis; investment evidence boundary)
3. Sourcing
| Channel | What the record supports | Repeatability and bias |
|---|---|---|
| Operator network | HP, Fairchild-linked, founder, executive, and technical relationships created proprietary referrals and fast technical context. Tandem came through a network in which Jimmy Treybig, Pitch Johnson, Gene Kleiner, and Perkins had distinct roles. [retrospective witness · moderate] | Repeatable for a well-embedded operator; difficult for a newcomer to copy. It also concentrates access by geography, employer pedigree, and social proximity. (Pitch Johnson oral history, printed pp. 39–45; Brook Byers oral history, printed pp. 9–10) |
| Portfolio referral | Existing founders and portfolio executives surfaced adjacent opportunities and talent. Compaq, for example, reached Kleiner Perkins through Sevin Rosen and was sponsored inside the firm by John Doerr; Perkins's later governance role must not be mistaken for origination. [documented behavior · moderate] | A compounding advantage once the portfolio works; also a source-concentration risk and a common path to person-versus-firm misattribution. (Kleiner Perkins, Compaq history) |
| Technical and university adjacency | Perkins's engineering and operating background improved access to technically legible propositions; Genentech connected academic science with a company-building vehicle. [researcher inference · moderate] | Transferable only with domain fluency or trusted expert access. The record does not show a systematic university-sourcing program. (Genentech founders; Genentech oral history, printed pp. 3–10) |
| Direct or cold approaches | The corpus contains much less evidence about cold inbound, broad outbound search, or systematic market mapping than about referrals. [researcher inference · insufficient public record] | Do not infer that absence in the surviving stories means the channel was unused. (source-map coverage) |
Sourcing rule. Record the route before evaluating quality: introducer, prior relationship, geography, institution, partner sponsor, competing access, and whether the same route repeatedly feeds the portfolio. Then compare referral-sourced decisions with cold or weak-tie decisions rather than treating network density as selection skill. [researcher inference · moderate] (source-map coverage)
4. Picking
Perkins's public picking algorithm is better represented as a sequence than as a static score:
| Step | Decision question | Evidence and boundary |
|---|---|---|
| 1. Maximum-upside test | If the product works, is the improvement or market creation large enough to support a venture-scale result? | Perkins used an order-of-magnitude frame in the Wharton report. It is a stated heuristic, not a measured cutoff. [investor-stated · moderate] (Wharton 2012 edited report) |
| 2. Decisive-risk test | Which single uncertainty most changes the decision now: mechanism, market need, regulation, financing, manufacturing, or people? | Tandem and Genentech allow partial reconstruction; original investment memoranda are not public. [researcher inference · moderate] (Tandem case; Genentech case) |
| 3. Smallest-proof test | What is the cheapest milestone that would distinguish a real opportunity from an attractive story? | Early Genentech work and staged company formation fit the model, but causation should not be assigned to Perkins alone. [researcher inference · moderate] (Genentech S-1, Business and Products, printed pp. 115–121) |
| 4. Team-in-sequence test | Who is essential before the proof, and which capabilities should be added only after it? | Perkins explicitly put opportunity before a complete team. The surviving record does not establish a universal hiring sequence. [investor-stated · moderate] (Red Herring 1993 interview mirror) |
| 5. Governance test | Can the investor acquire legitimate influence, recruit missing operators, finance milestones, and act under crisis? | Strong as a stated philosophy and visible in signature cases; weaker as a portfolio-wide causal claim. [researcher inference · moderate] (Bancroft/NVCA oral history, printed pp. 62–63) |
| 6. Price and ownership test | Does the entry price and ownership leave enough payoff after dilution, reserves, time, and loss probability? | Public evidence is much thinner than the opportunity-and-risk rhetoric. Exact entry ownership, dilution, reserve policy, and realized cash flows remain incomplete. [researcher inference · insufficient public record] (return-claims record) |
Selection boundary. A large technological discontinuity is not enough. The investable claim also needs a plausible buyer, a path through manufacturing or regulation, financing survivability, a team capable of learning, and terms that leave asymmetric payoff. The Apple pass illustrates a false negative; Snow Job illustrates how personal enthusiasm and an optimistic industry narrative can outrun disciplined underwriting. [researcher inference · moderate] (Apple pass; Snow Job)
5. Deal and portfolio mechanics
| Mechanic | Reconstructed practice | What remains unknown |
|---|---|---|
| Initial commitment | Commit enough to reach the next discriminating proof, especially when the decisive uncertainty is technical. [researcher inference · moderate] | Portfolio-wide initial-check distribution and written milestone covenants. (Genentech decision lab) |
| Follow-ons | The lead investor's willingness to continue can function as a powerful market signal and source of practical control. [investor-stated · moderate] (Bancroft/NVCA oral history, printed pp. 62–63) | Reserve ratios, pro-rata policy, exceptions, and the rate at which losers were stopped. |
| Ownership | Concentrated ownership and board involvement appear important, but exact ownership is available only episodically and must be tied to a named vehicle and date. [researcher inference · moderate] | A complete company-by-company ownership and dilution ledger. (return and ownership boundary) |
| Syndication | Relationships could provide domain help, price discovery, and later financing. Tandem also shows why sponsor, investor, board, and operating roles must be separated. [researcher inference · moderate] | Formal allocation rules and whether syndication improved outcomes versus merely widening access. (Pitch Johnson oral history, printed pp. 39–45) |
| Board and control | Formal votes, board seats, executive recruitment, and the shadow power of follow-on financing collectively shaped influence. [researcher inference · moderate] | Board minutes, dissent records, and portfolio-wide intervention frequency. (Bancroft/NVCA oral history, printed pp. 62–63) |
| Exit and stop | Public accounts emphasize IPOs and enduring winners more than sales, shutdowns, write-offs, or explicit kill criteria. [researcher inference · insufficient public record] | Complete holding periods, distributions, impairment decisions, and opportunity cost. (retrieval ledger) |
The economic 80/20 picture remains unresolved. Tandem and Genentech are the most defensible top-tier contributors in the public record, but no auditable cash-flow series establishes their rank as number one or two, their share of total value, or whether another company or vehicle belongs in the top tier. [researcher inference · moderate] (three Pareto views)
Portfolio construction is therefore only partially reconstructable. The evidence supports asymmetric-outcome intent, concentrated governance effort, and staged follow-ons; it does not support a precise loss ratio, reserve model, sector exposure, vintage attribution, or formula for portfolio count. Any modern replication should write these controls explicitly rather than import an inferred historical portfolio formula. [researcher inference · moderate] (portfolio-level forensics)
6. Vehicle and incentive fit
| Dimension | Historical evidence | Investment implication |
|---|---|---|
| Capital source and legal form | Perkins remembered the 1972 Kleiner & Perkins partnership as having about $8 million committed and about $4 million deployed; audited partnership records were not found. [retrospective witness · moderate] (Bancroft/NVCA oral history, printed pp. 31–35) | Finite outside capital made ownership, reserves, distributions, and conflicts part of the operating system. Exact LP composition and draw mechanics remain incomplete. |
| Duration and liquidity | The public record shows long company-building arcs and public-market exits, but not complete fund terms, extension provisions, distribution timing, or secondary liquidity. [researcher inference · insufficient public record] | Long technical development can fit patient capital only if fund duration and follow-on capacity match it. (vehicle evidence) |
| Fundraising pressure | Later partnerships existed, but the evidence reviewed does not quantify time spent fundraising or how fundraising altered decisions. [researcher inference · insufficient public record] | A claimed investment philosophy is incomplete unless fundraising and deployment incentives are modeled. (vehicle evidence) |
| Compensation and carry | Perkins recalled a shift from 20% to 30% carried interest around a later competing offer; the vehicle, hurdle, clawback, partner split, and documentary confirmation are unresolved. [retrospective witness · provisional] (Bancroft/NVCA oral history, printed pp. 36–37) | Higher carry may intensify asymmetric-outcome incentives, but the causal effect cannot be established from recollection alone. |
| Conflicts and distributions | Perkins described an “Investor's Bill of Rights” with restrictions on personal investing, prompt distributions, independent audits, and separate follow-on vehicles. The original 1972 document has not been recovered publicly, so compliance and exact language cannot be verified. [retrospective witness · provisional] (Bancroft/NVCA oral history, printed pp. 33–35; writings boundary) | Treat the ideas as reported governance design, not as proven contractual practice. |
| Governance and decision rights | Perkins emphasized individual sponsor accountability and lead-investor leverage; the record does not provide the full partnership voting agreement. [investor-stated · moderate] (Wharton 2012 edited report) | Underwriting accountability should identify sponsor, dissent, vetoes, and follow-on authority. |
| Geography and authority | Silicon Valley networks dominate the evidence. Formal geographic mandates and delegated authority are not public. [researcher inference · insufficient public record] | Separate historical network advantage from a portable geographic thesis. (vehicle evidence) |
| Regulation | Company-level biotechnology, medical-device, disclosure, and public-company rules mattered; fund-level regulatory design is not sufficiently documented here. [researcher inference · insufficient public record] | Regulatory risk belongs in the decisive-risk map, not in a generic “execution” bucket. (vehicle evidence) |
| Stakeholders and succession | KPCB added partners including Brook Byers and John Doerr while Perkins reduced active board work over time. Public records show a firm that outlived a founder, but not complete partner economics, succession votes, or key-person clauses. [researcher inference · moderate] (Brook Byers oral history; Kleiner Perkins, Compaq history) | Institutional continuity requires explicit role transfer, attribution, key-person, and conflict rules; brand continuity is not proof of process continuity. |
Modern fund comparison. The durable dimension is alignment among opportunity type, fund duration, reserve capacity, governance bandwidth, and incentives. It is not supportable to claim that the 1972 arrangement maps cleanly onto a current multistage platform, opportunity fund, SPV, evergreen vehicle, or corporate venture unit without examining the current agreement. [researcher inference · moderate] (vehicle evidence)
7. Team attribution and work with founders
The unit of analysis must be the role, not the legend:
| Case | Founder and operator roles | Investor and partner roles | Attribution boundary |
|---|---|---|---|
| Tandem | Jimmy Treybig and the operating team built the company; the decision-near record describes the non-stop computing proposition and organization. [contemporaneous record · strong] | Pitch Johnson, Gene Kleiner, Perkins, and others occupied different introduction, sponsorship, financing, and governance roles. [retrospective witness · moderate] | Do not collapse the outcome into “a Perkins deal.” (BusinessWeek artifact, physical PDF pp. 110–111; Tandem attribution note) |
| Genentech | Herbert Boyer and Robert Swanson were the founders; scientists, managers, and employees converted recombinant-DNA science into products and an organization. [contemporaneous record · strong] | Perkins contributed financing and governance; Brook Byers and others also had consequential roles. [researcher inference · moderate] | The S-1 supports roles and risks, not a single-person causal story. (Genentech founders; 1980 S-1 management, printed pp. 132–133; Genentech case) |
| Compaq | Founders and management built the company. [documented behavior · moderate] | Sevin Rosen referred the opportunity; John Doerr sponsored it inside Kleiner Perkins; Perkins later contributed governance. [documented behavior · moderate] | Origination, sponsorship, ownership, board work, and outcome are separate claims. (Kleiner Perkins, Compaq history) |
| Hybritech | Founder, scientific, and operating work was shared across the company. [retrospective witness · moderate] | Brook Byers was the central Kleiner Perkins partner in the surviving account. [retrospective witness · strong] | Firm portfolio membership does not establish Perkins-led selection or intervention. (Byers on formation, printed pp. 15–21; Byers on Perkins's mentorship and shared board, printed p. 28) |
Founder-work model. The supportable Perkins approach was high-contact: test the mechanism, recruit missing executives, preserve enough board authority to resolve crises, and use financing decisions to force explicit choices. Its legitimate form requires consent, role clarity, documented dissent, and an agreed escalation path. Without those controls, “active governance” can become founder displacement, shadow control, or retrospective credit capture. [researcher inference · moderate] (people, capital, and influence map)
8. Views on the craft
Perkins treated venture capital as a company-building craft rather than a capital-allocation service. That claim has four observable implications:
- Technical fluency changes diligence. An operator-investor should interrogate how the product works, which failure mode matters, and what proof would change the decision. [researcher inference · moderate] (Bancroft/NVCA oral history, printed pp. 10–27)
- The board is an operating forum. The job includes recruiting, financing, strategy, conflict, and crisis—not ceremonial oversight. [investor-stated · moderate] (Red Herring 1993 interview mirror)
- Judgment has an owner. Consensus should not erase the partner who made the call, the dissenting view, or the conditions for follow-on. [investor-stated · moderate] (Wharton 2012 edited report)
- Reputation compounds both access and error. Referrals improve information flow, while a celebrated portfolio can make later stories less skeptical and more founder-exclusive. [researcher inference · moderate] (sourcing synthesis)
These views are grounded in his interviews, but the craft must also be judged against contrary evidence: the Apple miss, Snow Job, an Acuson record with weaker public attribution, the HP investigation and disclosure process, and rhetoric that imposed reputational cost. [researcher inference · moderate] (talks synthesis; HP SEC order, paragraphs 4–15; HP corrective Form 8-K, Item 8.01)
Power law, fund size, access, and exits. The recurring focus on maximum upside and order-of-magnitude improvement is consistent with power-law selection, but no complete cash-flow denominator proves the actual distribution. Fund size should follow the opportunity perimeter and the capital required to buy decisive proofs, not a copied historical number. Access is a funnel input, not proof of selection skill. Exit evidence is thinner still: the record emphasizes IPO winners more than holding-period discipline, sales, shutdowns, or distributions. [researcher inference · moderate] (Pareto and denominator analysis; portfolio-level forensics)
What makes a great investor under this reconstruction. Greatness would require repeatable decision quality across wins, ordinary cases, losses, passes, and luck; legitimate improvement in company outcomes; disciplined vehicle economics; accurate role attribution; and durable conduct under governance stress. Fame, access, and a few gross company outcomes are insufficient. [researcher inference · strong] (searched denominator; attribution audit)
9. Signature bets and decision process
Use the cases as laboratories, not anecdotes:
| Case | Ex-ante decision problem | Process signal | What it can and cannot establish |
|---|---|---|---|
| Tandem | Could a fault-tolerant computing architecture create a large market, and could the team finance and build it? | Technical importance, staged company building, recruiting, and active governance. | Strong signature case; incomplete original terms, ownership, and complete cash-flow record. [documented behavior · strong] |
| Genentech | Could recombinant-DNA science be converted into reproducible products and a financeable company under scientific and regulatory uncertainty? | Separate scientific proof from organization scale; recruit after proof; govern financing and commercialization. | Strong primary company record; causal credit is shared and exact venture return remains unaudited. [documented behavior · strong] |
| Acuson | Could a new ultrasound architecture deliver a clinically meaningful improvement and survive capital/manufacturing risk? | Technical mechanism, milestone logic, and board involvement. | Useful but weaker: key reconstruction relies on a secondary synthesis of an inaccessible transcript; exact $1.25 million and role details are reported, not independently verified. [researcher inference · moderate] |
| Apple pass | Would an unfamiliar young company and personal-computing proposition justify engagement? | Shows the access-and-attention failure mode before diligence even begins. | A real false negative, not proof that the later winner was obviously investable at the time. [investor-stated · moderate] |
| Snow Job | Could an early K&P investment in a seasonal recreational product survive distribution, market-size, financing, and macro risk? | Demonstrates product enthusiasm and incomplete underwriting across technical, channel, market-size, and exact-vehicle boundaries. | A failure control; public evidence does not support a clean initial belief, exact K&P vehicle, complete terms, or loss calculation. [investor-stated · moderate] |
The ordinary-case evidence is less complete. Qume was brought to K&P by William Draper III of Sutter Hill in his surviving account, while Hybritech is Byers-led; their inclusion is useful precisely because it constrains Perkins-centric storytelling. [retrospective witness · moderate] (William Draper III oral history, printed p. 16; Byers on formation, printed pp. 15–21; Byers on Perkins's mentorship and shared board, printed p. 28)
10. Mental models and maxims
The full model definitions, evidence ladders, misuse conditions, falsifiers, and linked cases live in mental-models.md. The inventory below matches that file's nine substantive models; it is an index, not a second retelling. [researcher inference · strong]
| # | Model | Origin evidence | Short operating maxim and primary use |
|---|---|---|---|
| 1 | Seek order-of-magnitude improvement with venture-scale upside | Perkins stated an order-of-magnitude and risk frame. [investor-stated · moderate] (Wharton 2012 edited report) | Test maximum value before optimizing the base case; use for opportunity selection and payoff asymmetry. |
| 2 | Name the decisive risk and buy the smallest discriminating proof | Public risk remarks plus staged Genentech evidence. [researcher inference · moderate] (WBUR, 23:23–31:38; Genentech S-1, printed pp. 115–121) | Spend first on evidence that can change the decision; use for diligence and milestones. |
| 3 | Scale capital and build missing capabilities only after proof | Opportunity-before-team statements and signature cases. [researcher inference · moderate] (Red Herring 1993 interview mirror; Genentech decision lab) | Sequence burn and hiring behind learning; use for financing, recruiting, and follow-ons. |
| 4 | Use operator and technical fluency to interrogate the mechanism | Engineering, HP, and founder experience precede the investing work; causation remains inferred. [researcher inference · moderate] (Bancroft/NVCA oral history, printed pp. 5–27) | Ask how the system fails, not only how the pitch succeeds; use for technical diligence. |
| 5 | Make board service crisis work, with explicit formal and shadow decision rights | Board-work doctrine and lead-investor follow-on power. [investor-stated · moderate] (Red Herring 1993 interview mirror; Bancroft/NVCA oral history, printed pp. 62–63) | Map votes, financing power, information, and escalation before crisis; use for governance. |
| 6 | Use referral networks as filters, then audit access and source concentration | Participant and portfolio records show referral density and attribution risk. [researcher inference · moderate] (Pitch Johnson oral history, printed pp. 39–45; Kleiner Perkins, Compaq history) | Record who opened the door and which doors the network never sees; use for sourcing control. |
| 7 | Pursue asymmetric outcomes while preserving the loss, pass, luck, and ordinary denominator | Maximum-upside rhetoric sits beside incomplete cash flows, a pass, a loss account, and admitted luck. [researcher inference · strong] (return boundary; WBUR, 12:35–15:43 and 31:51–33:08) | Never let the winner erase the full decision set; use for portfolio learning. |
| 8 | Define the fund opportunity perimeter before quality is visible | Reported partnership and conflict rules, with the original 1972 artifact missing. [retrospective witness · provisional] (Bancroft/NVCA oral history, printed pp. 31–40; writings boundary) | Make mandate, ownership, reserve, conflict, and vehicle rules ex ante; use for vehicle design. |
| 9 | Attribute every outcome role by role before judging repeatability | Tandem, Compaq, Hybritech, and Qume expose distinct roles. [researcher inference · strong] (attribution audit) | Separate founder, sponsor, partner, firm, vehicle, and luck; use for performance attribution. |
The maxims are researcher-created compressions of the documented record, not verified Perkins quotations. [researcher inference · moderate] (full model field guide)
11. Evolution and contradictions
| Period | Observable evolution | Tension or contradiction |
|---|---|---|
| Pre-1972 operator and founder | Engineering, HP operating work, and University Laboratories created a mechanism-first, company-building frame. [retrospective witness · moderate] (CHM oral history, printed pp. 5–7) | Later stories can over-credit early operator experience because failed or mundane operating decisions are less visible. |
| 1972 partnership formation | Kleiner & Perkins paired operating reputations, finite outside capital, and reported conflict/distribution rules. [retrospective witness · moderate] (Bancroft/NVCA oral history, printed pp. 31–40) | The claimed 1972 governance artifact is not publicly recovered; remembered design cannot prove compliance. |
| Tandem and Genentech era | The signature record shows movement from technical assessment to recruitment, financing, and board-intensive scaling. [researcher inference · moderate] | Opportunity-first rhetoric coexisted with unusually powerful networks and exceptional founders; process and access are difficult to disentangle. (Tandem and Genentech labs) |
| KPCB institutionalization | New partners expanded sector reach and created a durable firm beyond its founders. [researcher inference · moderate] | Firm success increasingly cannot be assigned to Perkins, and later firm outcomes are not evidence of his personal selection edge. (team attribution) |
| Later governance and public role | Perkins continued to emphasize board judgment, risk, and strong individual accountability. [investor-stated · moderate] | HP's investigation and disclosure chronology shows that governance power without rigorous process can create serious institutional risk. (HP SEC order, paragraphs 4–15) |
| 2014 public rhetoric | He entered a public argument about wealth, taxes, and political treatment. [contemporaneous record · strong] | The remarks weakened the claim that elite access was balanced by stakeholder sensitivity and supplied a reputational counterexample to judgment. (New Yorker contemporaneous report; TIME contemporaneous report) |
Observed process-change boundary. The sources document institutional growth and role transfer, but not a reliable portfolio-wide process change after the Apple pass, Snow Job, Acuson dispute, HP episode, or 2014 criticism. Later retellings identify lessons; they do not prove that written selection, conflict, or governance procedures changed. [researcher inference · insufficient public record] (writings evolution; talks evolution)
12. Failures and limits
| Failure or limit | What it teaches | Evidence boundary |
|---|---|---|
| Apple pass | Access, attention, and social recognition fail before formal diligence; preserve declined-meeting and pass logs. | Perkins later recalled missing the meeting; the record does not establish that an investment was offered on investable terms. [retrospective witness · moderate] (Bancroft/NVCA oral history, printed p. 61; CHM panel, 01:08:03–01:08:43) |
| Snow Job | Product enthusiasm, distribution exposure, market-size error, and exact-vehicle ambiguity require explicit underwriting and portfolio controls. | A public interview supports a loss story, but not exact terms, a contemporaneous decision memo, or an audited return. [investor-stated · moderate] (WBUR, 12:35–15:43) |
| Satellite outcome described as luck | Maintain a luck ledger; favorable outcomes do not automatically validate process. | Perkins himself characterized the outcome as luck in an edited interview. [investor-stated · moderate] (WBUR, 31:51–33:08) |
| Acuson attribution boundary | Do not let a vivid board story substitute for an accessible primary transcript or a complete role map. | The key transcript is inaccessible in the reviewed public corpus; claims are downgraded accordingly. [researcher inference · insufficient public record] (Acuson note) |
| HP investigation and disclosure | Board influence needs lawful investigation design, privacy controls, independent oversight, and prompt disclosure. | The SEC order and HP corrective filing establish a public-company process failure; they do not prove every later characterization of Perkins's role. [contemporaneous record · strong] (SEC order, paragraphs 4–15; HP Form 8-K, Item 8.01) |
| Missing denominator | A selection philosophy cannot be economically validated from marquee winners. | Complete invested capital, distributions, write-offs, opportunity costs, and personal-versus-fund allocations are absent. [researcher inference · strong] (searched denominator) |
| Attribution drift | Portfolio-company fame often migrates from founders and lead partners to a famous senior investor or firm. | Hybritech, Qume, and Compaq show why company, firm, partner, and person ledgers must stay separate. [researcher inference · strong] (attribution audit) |
These are not side notes to a greatness narrative. They are the controls needed to decide whether the proposed edge was repeatable, portable, and net beneficial. [researcher inference · strong] (failure and anti-portfolio record)
13. Historical operating environment
| Era condition | How it could amplify Perkins's results | Why modern readers must adjust |
|---|---|---|
| Scarce organized venture capital | Operator reputation and a small number of trusted networks could create privileged access and strong financing leverage. [researcher inference · moderate] | In a market where capital is more abundant or faster, test whether exclusivity, price discovery, and a single lead's signaling power still behave the same way. [researcher inference · provisional] (historical context) |
| Concentrated Silicon Valley geography | Dense HP, Fairchild, Stanford, founder, and investor ties lowered search and reference costs. [researcher inference · moderate] | When teams or markets are geographically distributed, build explicit weak-tie discovery and remote-diligence systems rather than assuming the historical cluster advantage transfers. [researcher inference · provisional] (historical context) |
| Hardware and manufacturing intensity | Technical proof, reliability, and manufacturing milestones made staged risk reduction observable. [researcher inference · moderate] | For software or AI products that can ship before their full business is proved, separate product iteration from market, distribution, data-rights, and unit-economic evidence. [researcher inference · provisional] (Tandem decision lab) |
| Biotechnology's emerging commercial path | Public science could establish mechanism while company formation, IP, regulation, and financing remained uncertain. [contemporaneous record · strong] (Genentech S-1, risk factors, printed pp. 107–109) | Do not assume the 1976 platform maturity, regulation, capital intensity, or competitive clock; rebuild the risk sequence for the company under review. [researcher inference · provisional] |
| Long board tenures and concentrated governance | A lead investor could recruit, arbitrate, and finance through multiple company phases. [researcher inference · moderate] | Where syndicates, round cadence, independence requirements, information rights, or stakeholder scrutiny differ, write the actual decision rights instead of assuming historical lead power. [researcher inference · provisional] (Bancroft/NVCA oral history, printed pp. 62–63) |
| Less complete digital disclosure | Oral histories preserve insight but also permit winner selection, attribution drift, and chronology errors. [researcher inference · strong] | Treat versioned memos, pass logs, cap tables, board decisions, and cash-flow ledgers as the modern control; absence of the record should reduce confidence. [researcher inference · provisional] (searched denominator) |
Historical context is an explanatory variable, not a discount or an excuse. The correct test compares Perkins with similarly situated investors and opportunities from the same vintages, networks, sectors, and vehicles. [researcher inference · moderate] (historical context)
14. Modern VC translation (as of 2026-08-23)
The following translations are dated 2026-08-23. Each is an analyst-created application of the historical lesson, not a claim that Perkins addressed today's market.
| Major lesson | What transfers | Changed assumption and likely misuse | Observable signals and 2026 application |
|---|---|---|---|
| Order-of-magnitude importance | Start with maximum plausible customer or societal value. | A large demo improvement can be mistaken for durable willingness to pay, distribution, or defensibility. | Require a quantified before/after baseline, buyer evidence, retention or repeat use, and a path from product gain to venture-scale economics. For an AI product, separate benchmark lift from workflow adoption and gross-margin durability. [researcher inference · moderate] (model 1) |
| Name the decisive risk | Force the team and sponsor to identify the uncertainty that most changes the decision. | Teams may create long risk registers that avoid ranking, or label ordinary execution as the decisive risk. | IC memo names one primary risk, one rival hypothesis, one proof, a deadline, and a decision rule. [researcher inference · moderate] (model 2) |
| Buy the smallest discriminating proof | Spend for learning before scale. | Cheap software iteration may encourage premature growth spend; deep-tech investors may mistake a laboratory result for manufacturability. | Release capital against mechanism, customer, regulatory, manufacturing, and unit-economic milestones separately; do not let a proxy metric silently become the final proof. [researcher inference · moderate] (model 2) |
| Recruit after evidence, not after certainty | Add capability as the opportunity's bottleneck becomes visible. | Waiting for perfect proof can lose the market; hiring a complete executive team before product evidence locks in burn and politics. | Maintain a capability map tied to milestones: which hire is needed before the next proof, which after, and what evidence would change the sequence. [researcher inference · moderate] (model 3) |
| Interrogate the mechanism with relevant operator fluency | Domain experience can expose physical, manufacturing, customer, and organizational failure modes that a financial screen misses. | Operator confidence can become false authority when the sector, technology, or regulatory regime differs from prior experience. | Require the sponsor to state the competence boundary, name independent experts, reproduce the critical test, and document rival explanations. Analyst-created hypothetical: for AI infrastructure, pair systems expertise with independent evaluation of data rights, workload reliability, security, and unit economics rather than accepting a benchmark demo. [researcher inference · moderate] (model 4) |
| Make governance productive | Board work should improve recruiting, financing, strategic choices, and crisis response. | “Founder friendly” can become avoidance; “active” can become shadow management or coercion. | Write formal votes, information rights, follow-on discretion, escalation, independent investigation, founder transition, and dissent procedures before crisis. [researcher inference · moderate] (model 5) |
| Treat referrals as filters, not truth | Trusted networks reduce search and reference cost. | Network prestige can hide correlated access, pedigree bias, and repeated blind spots. | Dashboard source concentration by introducer, institution, geography, and partner; compare progression and outcomes for referrals, cold inbound, outbound, and weak ties. [researcher inference · moderate] (model 6) |
| Define vehicle perimeter ex ante | Match duration, reserves, ownership, and conflict rules to the opportunity. | Cross-fund investments, SPVs, opportunity vehicles, and personal holdings can create allocation and signaling conflicts not visible in a company story. | Every decision names vehicle, allocation rule, reserves, related interests, continuation rights, liquidity assumptions, and stakeholder approvals. [researcher inference · moderate] (model 8) |
| Preserve denominator, attribution, and luck | Learn from all decisions and credit the actual work. | Brand storytelling can treat gross company value as investor return and firm participation as personal authorship. | Maintain versioned pass/loss/luck logs; audited cash flows by vehicle and vintage; role-level attribution; PME and loss-ratio controls where appropriate; and explicit “reported” prefixes for unaudited figures. [researcher inference · strong] (model 7; model 9) |
The modern decision test is therefore not “Would Perkins invest?” It is: What is the maximum credible payoff, which uncertainty controls it, what is the smallest proof, who owns the judgment, which vehicle bears the risk, what governance rights are legitimate, and which evidence would stop the investment? [researcher inference · moderate] (full model field guide)
15. Comparison with completed peers
All three comparisons are dimension-valid and use completed peer dossiers. Arthur Rock is a same-era company-formation and concentrated-governance comparator; Pitch Johnson is a direct Tandem participant and relationship/syndication control; Peter Crisp is a same-era institutional-capital and Apple-adjacent control. These are analytical comparisons, not claims that the investors used identical terminology. [researcher inference · moderate] (Arthur Rock philosophy and evidence; Pitch Johnson oral history, printed pp. 39–45; Peter Crisp oral history, printed p. 16)
| Dimension | Tom Perkins | Arthur Rock | Pitch Johnson | Peter Crisp |
|---|---|---|---|---|
| Sourcing | Operator, technical, founder, and portfolio referrals; dense Silicon Valley access. | Trusted founder and executive networks, with a pronounced people-first reputation. | Relationship-rich syndication and local network development; directly relevant to the shared Tandem history. | Institutional and relationship sourcing through Rockefeller/Venrock networks and long-duration relationships. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
| Picking | Important discontinuity, decisive risk, smallest proof, then team and scale. | People, integrity, and reality testing carry greater explicit weight. | Entrepreneur, market, and syndicate judgment appear more relational and portfolio-system oriented. | People-and-market screening within an institutional capital context. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
| Ownership and portfolio | Concentrated governance and asymmetric winners; economic denominator unresolved. | Concentrated company formation, often across personal and partnership-capital eras. | More explicit portfolio and syndication development across a long investing career. | Patient institutional capital and specialist/team structure; Apple offers a useful shared-company comparison without identical roles. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
| Governance | Board work, recruiting, crisis, and follow-on signaling are central. | Direct founder partnership and concentrated governance are also central. | Long board relationships and collaborative syndication constrain single-person credit. | Long-horizon institutional governance and relationship continuity. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
| Vehicle and incentives | Early finite partnership; reported carry and conflict rules, but original terms and cash flows incomplete. | Personal and partnership capital must be separated by deal and era. | Vehicle evolution and partnership roles are important to attribution. | Rockefeller/Venrock capital source changes the stakeholder and duration comparison. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
| Failure control | Apple pass, Snow Job, HP process, luck admission, sparse ordinary denominator. | The peer dossier preserves misses and attribution limits rather than using only Intel/Apple-era fame. | Tandem role overlap makes Johnson a direct control on Perkins-centric attribution. | A different institutional route to company building controls for the claim that the Perkins model was the only successful form. [researcher inference · moderate] (Rock peer dossier; Johnson peer dossier; Crisp peer dossier) |
The most defensible comparative difference is not “technical investor versus nontechnical investor.” It is that Perkins made operator-technologist interrogation, staged technical proof, and board crisis power unusually explicit in his public doctrine. The strongest commonality is concentrated, relationship-driven company formation under scarce capital. Whether the Perkins variant produced superior risk-adjusted cash returns remains unproved without matched fund data. [researcher inference · moderate] (Arthur Rock comparison; Pitch Johnson participant evidence)
16. The strongest case against greatness
The skeptical case is substantial:
- The economic record is incomplete. There is no complete audited ledger of contributions, ownership, dilution, distributions, write-offs, fees, carry, opportunity cost, and personal-versus-fund allocation. Reported company values or headline multiples cannot substitute for realized cash flows. [researcher inference · strong] (searched denominator)
- The visible record is dominated by outliers. Tandem and Genentech may explain a large share of reputation and possibly economics; their exact share is unresolved. A philosophy inferred from winners can be elegant even when the full portfolio was ordinary. [researcher inference · moderate] (Pareto views)
- Access and era may explain much of the edge. HP experience, Silicon Valley density, scarce organized capital, and elite founder networks could create selection and governance power unavailable to a comparable outsider. [researcher inference · moderate] (historical context)
- Causal credit is divided. Founders, scientists, executives, Gene Kleiner, Brook Byers, John Doerr, Pitch Johnson, co-investors, employees, vehicles, and public markets contributed. Firm participation is not Perkins authorship. [researcher inference · strong] (attribution audit)
- The controls are adverse. Apple is a missed opportunity; Snow Job is a loss narrative; the satellite story is admitted luck; Acuson has a material access/attribution boundary; HP demonstrates governance-process risk; 2014 rhetoric demonstrates reputational and stakeholder judgment risk. [researcher inference · strong] (failure and anti-portfolio record)
- The doctrine may be retrospective compression. No complete set of original decision memos proves that maximum-upside, decisive-risk, smallest-proof, and team-sequencing rules were written and followed before outcomes were known. [researcher inference · moderate] (uncertainty ledger)
What would discriminate skill from legend? Recover complete fund and personal cash flows; original investment and pass memoranda; cap tables and vehicle allocations; board minutes and recruiting records; partner dissent and role maps; a matched set of ordinary and failed cases; founder, employee, and co-investor testimony independent of Perkins; and same-vintage controls adjusted for sector, geography, ownership, and access. Favorable results across those tests would strengthen the edge thesis. Failure to find the predicted ex-ante behavior or matched performance would falsify it. [researcher inference · strong] (uncertainty ledger)
The fair conclusion is conditional: Perkins belongs in the history of operator-led venture company building, and the public record supports several durable mechanisms. The public record does not permit a precise ranking of his economic performance or isolate how much came from repeatable judgment rather than partners, founders, access, outliers, and era. [researcher inference · strong] (return boundary)
17. Analyst study guide, glossary, and go deeper
Recommended learning path
- Start with _profile.md for identity, chronology, vehicles, role boundaries, and the evidence map.
- Read writings.md to separate authored artifacts, the unrecovered 1972 governance text, edited excerpts, and retrospective witness accounts.
- Read talks.md for the oral-source operating system and its access boundaries.
- Work through investments.md, beginning with the Pareto views and evidence boundary, then complete all five no-hindsight decision labs.
- Use the individual deal notes to inspect known, unknown, believed, disputed, terms, milestones, interventions, and outcomes without hindsight.
- Finish with mental-models.md, testing each model against its misuse condition and falsifier rather than memorizing the maxim.
Glossary
- Attribution ledger: A role-by-role record separating founder, introducer, sponsor, underwriter, board member, recruiter, firm, vehicle, syndicate, and luck.
- Decisive risk: The uncertainty whose resolution would most change the investment decision at the current stage.
- Discriminating proof: The smallest credible test that can distinguish the investment's key hypothesis from a live alternative explanation.
- Follow-on signaling power: Influence created when later investors and the company interpret an incumbent lead's willingness or refusal to invest as information.
- Investor's Bill of Rights: Perkins's retrospective label for reported 1972 conflict, audit, distribution, and follow-on rules; the original public artifact has not been recovered.
- Opportunity perimeter: The fund's ex-ante mandate across stage, sector, geography, check size, ownership, reserves, conflicts, duration, and adjacent vehicles.
- Reported return: A performance figure reproduced from a source but not recomputed from audited cash flows; it must remain prefixed “reported.”
- Searched denominator: A documented statement of which cash-flow, ownership, loss, pass, and portfolio records were sought, found, and still missing.
- Shadow decision right: Practical power arising from capital, information, relationships, or signaling rather than a formal vote.
Sourcing checklist
- Record the original route, introducer, prior relationship, institution, geography, and partner sponsor.
- Distinguish inbound, outbound, referral, portfolio, university, and cold channels.
- Measure concentration and progression by channel; inspect who never enters the funnel.
- Separate access advantage from selection advantage and role from authorship.
Diligence and investment-committee checklist
- State maximum credible upside and the assumptions that connect product improvement to economics.
- Freeze known, unknown, believed, and disputed facts at the decision date.
- Name the decisive risk, rival hypothesis, smallest proof, deadline, cost, and kill rule.
- Identify which team capability is needed now and which should wait for evidence.
- Record alternatives, price, security, ownership, dilution, reserves, follow-ons, and vehicle.
- Name sponsor, evaluators, dissent, conflicts, decision rights, and the evidence that would reverse the decision.
- Keep the pass, loss, ordinary, and luck record beside the winner case.
Governance checklist
- Map board votes, protective provisions, information rights, financing discretion, and shadow power.
- Agree recruiting, founder-transition, escalation, independent-investigation, disclosure, privacy, and conflict procedures before crisis.
- Tie follow-ons to named evidence; record deviations and who authorized them.
- Attribute each intervention to the people who performed it and test whether it changed the outcome.
- Define exit, distribution, impairment, and stop criteria at company and vehicle levels.
Best sources, in reading order
- Bancroft Library/NVCA oral history — broad first-person career and doctrine; valuable but retrospective.
- Genentech 1980 S-1 in the Middleton collection — decision-near company, risk, management, transaction, and ownership record.
- Computer History Museum Perkins oral history — operator and company-building detail with retrospective limitations.
- Decision-near Tandem BusinessWeek artifact, physical PDF pp. 110–111 — contemporaneous company context.
- Brook Byers oral history and Pitch Johnson oral history — participant controls on Perkins-centric attribution.
- Red Herring 1993 interview mirror and Wharton 2012 edited report — public doctrine; treat as edited access.
- WBUR interview — risk framework, failure, and luck controls with timestamps in the talk note.
- HP SEC order and HP corrective Form 8-K — primary governance and disclosure controls.
The unresolved research queue is explicit: original 1972 governance text; complete fund agreements and cash flows; deal-by-deal ownership and dilution; original decision and pass memoranda; portfolio-wide losses and ordinary outcomes; board minutes; independent founder and employee accounts; and matched peer data. Until those are found, preserve uncertainty rather than converting a coherent philosophy into a measured performance claim. [researcher inference · strong] (unresolved record; writings boundary)
Pareto 80/20 — what drove the record
This file reconstructs the public investment record around Tom Perkins without turning firm logos, company outcomes, or retrospective memories into audited personal returns. The 80/20 frame is a prioritization lens, not a literal percentage claim. Detailed case notes and the uniform evidence ledger sit in notes/; all links below are claim-local.
Directional economic contribution
No public cash-flow series supports a percentage allocation, TVPI, DPI, IRR, or partner-level return. The ranking is therefore directional and deliberately stops where ownership, dilution, distributions, and vehicle cash flows disappear.
| Rank | Candidate | Basis / measure | Attribution | Support / mechanism | Denominator caveat / case |
|---|---|---|---|---|---|
| Top tier (order unresolved) | Genentech | K&P held 14.5% immediately before the 1980 offering; secondary histories describe a very large marked outcome. | The stake belonged to K&P, not Perkins personally; founders, scientists, employees, and partners created company value. | [researcher inference · moderate] Ownership plus an IPO supports outlier status, not realized contribution. S-1, “Certain Shareholders,” printed pp. 137–138 | Dilution, distributions, sale timing, fees, carry, and taxes are unavailable. Case |
| Top tier (order unresolved) | Tandem Computers | Issuer records verify the 1977 offering; participant evidence identifies Tandem as the other exceptional Fund I result. | Founders and employees built Tandem; K&P, Perkins, Kleiner, Byers, Johnson, and co-investors supplied different capital and governance roles. | [researcher inference · moderate] A major company outcome supports outlier status, not a precise return rank. Tandem annual report, stock-price section | Initial ownership and the cash-flow chain through distributions are unavailable. Case |
| Residual | Residual Fund I portfolio | Perkins recalled seven or eight additional positions near break-even in aggregate, plus failures. | Partnership-level retrospective testimony; most company and sponsor identities remain unknown. | [retrospective witness · moderate] The recollection supports a concentrated shape, not company-level contribution. Bancroft/NVCA, printed pp. 38–40 | The 17-versus-18 denominator conflicts and residual economics cannot be ranked. Ledger |
Excluding either Genentech or Tandem leaves the other as the strongest visible contributor; the public record cannot order the pair. Excluding both leaves no public basis for claiming that the residual vehicle was exceptional. That is an outlier-sensitivity finding, not a calculated fund return. [researcher inference · strong] return-claim reconstruction
Historical importance
| Rank | Candidate | Basis / measure | Attribution | Support / mechanism | Denominator caveat / case |
|---|---|---|---|---|---|
| 1 | Genentech | Venture-backed biotechnology company and 1980 public-market milestone. | Swanson and Boyer founded it; Cohen–Boyer science, contract labs, employees, partners, and management were causal. | [researcher inference · strong] Founders S-1, appendix cover | Qualitative importance, not return contribution. Case |
| 2 | Tandem Computers | Durable fault-tolerant-computing franchise and staged company-building pattern. | Treybig, Green, Katzman, Loustaunou, Kleiner, Byers, co-investors, and employees share credit. | [researcher inference · strong] BusinessWeek, article pp. 64–65 | Qualitative importance, not return contribution. Case |
| 3 | Hybritech | Monoclonal-antibody diagnostics venture formation. | Primarily Byers-sponsored; Perkins supplied approval, mentorship, and board service. | [researcher inference · strong] Byers, printed pp. 37–42 | Attribution is shared and economics are unresolved. Ledger |
| 4 | Acuson | Differentiated medical imaging that reached a public-company acquisition. | Maslak supplied invention and operation; Perkins and KPCB supplied capital, recruitment, and governance. | [researcher inference · moderate] CHM, pp. 20–21 Siemens tender filing | Early chronology is substantially retrospective; KPCB economics are unknown. Case |
| 5 | University Laboratories | Pre-fund bridge from HP operator to startup owner and investor. | Perkins reports conception, household capital, and supervision; Packard and the operating team were also necessary. | [investor-stated · moderate] CHM, pp. 5–6 | Personal operating-company episode, not LP-fund performance. Ledger |
Decision-learning value
| Rank | Candidate | Basis / measure | Attribution | Support / mechanism | Denominator caveat / case |
|---|---|---|---|---|---|
| 1 | Genentech | Best decision record for staged scientific proof and partner-funded scale. | Shared among founders, scientists, management, K&P, institutions, and partners; Perkins supplied capital, board, and financing work. | [documented behavior · strong] S-1, “Business” and “Products,” printed pp. 115–121 | Initial terms and exact K&P economics remain unresolved. Case |
| 2 | Tandem Computers | Best decision record for staged architecture, financing, syndication, customer proof, and scale. | Sponsor, founder, technical, IC, co-investor, and operator roles are distinct. | [documented behavior · strong] BusinessWeek, article pp. 64–65 | Initial security and ownership remain unresolved. Case |
| 3 | Apple pass | Clearest intake-screen false negative before full diligence. | Perkins owns the non-meeting; other K&P decision rights are unknown. | [retrospective witness · moderate] CHM panel, 01:08:03–01:08:43 | Not a priced or voted IC pass. Case |
| 4 | Acuson | Technical-adjacency diligence, team construction, bridge financing, and later roadmap dissent. | Perkins narrates investor work; Maslak and employees supplied invention and operation. | [researcher inference · moderate] Evidence boundary | Key early chronology derives from a secondary reconstruction of an inaccessible transcript. Case |
| 5 | Snow-Job | Real failure testing market size, seasonality, distribution resilience, and exogenous-shock exposure. | Perkins later narrated the firm investment; sourcing, approval, and governance remain unknown. | [retrospective witness · moderate] WBUR, 00:12:35–00:15:43 | No decision-date memo, terms, company financials, or shutdown record. Case |
Evidence boundary and searched denominator
| Gate | Result |
|---|---|
| Publicly claimed or known | Fund I is described as 17 companies in a secondary historical reconstruction and 18 investments in a tertiary company history; Perkins later said 17 sounded too high. By 1984, KPCB and two successor partnerships testified to 86 companies. Later counts of 180, 450+, 500/525, and 1,000+ are firmwide, definition-incomplete, and span later partners and vehicles. [researcher inference · strong] count reconciliation |
| Candidate records searched | 15 company paths: 13 investments or leads, one missed opportunity, and one later-governance attribution control. [documented behavior · strong] uniform ledger |
| Candidate records researched | 15/15 reached at least one public source; several terminate at an honest access or attribution boundary. [documented behavior · strong] uniform ledger |
| Sufficient outcome attribution | University Laboratories, Tandem, Genentech, Hybritech, Acuson, and the Apple pass. [researcher inference · moderate] candidate analysis |
| Sufficient decision reconstruction | Tandem, Genentech, Acuson, Apple, and Snow-Job; only Tandem and Genentech have decision-near public records strong enough to constrain retrospective accounts. [researcher inference · strong] case notes |
| Unresolved | Exact Fund I roster, ordinary-company identities, vehicle assignments, complete fund cash flows, and most check/security/ownership/dilution fields. [researcher inference · strong] missing fields |
| Excluded | HP and News Corp governance; later firm-logo wins; the original Compaq selection as a Perkins-sponsored deal. A board seat is not an investment, and KP attributes original Compaq sourcing/sponsorship elsewhere. [researcher inference · strong] KP on Compaq |
| Coverage fraction | Not computable. A 15-company research list divided by 17 or 18 would mix unlike numerators and an unverified denominator. [researcher inference · strong] denominator note |
Historical market, regulation, and base rates
- K&P launched when institutional venture was small, private-company information asymmetry was severe, and the limited-partnership form enabled concentrated ownership, private governance, and flexible profit sharing. The same privacy that helped the form scale creates today's cash-flow evidence gap. [researcher inference · moderate] Federal Reserve staff study, pp. 11–23
- A 1981 contemporaneous account reported that roughly half of initial venture investments were written off and that firms without bank or public-market access depended heavily on private capital. That is a period base-rate control, not Fund I's measured loss ratio. [contemporaneous record · moderate] TIME, August 10, 1981
- The 1974 oil and market shock matters to Snow-Job and capital supply. Later capital-gains and pension-rule changes expanded the ecosystem after the earliest K&P decisions and must not be projected backward. [researcher inference · moderate] 1978 Congressional Record, p. 15697
- Genentech depended on federally supported university science, research institutions, licensing, regulators, contract labs, and pharmaceutical partners. Finance was one part of a production system. [researcher inference · strong] Genentech founders 1980 S-1
Vehicle economics and incentives
| Layer | What is public | What is not public |
|---|---|---|
| University Laboratories | Perkins says household capital financed the company while he remained at HP with Packard's permission. [investor-stated · moderate] CHM, pp. 5–6 | Capitalization, outside financing, equity split, merger consideration, taxes, and personal proceeds. |
| 1972 K&P partnership | Perkins recalled $8 million committed, roughly $4 million deployed, and the balance returned. [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–35 | Executed agreement, capital calls, GP commitment, fees, reserves, recycling, audited statements, waterfall, and liquidation. |
| Successor partnerships | Tandem's 1991 proxy identifies K&P (1972) and KPCB I–IV (1978, 1980, 1982, 1986) as separate general-partner roles. [contemporaneous record · strong] Tandem 1991 proxy | Company-to-vehicle map, commitments, ownership, cash flows, and partner carry shares. |
| Firm brand | By 1984 three partnerships had invested in 86 companies; later firm counts span many partners and vehicles. [contemporaneous record · strong] 1984 JEC hearing, p. 166 | Firm attribution cannot allocate sourcing, selection, governance, or economics to Perkins. |
| Company security | Issuer records verify boards, offerings, acquisitions, and some entity holdings. [researcher inference · strong] Genentech S-1 | Initial security, transfers, dilution, follow-ons, lockups, distributions, sale prices, and underlying LP economics. |
Perkins also recalled a later move from 20% to 30% carry around a Hillman offer. The effective vehicle, hurdle, catch-up, clawback, vesting, and partner allocation are not public; applying 30% to Fund I would be false precision. [investor-stated · provisional] Bancroft/NVCA, printed pp. 36–37
His account of an “Investor's Bill of Rights” describes no personal portfolio investing, immediate profit distribution, independent audits, and separated follow-on vehicles. The underlying document was not reproduced, so these remain participant-described incentives rather than inspected contract terms. [investor-stated · moderate] Bancroft/NVCA, printed pp. 33–35 CHM, pp. 17–18
People, capital, and influence map
- Eugene Kleiner: co-GP, consent partner, Fairchild operator, and network source. Early K&P results are partnership results. [retrospective witness · strong] Perkins oral history, printed pp. 31–42
- Tom Perkins: strongest supported roles are sponsor/incubator, technical interrogator, financing/board architect, chair, recruiter, and mentor. These roles vary by company and never imply sole causality. [researcher inference · strong] role ledger
- Frank Caufield and Brook Byers: later partners expanded sourcing, diligence, healthcare expertise, capital, and governance. Hybritech is principally Byers's deal. [retrospective witness · strong] Byers oral history
- Company builders: Treybig, Green, Katzman, and Loustaunou at Tandem; Swanson, Boyer, Cohen, scientists, and executives at Genentech; Maslak at Acuson. Investor narratives cannot absorb their invention and execution. [researcher inference · strong] case notes
- Co-investors and outside validators: Pitch Johnson, Bill Draper, Wilmington, Lubrizol, Lilly, research institutions, customers, and later capital providers supplied independent judgment, rights, distribution, or scale. [researcher inference · strong] candidate ledger
- LPs and institutions: LP capital bore vehicle risk; university and public research infrastructure created much of the scientific option set. Public records do not allocate returns or causal contribution precisely. [researcher inference · strong] denominator note
Uniform candidate ledger
| Candidate / status | Stage and vehicle | Thesis / decision | Capital and attribution | Outcome / evidence limit |
|---|---|---|---|---|
| University Laboratories / IN | Mid-1960s; Perkins household/personal company. | Adjustment-free laser. | Amount/equity unknown; Perkins claims invention, financing, supervision. | Merged into Spectra-Physics; consideration/proceeds unknown. [investor-stated · moderate] CHM, pp. 5–6 |
| Tandem / IN | 1974–75 formation; first K&P partnership. | Fault-tolerant computing; stage architecture before scale. | $50,000 recruiting authorization, then $1 million reported decision-near; security/ownership unknown. Perkins sponsor/chair; founders/operators and partners distinct. | IPO 1977, Compaq merger 1997; fund proceeds unknown. [documented behavior · strong] case |
| Qume / IN | Early K&P; vehicle/round unknown. | Public record preserves network sourcing, not thesis. | Bill Draper says Sutter Hill brought it to K&P. | Exit/economics unknown. [retrospective witness · moderate] Draper, p. 16 |
| Snow-Job / IN | Early K&P; exact vehicle unknown. | Motorcycle-to-snowmobile conversion; macro/distribution and market-size errors. | Check, security, sponsor, board rights unknown. | Failed; loss/recovery not independently reconstructed. [investor-stated · moderate] case |
| Andros / IN | Backed from 1972; exact vehicle/round unknown. | Artificial-heart commercialization with long development. | Firm attribution only. | Still pre-profit in 1981; later outcome not reconstructed. [contemporaneous record · moderate] TIME, p. 4 |
| Genentech / IN | Formed 1976; first K&P partnership. | Contract falsifiable proof, stage capital, use partners/licenses. | Initial accounts conflict at $100k–$250k; Swanson/Boyer founders, Perkins chair/finance. | IPO 1980; fund cash flows unknown. [documented behavior · strong] case |
| Cetus / IN | Position preceded/overlapped Genentech. | Existing biotechnology exposure; conflict later emerged. | Perkins recalled about $500,000 at too high a price; security/ownership unknown. | Position reportedly sold; proceeds unknown. [retrospective witness · moderate] Genentech oral, pp. 3–4 |
| Hybritech / IN | Formed 1977; later KPCB partnership unknown. | Monoclonal-antibody diagnostics. | Byers sponsor/chair; Perkins approval/mentor/director; check/ownership unknown. | IPO 1981, Lilly acquisition 1986; KPCB proceeds unknown. [retrospective witness · strong] Byers, pp. 37–42 |
| Collagen / IN | Initial K&P investment in Genentech year; vehicle unknown. | Purified animal collagen applications. | Byers identifies a Perkins investment; all capital terms unknown. | Later public company; proceeds unknown. [retrospective witness · moderate] Byers, p. 29 |
| Acuson / IN | Early 1980s medical imaging; KPCB vehicle unknown. | Acoustic-diffraction differentiation; back inventor, recruit team. | reported: $1.25m KPCB share of a $2.5m round; security/ownership unknown. | IPO 1986, Siemens $23/share tender in 2000; KPCB proceeds unknown. [retrospective witness · moderate] case |
| LSI Logic / IN | Early 1980s; exact KPCB vehicle unknown. | ASIC/fab strategy and conditional financing test. | Corrigan operated; Perkins/Valentine investor-directors; check/ownership unknown. | Outcome/economics not reconstructed. [retrospective witness · moderate] CHM panel, p. 34 |
| VLSI Technology / IN | Early ASIC formation; syndicate including KPCB. | Separate chip design from fabrication despite skepticism. | Firm participation supported; Perkins-personal sourcing not established. | Exit/economics not reconstructed. [retrospective witness · moderate] CHM panel, p. 8 |
| Unnamed satellite company / LEAD | Late-stage episode; company/vehicle unknown. | Capacity position facing a claimed loss; launch failure allegedly triggered insurance. | Amounts and role are firm-level self-report. | Identity and cash flow unverified; excluded from rankings. [investor-stated · provisional] WBUR, 00:31:51–00:33:08 |
| Apple / MISS | Intake meeting only; no full diligence or investment. | Categorized as another kit and outside K&P's technology-risk formation model. | No terms, vote, check, or ownership. | Missed opportunity; counterfactual return prohibited. [investor-stated · moderate] case |
| Compaq / CTRL | Original investment early 1980s; later Perkins governance. | Firm history credits Sevin Rosen referral and John Doerr sponsorship. | Not a Perkins-sponsored selection. | No Perkins investment-return claim. [researcher inference · strong] KP on Compaq |
No-hindsight decision lab 1 — Tandem Computers
Freeze before the full company check
Known: Treybig had identified continuous transaction systems as a customer problem; Green and Katzman had relevant software and architecture experience; no finished product or customer acceptance existed. Believed: modular redundancy could tolerate component failures at a usable price. Unknown: architecture, manufacturing, service cost, bank trust, sales cycle, warranty exposure, and exact financing terms. The decision-near record reports $50,000 to recruit/design and $1 million only after the architecture impressed the investors. [contemporaneous record · strong] BusinessWeek, article pp. 64–65
Stop here and answer without using the IPO or merger:
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome
K&P staged architecture/recruiting before company financing; later syndication brought independent capital and governance. Tandem went public in 1977 and merged into Compaq in 1997. Security, ownership, dilution, distributions, and fund proceeds remain unknown. [documented behavior · strong] full reconstruction
No-hindsight decision lab 2 — Genentech
Freeze at the 1976 seed decision
Known: Cohen–Boyer recombinant-DNA methods existed, but Genentech had no product, lab, production, clinical program, approval, or distribution. Believed: contract labs could produce a shorter peptide proof before an integrated build. Unknown: expression, rights, regulation, patents, scale, safety, market acceptance, and total capital. Initial amounts conflict between $100,000 and about $250,000; the defensible fact is staged experimentation, not a clean seed number. [researcher inference · strong] S-1 risk factors Perkins oral, pp. 4–7
Stop here and answer without using the scientific milestones, IPO, or Roche:
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome
K&P financed a contracted proof; Genentech demonstrated somatostatin in 1977 and human insulin in 1978, partnered with Lilly, and sold one million company shares at $35 in the 1980 IPO. Swanson and Boyer remained founders; K&P's later cash flows are not public. [documented behavior · strong] full reconstruction
No-hindsight decision lab 3 — Acuson
Freeze at the January 1982 institutional round
Known: Maslak claimed an ultrasound architecture that handled acoustic diffraction differently; he had prototype work and prior angel/founder capital but no established CEO record. Believed: technical differentiation could create materially better images and a recruitable company. Unknown: clinical validation, manufacturability, sales/service economics, regulatory path, product cadence, and all KPCB security/ownership terms. Public reconstruction reports: KPCB supplied about $1.25 million of a $2.5 million round; a later bridge/loan followed an over-budget prototype. [researcher inference · moderate] full reconstruction
Stop here and answer without using the IPO or Siemens tender:
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome
KPCB backed the technical program, helped recruit a team, and later supplied bridge capital according to the secondary reconstruction. Acuson began sales, went public in 1986, and became subject to Siemens's $23-per-share tender in 2000. Maslak's invention and operation remain distinct; KPCB proceeds are unknown. [researcher inference · moderate] case note SEC tender
No-hindsight decision lab 4 — Apple pass
Freeze at the intake decision
Known: a meeting opportunity existed; no reviewed memo shows a full founder meeting, product demonstration, customer data, pricing, terms, or IC vote. Believed: Perkins categorized the opportunity as another computer kit with insufficient technology risk for K&P's company-formation model. Unknown: nearly everything a real underwriting would require. The decision under review is therefore whether to collect information, not whether to invest at a documented price. [investor-stated · moderate] Bancroft/NVCA, p. 61
Stop here and answer without using Apple's later outcome:
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome
Perkins did not take the meeting and later acknowledged the miss. This was an intake false negative, not a priced investment rejection; no counterfactual fund return is supportable. [investor-stated · moderate] case note
No-hindsight decision lab 5 — Snow-Job
Freeze before launch
Known: a 1973 advertisement shows a motorcycle-to-snow-vehicle conversion being marketed; the precise investment date is unknown, so the October 1973 embargo cannot be treated as known at commitment. Believed: no contemporaneous investment thesis survives; Perkins later recalled that the product worked. Unknown: check, security, ownership, unit economics, demand by region, channel commitments, inventory, runway, and shutdown rules. [researcher inference · moderate] WBUR, 00:12:35–00:15:43
Stop here and answer without using the failure:
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome
The company failed around the oil embargo; Perkins later conceded that even success would not have produced a large outcome. That makes market size, distribution resilience, and explicit stop rules the portable lessons. Exact loss and recovery remain unknown. [investor-stated · moderate] case note
Portfolio forensics
- Selection: the best-supported pattern is technical interrogation paired with willingness to form a team around an unproven product. Apple shows the shadow side: a screen tuned to technical formation can reject a category before collecting enough evidence. [researcher inference · strong] Tandem Apple
- Staging: Tandem and Genentech both reduced an integrated-company problem into an information-producing first check. The transferable mechanism is milestone design, not the celebrated sector. [researcher inference · strong] Genentech Tandem
- Syndication and external proof: co-investors, laboratories, partners, and customers did more than supply money; they added independent validation and capabilities. [researcher inference · strong] people map
- Governance: close board work, recruiting, financing, and follow-on signaling were central, but founder/operator credit and potential dependence must remain visible. Formal protective terms are unavailable. [researcher inference · strong] Acuson Tandem
- Concentration: the public record supports two exceptional Fund I outliers and an ordinary/failing residual, but no exact loss ratio or outlier share. [researcher inference · strong] return claims
- Liquidity: IPO and acquisition dates are company outcomes, not fund cash flows. Distribution timing, lockups, sale decisions, carry, and partner proceeds remain unavailable. [researcher inference · strong] missing fields
Failure and miss postmortems
| Case | Error class | What the record supports | Process repair | Boundary |
|---|---|---|---|---|
| Snow-Job | Macro/distribution shock plus small potential market | Perkins says launch collided with the oil embargo and later judged the upside too small. [investor-stated · moderate] case | Stress channel availability, unit economics, geographic demand, runway, and whether success can matter at fund scale. | Exact company identity, check, loss, and shutdown record unresolved. |
| Apple | False negative before diligence | Category fatigue and a technology-risk screen prevented a full meeting. [investor-stated · moderate] case | Separate “outside current model” from “unworthy of information”; require a cheap evidence-gathering gate for anomalous founder/product pull. | Not a priced IC pass; counterfactual return prohibited. |
| Acuson roadmap dispute | Governance/product-cadence disagreement | Perkins later opposed a multi-year product gap; the company outcome alone does not prove his preferred roadmap. [investor-stated · moderate] case | Predefine product-cadence evidence, dissent documentation, and board escalation. | Board minutes and alternative counterfactual absent. |
| Unnamed satellite | Luck/selection contamination | Perkins described insurance after a launch failure averting a large claimed loss. [investor-stated · provisional] WBUR, 00:31:51–00:33:08 | Record identity, underwriting, insurance, and counterfactual loss before labeling skill or luck. | Identity and cash flow unverified; excluded from rankings. |
| Ordinary Fund I positions | Survivorship and denominator gap | Perkins recalled seven or eight companies around break-even, but public histories foreground Tandem and Genentech. [investor-stated · moderate] Bancroft/NVCA, pp. 38–40 | Preserve the full portfolio schedule, write-offs, reserves, and distributions; sample ordinary cases, not only outliers. | Names and economics insufficient for a fourth independent ordinary-case lab. |
Return claims: record, do not calculate
Public sources report $8 million committed and roughly $4 million deployed; reported: “hundreds of millions” returned; reported: 51% annualized and 16% without two outliers in a secondary reconstruction; reported: 41.5% over ten years in a tertiary history; reported: approximately 40% firmwide in Perkins's later account; and reported: more than $35 billion returned across five decades in a current firm announcement. These claims use different vehicles, periods, definitions, and evidence quality. None supplies the cash-flow series needed to calculate or audit a Perkins-personal or Fund I return. [researcher inference · strong] claim-by-claim ledger
Uncertainty and retrieval ledger
Processed in full or at document level
- Bancroft/NVCA Perkins oral history; Genentech oral history and 1980 S-1 appendix; CHM Perkins transcript; full Red Herring interview; official WBUR audio-derived transcript; Tandem issuer and decision-near press records; Acuson SEC transaction; Compaq proxy; 1984 congressional testimony; and official company/firm histories. [documented behavior · strong] source map
Highest-value missing artifacts
- Executed 1972 partnership agreement and subscriptions; audited statements and capital accounts; complete Fund I portfolio schedule; company cap tables and transfer ledgers; distribution notices; partner carry allocations; original investment memos, term sheets, board minutes, customer contracts, and the reportedly recovered Hillman copy of the original plan. [researcher inference · strong] retrieval note
Hard conclusions that remain prohibited
- Exact Fund I company count; company-level or person-level IRR/TVPI/DPI; Perkins's net worth or personal proceeds; ownership inferred from company valuation; solo founder/causality claims; return attribution from a board seat; and counterfactual Apple gains. [researcher inference · strong] attribution controls
Coverage and evidence base
The seven priority routes in sources.md collapse into three candidate Perkins-authored or Perkins-described writing families plus one indispensable written governance control. The candidate families are the 2007 memoir Valley Boy, the conflict rules Perkins said appeared in Kleiner & Perkins's missing 1972 business plan under the name “Investor's Bill of Rights,” and his January 2014 Wall Street Journal letter. Hewlett-Packard's September 2006 corrective 8-K is an issuer-authored record, not a Perkins letter: the filing expressly says he supplied no written explanation at resignation. annotated source map HP 8-K, Item 8.01, paragraph beginning “On May 22”
| access class | routes | deeply analyzed work/control families | exact boundary |
|---|---|---|---|
| full public body | 2 | 2 | Bancroft/NVCA transcript containing Perkins's oral description of the missing 1972 plan; HP's issuer-authored corrective 8-K, controlled by the SEC order. |
| partial public body | 2 | 2 | Valley Boy limited preview; authenticated fragments of the 2014 letter and later remarks in contemporaneous coverage. |
| metadata-only | 3 | 0 | Valley Boy publisher and audiobook manifestations; Sex and the Single Zillionaire bibliographic record. |
| total | 7 | 4 | Every route is processed to its public boundary; manifestations are not counted as extra works. |
The Valley Boy preview did not expose a stable public page inventory during this run, so no unseen memoir proposition is analyzed. Publisher pages authenticate edition and scope only. The novel's body is unavailable and its investment relevance is low. The 2014 original is paywalled; only passages reproduced by contemporary sources are treated as direct. No private IC memo, LP letter, board correspondence, partnership agreement, or complete prediction ledger surfaced. [researcher inference · strong] memoir note 2014 note
Source-by-source analysis
1972 plan, orally described in 2009 — the purported “Investor's Bill of Rights”
Authorship, context, and access
Perkins told Bancroft interviewer Sally Smith Hughes that he wrote Kleiner & Perkins's original business plan, but also said he did not know where it was. He then orally described a section he and Kleiner called an “Investor's Bill of Rights.” The public record therefore contains Perkins's 2009 recollection of a purported 1972 text—not the text itself. [investor-stated · provisional] Bancroft/NVCA, printed pp. 32–35, especially p. 34 source note
Argument, reasoning, and examples
The described framework targets conflicts between general partners and their fund investors. Perkins listed four practices he considered abusive: acquiring founder stock cheaply before putting the fund in at a higher price; diverting an attractive opportunity away from the fund; failing to supply audited statements; and cherry-picking personal co-investments, potentially at preferential prices. He said their response was not to make any personal investment the fund could have made. [investor-stated · moderate] Bancroft/NVCA, printed pp. 34–35
The reasoning chain is incentive-based: if GPs can exploit information or choose after observing quality, they can privatize upside and leave weaker risk with LPs; eliminating eligible personal investments removes one route for selection abuse. Perkins later said he and Kleiner also invested as LPs in the partnership, so every partnership investment gave them automatic participation. That is alignment evidence, not proof of equal exposure after carry, fees, dilution, tax, or later vehicles. [investor-stated · moderate] Bancroft/NVCA, printed p. 40
Assumptions, application, and contradictions
The prescription assumes “anything the fund might invest in” can be defined before a deal and that audits plus a personal-investment ban control allocation conflicts. Modern multi-fund, scout, SPV, continuation, and cross-stage structures make that perimeter harder to draw; even in 1972, the missing plan and agreements prevent a clause-by-clause compliance test. [researcher inference · moderate] oral description
Perkins claimed this contribution “cleaned up” the industry and set a standard once K&P succeeded. That causal claim is not independently established: no peer denominator, adoption series, governing agreements, or enforcement record was found, and K&P's success could make a pre-existing norm more visible without causing it. [investor-stated · insufficient public record] Bancroft/NVCA, printed pp. 34–35
2006 — HP resignation and disclosure record, not a Perkins-authored letter
Authorship, context, and chronology
HP's corrective filing says Perkins gave no written explanation when he resigned on May 18. It reports that he objected to bringing the leak matter before the full board, believed he and the chair had agreed to handle the implicated director privately, and opposed seeking that director's resignation for what he viewed as a relatively minor offense. This reverses a common compressed account that casts his immediate objection as a demand for full-board process. [contemporaneous record · strong] HP 8-K, Item 8.01, paragraphs beginning “On May 22” and “George A. Keyworth” SEC order, ¶¶6–9 source note
Reasoning and observable conduct
The SEC order records Perkins's logic more fully: confidential, “off-line” warning by the chair and governance-committee chair; proportionality to Keyworth's contribution and perceived offense; objection to the meeting process; then resignation after the board voted by secret ballot to request Keyworth's departure. The observable action supports costly dissent and a preference for private resolution. It does not show that his preferred procedure was legally or institutionally superior. [documented behavior · strong] SEC order, ¶¶8–9
The public record first documents Perkins investigating the investigative methods after resignation. HP says that on June 19 he sought information and asserted improper recording; HP then disclosed that pretexting had been used and he requested an inquiry. Thus the later privacy/ethics objection cannot be projected backward as the complete May 18 reason. [contemporaneous record · strong] HP 8-K, Item 8.01, paragraph beginning “On June 19”
Outcome, contradiction, and decision relation
The SEC found HP should have disclosed Perkins's disagreement because it concerned operations, policies, or practices; HP settled without admitting or denying the findings except jurisdiction and accepted a cease-and-desist order. The respondent was HP, not Perkins. The record therefore supports disclosure and investigation controls, but not a personal securities-law sanction or a general claim that Perkins “whistleblew” every issue at resignation. [contemporaneous record · strong] SEC order, ¶¶1, 12–15 and §IV
For investing, the case yields a falsifiable governance checklist: identify who may investigate directors; when a matter stays private or reaches the board; what evidence and counsel are required; when recusal applies; how proportionality is assessed; and which disagreements must be disclosed. It does not validate Perkins's preferred answer to each question. [researcher inference · strong] HP/SEC note
2007 — Valley Boy: The Education of Tom Perkins
Authorship, publication context, and access
The participant memoir was published after Perkins's HP resignation and three decades after the defining Tandem and Genentech decisions. Its title and publisher description frame technical education, company formation, venture partnership, governance, and personal wealth as one “education.” Only a limited preview, edition metadata, interviews about the book, and independently locatable oral-history accounts were lawfully available; the full memoir was not reconstructed from snippets or reviews. [investor-stated · provisional] Google Books limited preview publisher description source note
Observable argument and reasoning chain
The accessible framing proposes an operator-to-investor sequence: MIT/HBS and Doriot; HP technical, sales, and management work; University Laboratories as firsthand company creation; then an involved venture practice. Independent direct interviews record the same sequence—Doriot's course at printed pp. 5–7, HP at pp. 10–27, and the partnership plan at pp. 31–35—but those interviews are controls on self-narration, not visible memoir pages. [investor-stated · moderate] Bancroft/NVCA, printed pp. 5–7 Bancroft/NVCA, printed pp. 10–27
The implied rule is to turn operating exposure into better product, organizational, financing, and board questions. University Laboratories is the strongest application because Perkins actually built and sold a technical company before K&P. Yet public evidence does not show which exact lesson changed a later decision or how the same process performed across losing and ordinary investments. [researcher inference · moderate] CHM Perkins history, pp. 6–7
Selection, attribution, and contradictions
Memoir evidence is outcome-selected. Tandem and Genentech can illustrate a process, but the public first-fund denominator conflicts at 17 versus 18 companies and lacks audited LP cash flows. Founder, scientist, operator, Kleiner, Byers, Johnson, co-investor, LP, and later-partner work must not collapse into a single hero narrative. [researcher inference · strong] coverage assessment
The memoir is therefore valuable for intellectual development and hypothesis generation, not for proving personal alpha, complete portfolio performance, or causality. A valid transfer must specify the operating lesson, the decision it should change, contemporaneous evidence, and a counterexample. [researcher inference · strong] memoir note
2014 — wealth letter, reported retraction, and weighted-vote remarks
Text, context, and access
The paywalled January 24 letter compared criticism of rich Americans with Nazi persecution of Jews. The New Yorker reproduces its central question and records Perkins defending the minority-hatred premise on Bloomberg; TIME reproduces the central passage and reports that he later called the historical term a terrible choice. These stable reproductions authenticate the controversy but are not a complete original-text edition. [investor-stated · strong] New Yorker, paragraphs beginning “On Saturday” and “The letter was controversial” TIME, paragraphs beginning “Perkins’ latest comments” and “Perkins would later retract” source note
Reasoning, assumptions, and counterevidence
Perkins's reasoning treats criticism of a wealthy minority as a potentially escalating form of group hatred. That chain assumes relevant equivalence between redistributive rhetoric/protest and state-enabled racial persecution, and it omits power, coercion, historical mechanism, and proportionality. Calling the term a poor choice narrows the analogy but retains the class-hostility thesis. [researcher inference · strong] New Yorker, paragraphs 37–42 in the accessible article body
At a February Commonwealth Club event, Perkins suggested tax payment as a voting qualification and tax amount as a vote multiplier. TIME places the remarks around 1:03:00 and reports both his later “provocative” description and a reporter's account that he said he was not serious. The idea is therefore authenticated but seriousness is ambiguous; it remains a public-judgment and governance signal, not an investment-result claim. [investor-stated · moderate] TIME, paragraphs beginning “Tom Perkins,” “But that's not all,” and timestamped paragraph beginning “Later in the interview”
Decision relation and contradiction
The episode is a counterexample to a global-good-judgment thesis: demonstrated technical or board success does not validate reasoning in an unrelated historical/political domain. It does not prove that Perkins's technical calibration always failed to transfer, nor does it retroactively change a fund return. [researcher inference · strong] 2014 source note
KPCB publicly distanced itself from the comments and stated Perkins had not been involved in years. That response protects the boundary between retired co-founder and contemporary firm, just as later KPCB portfolio lists must not be assigned to him personally. [documented behavior · strong] New Yorker, paragraphs beginning “Soon after”
Cross-source synthesis
- Conflict design is the clearest described written rule. The purported 1972 plan attacks GP self-dealing, deal diversion, unaudited reporting, and cherry-picked co-investment. Its causal path is aligned allocation → less hidden extraction → more trustworthy LP relationship; only the prescription and Perkins's own-LP statement are public, not portfolio-wide compliance or return impact. [researcher inference · strong] Bancroft/NVCA, printed pp. 34–35 and 40
- Operating experience is an input, not a universal warrant. The memoir framing and oral histories connect HP/University Labs to later involvement. The 2014 record supplies a domain-transfer counterexample; modern use should require relevant expertise and decision-specific evidence. [researcher inference · strong] CHM Perkins, pp. 6–7 2014 note
- Costly dissent is observable, but the preferred process still needs testing. Resignation makes Perkins's HP objection concrete. The verified objection favored private handling and opposed the board's sanction; it does not establish that private resolution was correct. [researcher inference · strong] SEC order, ¶¶8–9
- Governance must distinguish allocation, investigation, and disclosure. The 1972 recollection addresses GP-LP conflicts; HP addresses board investigation and issuer disclosure; the 2014 episode addresses public power and firm identity. Combining them into a generic “investor rights” doctrine would erase the decision-maker and harmed party in each. [researcher inference · strong] Bill note HP note
Intellectual chronology, evolution, and contradictions
| period | observable writing or written-record emphasis | unresolved tension |
|---|---|---|
| circa 1972, recalled 2009 | Ban GP opportunity diversion, preferential personal purchases, unaudited reporting, and selective co-investment. [investor-stated · moderate] Bancroft/NVCA, pp. 34–35 | Original plan and agreements are missing; industry-standard-setting claim is unverified. |
| 2006 | Private handling and proportionality for a director leak dispute; later inquiry into pretexting; issuer duty to disclose material disagreement. [contemporaneous record · strong] HP 8-K SEC order, ¶¶8–15 | Private resolution, full-board authority, lawful investigation, and shareholder disclosure pull in different directions. |
| 2007 | Operator apprenticeship organized as venture education. [investor-stated · provisional] memoir record | Outcome selection and incomplete ordinary-case economics prevent causal proof. |
| 2014 | Class-hostility warning, partial retraction of historical wording, and provocative tax-weighted voting remarks. [investor-stated · moderate] TIME, final article paragraphs | Public reasoning conflicts with a reputation for calibrated, domain-specific risk judgment. |
Decision playbook derived from the corpus
| historical rule or reconstruction | diagnostic questions | boundary and failure mode |
|---|---|---|
| Define the opportunity-allocation perimeter before seeing deal quality. [researcher inference · strong] Bancroft/NVCA, pp. 34–35 | Which fund, SPV, account, partner, scout, or affiliate is eligible? Who verifies price and allocation? | A vague “fund might invest” test can be rewritten after the winner is visible. |
| Eliminate preferential personal economics and audit the residual conflicts. [researcher inference · strong] Bill note | Are security, price, timing, fees, carry, information and follow-on rights equivalent? What is disclosed and independently reviewed? | Co-investment at the same headline price may still have better timing, information, liquidity, governance, or fee terms. |
| Separate confidential triage, board authority, investigation, and public disclosure. [researcher inference · strong] SEC order, ¶¶3–15 | Who investigates whom? When must counsel, full board, recusal, regulator, or owners be involved? What evidence triggers escalation? | Private handling can preserve trust or conceal misconduct; automatic full-board exposure can impair confidentiality and fairness. |
| Treat resignation as evidence of disagreement, not evidence of correctness. [researcher inference · strong] SEC order, ¶¶8–9 | Was the objection recorded before outcome? What alternative procedure and predicted harm were specified? | Costliness may reveal conviction, self-protection, temperament, or error; it does not identify which. |
| Keep expertise and analogies domain-bounded. [researcher inference · strong] 2014 note | Are mechanisms, base rates, affected parties and power relations actually comparable? What would falsify the analogy? | Reputation in technology or finance can substitute for absent historical, political, or ethical evidence. |
Modern VC translation as of 2026-08-23
- Cross-fund and SPV allocation. [researcher inference · strong; analyst-created scenario] Imagine one manager operating seed, growth, opportunity, scout and personal vehicles. Transfer the 1972 anti-cherry-picking principle by publishing ex-ante eligibility, price, allocation, follow-on and exception rules; log every exception; and require independent conflicts review. What changed is vehicle multiplicity and information speed. Signals are pre-deal timestamps, consistent allocation, equivalent terms and audited exceptions. Misuse is declaring all conflicts cured because vehicles pay the same headline price. historical basis
- AI-company incident governance. [researcher inference · strong; analyst-created scenario] For a hypothetical model company, pre-agree confidential triage, independent technical investigation, evidence preservation, full-board escalation, regulator/customer disclosure, director recusal and sanctions. What transfers from HP is the need to separate these decisions; what does not transfer is Perkins's preferred private resolution. Signals are written authority, counsel independence, tamper-evident logs and decision-date minutes. Misuse is calling secrecy “trust” or publicity “transparency” without due process. historical basis
- Operator-to-investor learning. [researcher inference · moderate; analyst-created scenario] For a former AI-infrastructure executive assessing a biotech platform, transfer process discipline—identify the decisive bottleneck and evidence—not domain confidence. Changed assumptions include science, regulation, capital duration and validation institutions. Signals are explicit outside experts, prewritten falsifiers and ordinary-case scoring. Misuse is treating prior operating prestige as portable selection alpha. historical basis
- Public claims by fiduciaries and firm elders. [researcher inference · strong] Require mechanism, proportionality, affected-party, historical-expert and firm-attribution checks before publication. The 2014 record shows how a retired co-founder's rhetoric can trigger a current-firm disavowal. Signals are explicit personal/firm capacity and documented review for high-impact claims. Misuse is treating provocation or attention as evidence. historical basis
These are analyst-created applications, not claimed Perkins rules. They deliberately avoid naming current companies because no current case is needed to test the checklist and because a historical analogy should not manufacture contemporary facts.
Highest-value writings and written controls
- Purported 1972 “Investor's Bill of Rights,” orally described at printed pp. 34–35 — most explicit investment-conflict rule, with missing original and compliance record. [researcher inference · strong]
- HP corrective 8-K plus SEC order — strongest decision-near governance control, explicitly not authored by Perkins. [researcher inference · strong]
- Valley Boy limited preview — central authored self-conception, but only partially accessible and outcome-selected. [researcher inference · moderate]
- 2014 wealth letter fragments and response — strongest adverse public-judgment record. [researcher inference · strong]
- No fifth substantive investment/governance writing surfaced. The novel is metadata-only and low-relevance; publisher/audiobook pages are memoir manifestations. Promoting one would fabricate depth. [researcher inference · strong]
Unresolved written record
- The 1972 K&P business plan and purported “Investor's Bill of Rights” original are missing; only Perkins's printed-pp. 34–35 oral description is public.
- No lawful public full text or stable visible-page inventory of Valley Boy surfaced; the full body is not inferred from reviews, snippets, or later interviews.
- The 2014 WSJ original is not openly accessible; authenticated excerpts and reported retraction/context are processed, not an invented complete edition.
- HP's corrective 8-K is issuer-authored and says Perkins provided no written reason at resignation; a “Perkins resignation letter” was not found.
- The public corpus is not a lifetime bibliography. Private memos, LP letters, board correspondence, partnership files, and later revisions may exist.
- No writing supplies a complete screening, pass, investment, ownership, return, carry, forecast, or attribution denominator. The corpus supports conflict and governance diagnostics more strongly than repeatable person-level alpha.
Coverage and evidence base
The priority spoken corpus contains 16 mapped routes representing 15 underlying appearance families. All 15 families were processed: eight full-text or full-audio families, three partial or edited-text families, and four metadata/access-boundary families. The TechCrunch video/report routes are one appearance; the three oral histories are separate interviews; and the Smithsonian collection/item routes identify one raw Perkins session. [researcher inference · strong] Source-map denominator
“Full” means that the accessible transcript, article body, or official audio was processed end to end; it does not make retrospective testimony independent or decision-near. The 45:59 WBUR recording was transcribed through the OpenAI API with whisper-1, and its machine wording is paraphrased against timestamps. Venture Voice, Commonwealth Club, Risk & Reward, and the raw Something Ventured interview produced access-boundary notes rather than inferred speech. [researcher inference · strong] WBUR access note access-boundary ledger
Source-by-source analysis
1993 — Red Herring, “Venture Pioneer Tom Perkins Unveils Some Secrets”
The original-interviewer mirror is the earliest substantial direct record in the mapped corpus. Perkins says K&P looked for large-scale importance, screened ideas before people, used board involvement as its product, and had participated in roughly 180 companies; the count is firm-level and definition-free, not a personal portfolio denominator. [investor-stated · moderate] 1993 Q&A, “biggest hits,” “elements to success,” and “competitive advantage” Source note
- Decision evidence: Snow-Job exposes both external shock and weak maximum upside; Perkins later says the company would not have mattered much even had it worked. [investor-stated · provisional] 1993 Q&A, “memorable failures”
- Governance evidence: he treats crisis response and board work as venture capital’s non-cash product, while providing no ordinary-company or founder-satisfaction denominator. [investor-stated · moderate] 1993 Q&A, “summarize the role” and “hardest part”
- Attribution control: early K&P was a partnership with Eugene Kleiner; “we” does not prove that Perkins sourced, approved, or led every referenced company. [researcher inference · strong] 1993 Q&A, “get together with Eugene Kleiner”
2001–2002 — Genentech oral history
The Bancroft Genentech interview is the deepest mapped participant account of Genentech’s staged formation. It covers Robert Swanson’s commercial initiative, Herbert Boyer and the scientific network, subcontracted proof, K&P financing, the board, licensing, the IPO, and later financing structures. [investor-stated · moderate] Genentech oral history, printed pp. 3–13 and 21–41 Source note
- Framework: replace a capital-heavy integrated plan with the smallest contracted scientific proof that can answer the decisive question. [researcher inference · strong] Genentech oral history, printed pp. 4–7
- Attribution: Swanson drove commercialization, Boyer supplied scientific partnership, laboratories performed the proof, and Perkins chaired and financed; none can be collapsed into a lone-founder story. [researcher inference · strong] Genentech oral history, printed pp. 3–10
- Economics boundary: recalled $100,000, $200,000, and $250,000 figures may describe different tranches or totals; no public cap-table-to-distribution chain resolves them. [researcher inference · strong] Genentech oral history, printed pp. 4–7 and 13
2002 event / 2017 upload — CHM “Pioneers of Venture Capital”
The CHM record establishes Perkins on a six-investor panel with William Draper III, Pitch Johnson, Burton McMurtry, Arthur Rock, and Donald Valentine. The event occurred in 2002 and the official video was uploaded in 2017; no 2010 event for this title was authenticated. [contemporaneous record · strong] CHM event record Source note
- Three moderator-bounded Perkins turns support his operator-to-investor account, distributed Tandem attribution, and admission that he never took the scheduled Apple meeting; machine captions were paraphrased, not quoted. [investor-stated · moderate] CHM video, 00:35:52–00:39:11, 00:50:54–00:53:18, and 01:08:03–01:08:43
2007 — CBS, “The Captain of Capitalism”
The CBS profile combines a 13:26 broadcast with an edited narrative report. It records Perkins’s Genentech description, a rough two-in-ten home-run intuition, yacht-building temperament, childhood self-interpretation, and an adversarial HP exchange. [contemporaneous record · moderate] CBS report, paragraphs beginning “He and his partners launched,” “And it took somebody,” and “Is it possible” Source note
- Useful tension: the build-stage technical challenge appears motivating, but a personality narrative cannot establish a portfolio rule. [researcher inference · provisional] CBS report, yacht sections
- HP control: Perkins acknowledges anger and possible error in how he resigned; his denials and Patricia Dunn’s allegations remain party positions, not adjudicated chronology. [researcher inference · strong] CBS report, paragraphs beginning “After he resigned” and “Asked if he misses”
- Return boundary: CBS’s reported: $250,000-to-$200 million Genentech summary lacks security, ownership, dilution, distribution, and LP-return records. [researcher inference · strong] CBS report, Genentech return paragraph
2007-11-28 — WBUR On Point
The official WBUR episode is the fullest public audio treatment in the mapped corpus; the following claims are timestamped paraphrases from a machine transcript, not publication-transcript quotations. [researcher inference · strong] WBUR access note
- 02:55–07:57: Fund I is described as $8 million committed, about $4 million invested, with Tandem and Genentech as the two major hits; idea, decisive risk, and staged proof precede team build-out. [investor-stated · moderate] official WBUR episode
- 12:35–15:43: Snow-Job is a candid failed case and a market-size correction. [investor-stated · moderate] official WBUR episode
- 22:08–22:16: Perkins predicts executive exit packages will taper as boards respond, but supplies no horizon, metric, or comparison set. [investor-stated · provisional] official WBUR episode
- 23:23–31:38: technical, manufacturing, people, patent, and customer risks should be separated; burn and headcount remain low until the named risk is resolved. [investor-stated · strong] official WBUR episode
- 25:51–26:58: former portfolio executives are described as a referral network. This is an access mechanism and a possible exclusion mechanism, not proof of meritocratic reach. [researcher inference · moderate] official WBUR episode
- 31:51–33:08: an unnamed satellite investment is admitted as luck after insurance rescued a looming loss. [investor-stated · provisional] official WBUR episode
- 37:22–38:53: Perkins’s “we started Compaq” formulation conflicts with KPCB’s attribution to Sevin Rosen and John Doerr; his supported role belongs to later governance/M&A. [researcher inference · strong] WBUR episode KPCB Compaq history
- 43:23–45:31: aspiring VCs are told to build a company first, consistent with the University Laboratories operator bridge. [investor-stated · moderate] official WBUR episode
2007 — Commonwealth Club
The Hoover catalog identifies a 59:27 Perkins speech dated 2007-11-14, while ABC7 says it was recorded 2007-10-28. No public body was recovered, so the date conflict is preserved without speech claims. [contemporaneous record · moderate] Hoover record ABC7 metadata Source note
2007-12-05 — Charlie Rose
The official speaker-labeled transcript adds capital-supply forecasts and Perkins’s decision-near HP chronology. [contemporaneous record · strong] Charlie Rose transcript Source note
- 09:52–10:40: Perkins argues that $20–40 billion of available venture capital cannot collectively earn a universal 100x aspiration and will inflate prices. Vintage, net/gross metric, and horizon are undefined. [investor-stated · provisional] official transcript
- 11:26–14:16: he credits John Doerr’s earlier Internet optimism, then predicts large green and biotechnology opportunity. The forecasts lack investable definitions and scored outcome windows. [investor-stated · provisional] official transcript
- 14:23–16:13: he claims K&P’s conflict rules were copied across the field. The original plan and adoption/compliance denominator remain unavailable. [investor-stated · insufficient public record] official transcript
- 16:53–17:14: the firm’s roughly 525-company network is described as a sourcing engine, not a Perkins portfolio. [researcher inference · strong] official transcript
- 17:25–18:13: he predicts long-term opportunity in China while describing KPCB’s approach as cautious; no vehicle, attribution, horizon, or outcome measure is supplied. [investor-stated · provisional] official transcript
- 21:42–26:08: he says he wanted HP’s leak issue handled privately before opposing a full-board process and resigning. This is narrower than a later whistleblower simplification and aligns with the SEC chronology on the private-resolution preference. [researcher inference · strong] official transcript SEC order ¶¶4–9
2008-01-17 — Creative Capital, “Valley Boy”
The full edited Q&A repeats fail-cheap/power-law reasoning, distinguishes startup entry from public-market timing, treats money as an undifferentiated VC input, records MySpace board advice, and attributes succession to Ray Lane, John Doerr, and Brook Byers. [investor-stated · moderate] Creative Capital Q&A, “worst investment,” bubbles, MySpace, and succession questions Source note
- Perkins predicts that green technology will matter, KPCB will raise a thirteenth fund in 2008, and China expansion will remain experimental amid positive US–China relations. None is supplied with a scoring horizon, defined market, attributed vehicle outcome, or return metric. [investor-stated · provisional] Creative Capital Q&A, green, succession, and China questions
2008-10-08 — Something Ventured raw interview
The Smithsonian item record identifies an in-person Perkins interview and transcript within the film archive. The item requires in-person research; the edited documentary cannot substitute for the raw session, so no speech claim is drawn from it. [contemporaneous record · strong] Smithsonian item NMAH.AC.1467_ref12 Source note
2009 — Bancroft/NVCA oral history
The 70-page Bancroft/NVCA oral history is the mapped corpus’s biography and philosophy spine. It covers childhood, education, HP and University Laboratories, K&P formation, Fund I, major deals, boards, conflicts, and failure; its publication restriction is respected through paraphrase and printed-page locators. [researcher inference · strong] Bancroft transcript, printed pp. 5–65 Source note
- Strongest portable rule: define the opportunity and decisive risk; buy the smallest proof; increase capital after proof; recruit the missing team around the validated opportunity. [researcher inference · strong] Bancroft transcript, printed pp. 33–35 and 45–49
- Strongest falsifier: Apple was rejected without a founder meeting because Perkins saw no technical risk and treated marketing risk as unstageable. [researcher inference · strong] Bancroft transcript, printed p. 61
- Strongest concentration evidence: Perkins says seven or eight non-outlier Fund I deals were only barely profitable, leaving Tandem and Genentech as the remembered engines; this is not an audit. [investor-stated · provisional] Bancroft transcript, printed pp. 32–33
- Strongest control warning: Perkins says lead-investor follow-on signaling creates effective control even with low ownership; that can enable rescue or create founder dependence. [researcher inference · moderate] Bancroft transcript, printed pp. 62–63
- Document conflict: in 2009 Perkins says the 1972 plan was missing; in 2011 he says Henry Hillman recovered a copy. The public corpus still does not expose the artifact. [researcher inference · strong] Bancroft transcript, printed p. 33 CHM transcript, pp. 17–18
2011-07-22 — Computer History Museum oral history
The CHM transcript separately records the operator apprenticeship and risk-decomposition algorithm, and supplies Acuson as an ordinary-success case. It is a second Perkins interview, so repetition is same-narrator consistency rather than independent corroboration. [researcher inference · strong] CHM transcript, pp. 17–21 Source note
- Acuson: Perkins interrogated the physical mechanism rather than trusting a long plan, backed an untested manager, recruited the team, and later challenged a roadmap that endangered the installed franchise. [investor-stated · moderate] CHM transcript, pp. 20–21
- Scope control: acknowledged failures have no complete denominator; the famous-case selection cannot establish personal hit rate or repeatable alpha. [researcher inference · strong] CHM transcript, p. 18
- Plan update: the recovered-plan recollection updates the 2009 missing-plan claim but does not make the plan public. [researcher inference · strong] CHM transcript, pp. 17–18
2012 — Knowledge at Wharton
The institutional report is edited rather than transcript-complete. It ties Packard’s order-of-magnitude contribution heuristic to Genentech staging, individual decision accountability, University Laboratories, small IPO pricing, team sequence, and referral sourcing. [contemporaneous record · moderate] Wharton report, opening, “Getting the Right Answers,” and Q&A sections Source note
- A dramatic improvement is a search heuristic, not a sufficient condition. [researcher inference · strong] Wharton report, opening paragraphs
- Risk up front means financing a discriminating experiment before building a full organization. [researcher inference · strong] Wharton report, “Getting the Right Answers”
- Anti-committee rhetoric can counter conformity or suppress dissent; the appearance provides no comparative decision dataset. [researcher inference · moderate] Wharton report, HP and corporate-VC paragraphs
- The claim that Silicon Valley is meritocratic is not established by a self-described closed referral network. [researcher inference · strong] Wharton report, closing Q&A
2013-09-11 — TechCrunch Disrupt with Don Valentine
The mapped video route and contemporaneous report are one appearance. Only the edited report and agenda were accessible; the hosted video, captions, and transcript did not surface. The report preserves Perkins’s idea-first and size-screen fragments plus the Apple admission, while its other quoted rules belong to Don Valentine and are not reassigned. [researcher inference · strong] TechCrunch report, opening Perkins paragraph and closing Apple paragraph official agenda Source note
Venture Voice and Risk & Reward — access boundaries
The Venture Voice slug now resolves to a generic landing page whose roster authenticates Perkins as a guest but exposes no episode body, date, audio, or transcript. The Risk & Reward route returned a 114-byte empty body with no identity, date, or media proof; neither route supports a speech claim. [contemporaneous record · insufficient public record] Venture Voice route Venture Voice note Risk & Reward route Risk & Reward note
Recurring stories and frameworks
1. Opportunity first, team in sequence
Across the direct record, Perkins begins with an opportunity large enough to matter rather than a complete executive team. The operational version is opportunity → named uncertainty → cheap proof → more capital → missing management. Tandem, Genentech, Hybritech, and Acuson show that this is idea-first sequencing, not indifference to people. [researcher inference · strong] Red Herring, risk and people answers Genentech oral history, printed pp. 4–10 CHM, pp. 20–21
The failure mode is moral and commercial overreach: a good idea does not prove good character, and Apple shows that a great opportunity may contain market-creation skill rather than a cheap technical experiment. [researcher inference · strong] TechCrunch idea/character and Apple fragments Bancroft, printed p. 61
2. Risk is an asset only when decomposable
“High risk” is attractive only when it marks a barrier and can be split into testable technical, market, customer, legal, patent, manufacturing, science, or people questions. The investor should fund the next discriminating fact, not celebrate uncertainty. Genentech is the signature proof; Apple is the false negative; Snow-Job shows poorly bounded market/exogenous risk; the unnamed satellite shows luck. [researcher inference · strong] WBUR, 06:58–07:57 and 23:23–33:08 Genentech oral history, printed pp. 4–7
3. The investor’s product is crisis work
Perkins repeatedly defines board service, recruiting, financing, and early trouble detection as the return for equity. A practical test is whether the director helps during a broken experiment, financing gap, leadership dispute, or strategic fork. The corpus lacks a founder, employee, or ordinary-company denominator, so the service claim is behaviorally plausible rather than portfolio-proven. [researcher inference · moderate] Red Herring, role and hardest-part answers Bancroft, printed pp. 62–65
4. Lead power substitutes for formal ownership
Perkins says first-investor follow-on signaling can confer effective control at low ownership. It can keep a company financeable and speed intervention, or create coercive dependence and suppress founder agency; modern use therefore needs explicit reserved matters, board independence, financing alternatives, and recorded dissent. [researcher inference · moderate] Bancroft, printed pp. 62–63 CHM, p. 19
5. Networks are both sourcing advantage and access risk
The firm’s shifting 180/450/525/500-plus counts accompany a consistent referral story: portfolio executives filter and introduce opportunities. Repetition supports narrative consistency, not the exact count. The mechanism can exclude outsiders, so analysts should measure route, response, conversion, and outcome by source rather than accept a meritocracy claim. [researcher inference · strong] Red Herring, sourcing answer WBUR, 25:51–26:58 Wharton, closing Q&A
6. Company building precedes financial craft
Perkins attributes his edge to building University Laboratories and operating at HP. Technical adjacency enabled questions at Tandem, Genentech, and Acuson, while board confidence came from operating responsibility. The corpus does not establish that every good investor must found a company or that operating experience transfers across sectors. [researcher inference · moderate] Bancroft, printed pp. 25–31 and 38–57 CHM, pp. 5–21
7. Power-law memory requires denominator discipline
Tandem and Genentech dominate the story, while Snow-Job, Apple, ordinary outcomes, and luck appear less often. Fund I is variously described as 17 or 18 companies; later firm counts change definitions; no personal attribution ledger or net-LP cash-flow series is public. Famous-survivor repetition therefore cannot prove hit rate or alpha. [researcher inference · strong] Bancroft, printed pp. 32–33 WBUR, 02:55–03:56 and 31:51–33:08
Evolution, tensions, and falsification
| Tension | Spoken evidence | Required control |
|---|---|---|
| Idea-first vs people-intensive execution | Perkins starts with the opportunity but spends years recruiting, mediating, and replacing leaders. [researcher inference · strong] Genentech, printed pp. 16–18 and 42–47 | Decision-date team gaps, recruiting actions, founder/employee accounts, and outcome denominator. |
| Cheap proof vs Apple | Staging works in Genentech; Apple was dismissed because its key edge was not treated as a testable technical risk. [researcher inference · strong] Genentech, printed pp. 4–7 Bancroft, printed p. 61 | Pass ledger, rejected-deal memos, and false-negative rate by risk type. |
| Hands-on help vs effective control | Crisis work and follow-on signaling can support companies while concentrating authority. [researcher inference · moderate] Bancroft, printed pp. 62–65 | Board minutes, reserved matters, financing alternatives, founder accounts, and ordinary cases. |
| LP fairness vs personal participation | The missing plan purportedly banned cherry-picking; Perkins describes personal participation in Genentech R&D partnerships with disclosed conflict. [researcher inference · strong] Bancroft, printed pp. 33–35 and 56–57 | Plan text, partnership agreements, approvals, allocations, and participant economics. |
| Field-building claim vs collective causation | Perkins says “we” started biotechnology and sometimes Compaq; source controls preserve Swanson, Boyer, scientists, operators, Doerr, Sevin Rosen, and others. [researcher inference · strong] WBUR, 03:47–04:01 and 37:22–38:53 Genentech founders | Contemporaneous formation records and role-by-role decision maps. |
| HP private process vs whistleblower memory | Decision-near direct speech favored a private approach before full-board escalation; later summaries compress the chronology. [researcher inference · strong] Charlie Rose, 21:42–26:08 SEC order ¶¶4–9 | SEC order, HP 8-K, hearing record, board evidence, and exact dates. |
| Capital-supply forecast vs outcome | Perkins forecast excess capital and price inflation without defining vintage, return measure, or horizon. [investor-stated · provisional] Charlie Rose, 09:52–10:40 WBUR, 34:31–35:15 | Predefined market series, fund-vintage cash flows, valuation data, and a scoring date. |
Biography and decision turns
- Technical formation: scholarship-enabled MIT engineering, HBS/Doriot, and HP supplied Perkins’s self-described prepared mind. [investor-stated · moderate] Bancroft, printed pp. 5–25
- Operator bridge: University Laboratories converted an internal inventor into a founder, capital allocator, supervisor, and seller. [researcher inference · moderate] Bancroft, printed pp. 25–31
- Institution formation: Perkins and Eugene Kleiner combined complementary operating/technical experience and outside commitments; later partners and vehicles retain separate credit. [researcher inference · strong] Bancroft, printed pp. 31–40
- Company formation: Tandem and Genentech made staged company-building the house story, but the company-specific contributors are indispensable. [researcher inference · strong] Bancroft, printed pp. 38–51 Genentech, printed pp. 3–10
- Governance turn: long board tenures and later HP/Compaq/News Corp roles moved Perkins toward public-company governance, where process and independence became as important as technical judgment. [researcher inference · moderate] Bancroft, printed pp. 62–65 Charlie Rose, 21:42–26:08
- Late public persona: memoir promotion, yachts, the 2014 wealth letter, and tax-weighted-voting comments are judgment evidence, not investment-performance evidence. [researcher inference · strong] CBS profile 2014 writing/adverse note
Decision playbook
- Write the maximum-upside case before admiring the technology. Name the customer, order-of-magnitude improvement, barrier, and realistic scale; reject a clever product whose successful outcome still cannot matter. [researcher inference · strong] Red Herring, failure and swing-for-the-fence answers Wharton, opening paragraphs
- Name one decisive uncertainty. Classify it as technical, science, market, customer, legal, patent, manufacturing, or people risk; state what evidence would change the decision. [researcher inference · strong] WBUR, 06:58–07:57 and 23:23–31:38
- Buy the smallest discriminating proof. Set cost, deadline, owner, pass/fail result, and the action after each outcome; keep organization and burn below the uncertainty-resolution rate. [researcher inference · strong] Genentech, printed pp. 4–7 WBUR, 23:23–25:23
- Increase capital only when evidence improves. Record which risk fell, which new risk emerged, why valuation changes, and how reserves remain sufficient. [researcher inference · strong] WBUR, 29:31–31:38
- Recruit to the proved opportunity. Treat missing executives as a staged task, while testing integrity and learning ability before technical success becomes a character proxy. [researcher inference · moderate] Wharton, closing Q&A CHM, pp. 20–21
- Map decision rights before taking the board seat. Separate formal ownership, board votes, follow-on signaling, reserved matters, and practical coercion; create independent escalation routes. [researcher inference · moderate] Bancroft, printed pp. 62–65
- Run attribution before telling the story. Record who sourced, invented, validated, approved, operated, recruited, financed, governed, and exited; “we” is not a role. [researcher inference · strong] Genentech, printed pp. 3–10 and 36–41 WBUR, 37:22–38:53
- Maintain a loss, pass, and luck ledger. Include cheap failures, later winners rejected, ordinary outcomes, rescue by external events, and the contemporaneous reason. [researcher inference · strong] WBUR, 12:35–15:43 and 31:51–33:08 Bancroft, printed p. 61
- Audit conflicts as implemented behavior. Inspect allocation, audit, co-investment, related-party participation, disclosure, and approval records rather than relying on an ethical origin story. [researcher inference · strong] Bancroft, printed pp. 33–35 and 56–57
Modern VC translation — as of 2026-08-23
- AI and deep tech: fund a benchmark that tests the claimed technical or cost discontinuity before scaling a sales organization; define model quality, inference cost, reliability, security, customer workflow, and incumbent response separately. [researcher inference · provisional] Genentech staging, printed pp. 4–7
- Market-creation risk: Apple is the guardrail. When the edge is distribution, design, or category creation, replace a technical proof with observable retention, willingness to pay, channel economics, or community formation; do not reject a case merely because risk is not laboratory-shaped. [researcher inference · provisional] Bancroft Apple account, printed p. 61
- Capital abundance: separate company quality from entry price. Run ownership, dilution, reserve, liquidation, exit-value, and time-to-liquidity cases; no opportunity is large enough at every price. [researcher inference · provisional] Charlie Rose, 09:52–10:40 Creative Capital, bubbles/value questions
- Founder governance: pair high-involvement support with independent directors, information rights, documented dissent, and financing alternatives; crisis usefulness does not excuse unchecked control. [researcher inference · provisional] Bancroft, printed pp. 62–65 Charlie Rose, 21:42–26:08
- Sourcing: measure referral concentration and cold-inbound conversion. A portfolio network is an advantage only if it continues to surface non-consensus evidence rather than clone the incumbent network. [researcher inference · provisional] WBUR, 25:51–26:58 Wharton, closing Q&A
- Falsification metric: report how often the decisive-risk experiment predicted success, failed cheaply, produced a false positive, or excluded a later winner. [researcher inference · provisional] Bancroft Apple and Fund I evidence, printed pp. 32–33 and 61 WBUR, 12:35–15:43
Best public listening and reading order
- Bancroft/NVCA oral history — best complete biography, Fund I, Tandem, Genentech, Apple, boards, and philosophy spine; read printed pp. 31–65 first. [researcher inference · strong] transcript, printed pp. 31–65
- Computer History Museum oral history — best concise separate official transcript for University Laboratories, the risk algorithm, control, and Acuson; read pp. 17–21. It is not narrator-independent corroboration. [researcher inference · strong] transcript, pp. 17–21
- Genentech oral history — best company-specific formation, staging, licensing, financing, and attribution record. [researcher inference · strong] transcript, printed pp. 3–41
- WBUR On Point — best full public audio; prioritize 02:55–07:57, 12:35–15:43, 23:23–38:53, and 43:23–45:31. [researcher inference · strong] official episode
- Red Herring 1993 interview — best early direct control before the memoir-era narrative hardened. [researcher inference · moderate] 1993 Q&A
Then use Charlie Rose for dated forecasts and HP chronology, CBS for adversarial personality/governance context, and Wharton for a compact later statement of the method. [researcher inference · strong] Charlie Rose transcript CBS report Wharton report
Retrieval limits and exclusions
- No inaccessible full text, audio, or archive body was reconstructed from snippets. [researcher inference · strong] access-boundary ledger
- Bancroft-restricted interviews are paraphrased; short quotations elsewhere remain source-limited. [contemporaneous record · strong] Bancroft rights/access note
- Machine-transcribed WBUR wording is not treated as a publication transcript; timestamps route verification. [researcher inference · strong] WBUR access note
- Edited reports do not reveal omitted questions or exchanges. [researcher inference · strong] CBS access note Wharton access note
- The 15-family denominator measures research coverage, not completeness of every appearance Perkins ever made. [researcher inference · strong] source-map denominator
- Talks generate hypotheses for investment analysis; they do not establish audited fund returns, personal cash flows, or sole causal credit. [researcher inference · strong] Bancroft economics boundary Genentech economics boundary
Seek order-of-magnitude improvement with venture-scale upside
- Rule — Start with a customer problem important enough to support an exceptional company and demand a discontinuous improvement, not merely an ingenious product. Before admiring the technology, write both the maximum-upside case and the reason an incumbent cannot deliver the same result. [researcher inference · strong]
- Evidence status — Perkins repeatedly states the importance and dramatic-improvement tests; Tandem, Genentech, and Acuson are compatible documented behaviors, while Snow Job is his own counterexample. The rule's articulation is [investor-stated · strong]; its portfolio efficacy is [researcher inference · moderate] because no complete deal or return denominator survives. Red Herring, “biggest hits,” “elements to success,” and failure answers
- Provenance and origin — Perkins retrospectively traced the heuristic to David Packard's demand for products materially better than alternatives; the earliest substantial public articulation recovered is the 1993 Red Herring interview. Wharton's 2012 report explicitly connects the dramatic-improvement heuristic to his later venture method. [investor-stated · moderate] Wharton, opening and “Getting the Right Answers” Red Herring, “elements to success”
- Mechanism and reasoning — A material customer discontinuity can support adoption despite startup risk, while a large attainable market gives one success enough weight to offset venture losses. A technically working product with a small ceiling consumes the same scarce partner attention without a fund-shaping payoff. [researcher inference · strong] WBUR, 00:12:35–00:15:43
- Historical and vehicle context — Perkins recalled the 1972 partnership as $8 million committed and roughly $4 million deployed, with Tandem and Genentech dominating the favorable narrative. In such a small, concentrated vehicle, the successful case had to be very large, but the missing cash-flow schedule prevents a numeric hurdle or outlier share. [investor-stated · moderate] Bancroft/NVCA, printed pp.31–38 Denominator reconstruction
- When to use — Use at initial screen and again before each scale round in deep technology, infrastructure, hardware, life sciences, and other cases where technical novelty can distract from customer value. Do not use it as a slogan to waive price, adoption, regulation, capital intensity, or distribution risk.
- How to apply it — State the customer and current alternative; quantify the claimed performance/cost/reliability improvement; name the durable barrier; build a bottom-up reachable market; calculate whether realistic ownership and dilution could matter to the fund; and kill or redesign any case whose success remains immaterial. [researcher inference · strong]
- Examples —
- Genentech — documented application: [documented behavior · strong] K&P financed a path toward recombinant production of human proteins, a potential platform rather than one narrow laboratory instrument; the 1980 filing independently documents several products, licenses, partnerships, and substantial scientific/regulatory risk. Genentech S-1, “Business,” “Products,” and risk factors, printed pp.115–132
- Tandem — researcher-applied inference: [researcher inference · strong] continuous transaction processing made downtime economically important, while modular fault tolerance promised a system-level discontinuity; the decision-near article still showed an engineering prototype, not proven customer adoption. BusinessWeek, article pp.64–65
- Acuson — researcher-applied inference: [researcher inference · strong] Perkins probed whether Maslak's acoustic-diffraction architecture created a genuine imaging advantage rather than accepting the plan's projections. CHM oral history, pp.20–21
- Snow Job — documented application (counterexample): [investor-stated · moderate] Perkins later conceded that even a functioning motorcycle conversion would not have created a company of Genentech-like importance. WBUR, 00:12:35–00:15:43 Failure lab
- Evidence — The rule has more than four useful citation placements across independent direct voice, decision-near press, an issuer filing, and an official oral history: Red Herring, scale/risk answers, Wharton, opening paragraphs, Tandem decision-near record, Genentech issuer record, and CHM Acuson account.
- Limits, failure modes, and exception ledger — “Large” can become an unfalsifiable story; an order-of-magnitude technical gain may not produce willingness to pay; realistic ownership may be too small; regulation or distribution can absorb the benefit; and a small market may still suit a smaller vehicle. Snow Job exposes a low ceiling, while Apple shows that a major opportunity may look technically ordinary at intake. Missing Fund I ownership and cash flows prevent proof that Perkins consistently enforced the screen. Apple miss lab Snow Job lab
- Evolution and contradictions — The scale language is stable from the 1993 interview through later talks, but Perkins's early consumer bets show it was not fully formed or consistently applied from the partnership's first day. Apple is the inverse error: a scale opportunity was dismissed before the product and founders were inspected. [researcher inference · strong] TRED 2 and Snow Job controls Bancroft Apple account, printed p.61
- Modern VC translation (as of 2026-08-23) — Transfers: retain the maximum-upside screen. Changed assumptions / what fails: a literal technical-gain threshold alone does not establish adoption, distribution, price, ownership, or fund relevance. Observable signals: a dated baseline, reproducible gain, paid retention, bottom-up reachable market, and dilution-adjusted ownership. Likely misuse: calling a benchmark win venture-scale. Adaptation: connect product change to customer economics and the fund return case. Analyst-created hypothetical: an AI-inference company must show reproducible quality/cost/latency improvement in its target workflow and paid retention. [researcher inference · provisional] Historical premise: Wharton, opening and “Getting the Right Answers” Denominator boundary
Name the decisive risk and buy the smallest discriminating proof
- Rule — Decompose an apparently high-risk company into named technical, scientific, customer, market, legal, patent, manufacturing, and people uncertainties; fund the cheapest time-bounded experiment that can falsify the most decisive one before financing the integrated company. [researcher inference · strong]
- Evidence status — The generic algorithm is [investor-stated · strong], and Tandem and Genentech provide [documented behavior · strong]. Its universal efficacy is unproven because the public record lacks a pass ledger, false-positive rate, and ordinary-deal denominator. CHM oral history, pp.18–19
- Provenance and origin — Decision-near Tandem behavior preceded the public formula: “We try to isolate whatever the biggest risk is in a given deal.” [investor-stated · strong] BusinessWeek, article p.64 Red Herring, “risk-formula” answer
- Mechanism and reasoning — Information has option value. A small experiment limits loss if the thesis is false, increases justified valuation if true, and prevents premature headcount and fixed-cost buildup. The experiment must discriminate between hypotheses; activity, a demo without controls, or elapsed time is not proof. [researcher inference · strong] WBUR, 00:23:23–00:31:38
- Historical and vehicle context — Scarce 1970s venture capital, a small first partnership, immature recombinant-DNA commercialization, and expensive computer hardware rewarded outsourcing and staged commitments. The historical advantage partly depends on being able to isolate a risk before a full organization exists. Federal Reserve historical context Genentech formation record
- When to use — Use when one uncertainty dominates and a test can change the decision within the company's runway. Do not force this model onto reflexive category creation, network effects, regulated systems whose risks interact, or products whose value appears only after a full-system build.
- How to apply it — Write the hypothesis and competing explanation; name one decision-changing metric, owner, budget, and deadline; pre-register pass, fail, and ambiguous actions; cap the organization needed for the test; and specify the check, valuation, and newly dominant risk after each result. [researcher inference · strong]
- Examples —
- Tandem — documented application: [documented behavior · strong] K&P first financed recruiting and architecture; the larger commitment followed a workable design, while production and customer trust remained for later stages. BusinessWeek, article pp.64–65 Tandem lab
- Genentech — documented application: [documented behavior · strong] the team used outside laboratories and a shorter peptide target to test recombinant expression before an integrated pharmaceutical build; the S-1 later records somatostatin and insulin milestones. Perkins Genentech oral history, printed pp.4–7 Genentech S-1, business chronology
- Hybritech — documented application: [retrospective witness · strong] Byers reports $300,000 of initial capital and a six-month monoclonal-antibody milestone; he also admits the contemporary diligence base was thin. Byers oral history, PDF pp.27–28
- Apple — researcher-applied inference (documented counterexample): [researcher inference · strong] Perkins did not take the meeting because the opportunity did not fit his technology-risk screen; no cheap information gate tested whether the category prior was wrong. CHM panel, 01:08:03–01:08:43 Apple lab
- Evidence — Independent support includes Tandem's decision-near article, Genentech's issuer filing, Byers on Hybritech, Perkins's CHM generic algorithm, and WBUR's risk taxonomy, 00:23:23–00:31:38.
- Limits, failure modes, and exception ledger — A convenient experiment can test the wrong risk; milestones can be gamed; sequential capital can underfund platform work; a negative may be a false negative; and multiple interacting uncertainties may not be cheaply separable. Apple is the clearest false negative, Snow Job leaves exogenous/channel risk poorly bounded, and the unnamed satellite story shows outcome luck. Snow Job lab Satellite evidence boundary
- Evolution and contradictions — Tandem and Genentech show the method before Perkins articulated it publicly. Later interviews make it increasingly clean and general, creating hindsight risk. Acuson's later roadmap dispute also shows that one early technical proof does not eliminate product-cadence and market risk. CHM Acuson account, p.21
- Modern VC translation (as of 2026-08-23) — Transfers: isolate the decision-changing uncertainty and buy falsifiable evidence before scale. Changed assumptions / what fails: one cheap test fails where risks interact or value appears only in the full system. Observable signals: a ranked hypothesis, live rival explanation, pre-registered threshold, budget, deadline, and next action. Likely misuse: milestone theater or choosing the easiest test rather than the decisive one. Adaptation: stage capital against sector-specific evidence and re-rank risk after every result. Analyst-created hypothetical: release an AI-agent round against task success, inference cost, escalation rate, and paid workflow retention rather than demo quality. [researcher inference · provisional] Historical premise: WBUR, 00:23:23–00:31:38 Tandem decision-near control
Scale capital and build missing capabilities only after proof
- Rule — Treat the initial opportunity as a nucleus, not a finished company. After the decisive proof improves, add the managers, laboratories, commercial partners, manufacturers, co-investors, and capital required for the next bottleneck—without rewriting their contribution as the investor's. [researcher inference · strong]
- Evidence status — The idea-first/team-in-sequence formulation is [investor-stated · strong] and the capability sequence is [documented behavior · strong] in Tandem and Genentech. Causal outcome attribution remains moderate because operators, scientific institutions, partners, and markets were indispensable. Red Herring, people and role answers
- Provenance and origin — Perkins says the original K&P plan committed the partners to earliest-stage formation and team building; the plan itself is unavailable. The 1975 Tandem article is the earliest decision-near control, showing named technical, finance, and operating recruits assembled around Treybig's market concept. CHM oral history, pp.17–18 BusinessWeek, article pp.64–65
- Mechanism and reasoning — Delaying full organizational build reduces burn and keeps role specifications tied to evidence rather than a generic org chart. External specialists can transform fixed capital needs into contracts, add independent validation, and cover capabilities an investor or founder lacks. [researcher inference · strong] Genentech oral history, printed pp.4–13
- Historical and vehicle context — K&P's first partnership lacked the capital to vertically integrate every technical venture. Contract research, pharmaceutical licenses, co-investor syndicates, and recruited executives allowed a small fund to hold options on much larger companies. The same structure also dispersed credit and created contractual dependencies. Genentech S-1, “Collaborative Arrangements” and risk factors
- When to use — Use when the opportunity is strong but a complete team or integrated supply chain would be premature. Do not defer a capability that is itself the decisive risk, or assume a missing CEO, regulator, manufacturer, or channel can always be recruited after technical proof.
- How to apply it — Map the capability required at each milestone; identify build, buy, contract, partner, and syndicate alternatives; assign decision rights and IP; set recruiting triggers; budget integration cost; and record what each contributor—not only the lead investor—must deliver. [researcher inference · strong]
- Examples —
- Tandem — documented application: [documented behavior · strong] the sequence recruited Green and Katzman for architecture, Loustaunou for finance, then external investors for prototype-to-market scale; issuer records later verify Perkins, Treybig, and Pitch Johnson board tenures. BusinessWeek, pp.64–65 Tandem 1991 proxy, printed pp.2–5
- Genentech — documented application: [documented behavior · strong] Swanson and Boyer formed the company, outside laboratories produced early proof, Lilly supplied commercialization capacity, and later executives supplied finance and operations; K&P chaired and financed rather than performing all functions. Genentech founders S-1, products and collaborative arrangements
- Hybritech — documented application: [retrospective witness · strong] Byers became founding CEO, Ivor Royston supplied scientific access, Perkins mentored and joined the board, Ted Greene later operated, and Pitch Johnson co-invested. Byers oral history, PDF pp.27–28 and 37–39 Perkins oral history, printed pp.58–59
- LSI Logic — documented application (adverse case): [retrospective witness · moderate] founders changed the technology plan after customer calls without clearly informing VCs, showing that staged capability building fails when information and decision rights lag the operating team. LSI oral-history panel, PDF pp.12–14
- Evidence — The model clears the breadth target through Tandem decision-near reporting, Tandem's issuer proxy, Genentech's S-1, Byers's Hybritech account, and the LSI participant panel.
- Limits, failure modes, and exception ledger — Recruiting may fail; contractors can hold critical IP; partners can acquire bargaining power; syndication can blur accountability; staged financing can create founder dependence; and a small team can miss regulatory, safety, manufacturing, or customer work. Hybritech's Byers explicitly recalls thin diligence and luck, while LSI shows an information gap between board and operators. Byers, PDF pp.27–28
- Evolution and contradictions — Perkins's later retellings increasingly center the investor's team-building role, while the company records make the distributed system visible. Acuson complicates a founder-replacement default because Maslak reportedly grew into the CEO role; Apple shows that refusing to meet an incomplete team can destroy the option to build one. CHM Acuson, pp.20–21 Apple lab
- Modern VC translation (as of 2026-08-23) — Transfers: sequence people, partners, and capital behind evidence. Changed assumptions / what fails: delaying a capability that is itself the decisive safety, regulatory, manufacturing, or distribution risk can destroy the option. Observable signals: each milestone has a named owner, build/buy/partner choice, integration cost, dependency, and recruiting trigger. Likely misuse: disguising chronic understaffing as capital discipline or crediting the investor for work done by operators and partners. Adaptation: review the capability map with every evidence gate. Analyst-created hypothetical: after a robotics team proves manipulation reliability, release the scale round only with manufacturing leadership, a safety-certification plan, and a second-source roadmap. [researcher inference · provisional] Historical premise: Genentech S-1, collaborative arrangements Adverse control: LSI panel, PDF pp.12–14
Use operator and technical fluency to interrogate the mechanism
- Rule — Use relevant operating and technical knowledge to get beneath forecasts and identify the physical, scientific, product, customer, and organizational mechanism; then seek outside expertise where personal fluency ends. Never treat success in one domain as a general warrant. [researcher inference · strong]
- Evidence status — Perkins consistently attributes his style to University Laboratories and HP, and Acuson supplies a detailed self-reported application. Tandem and Genentech show technically literate participation but also indispensable specialist work. [investor-stated · moderate] for origin; [documented behavior · moderate] for repeatable application. CHM oral history, pp.5–21
- Provenance and origin — Perkins built and supervised University Laboratories while employed at HP, then worked across HP engineering, sales, division management, budgeting, and internal company formation. He presents this as preparation for venture work; chronology supports exposure, not causality by itself. Bancroft/NVCA, printed pp.10–31 CHM, pp.5–17
- Mechanism and reasoning — Mechanism-level questions can expose an attractive spreadsheet attached to weak physics, a scientific result without rights or scale, or a founder who cannot explain the causal chain. Operating memory can also identify recruiting, manufacturing, financing, and board risks earlier—but only inside a relevant domain. [researcher inference · strong] CHM Acuson account, pp.20–21
- Historical and vehicle context — Early venture firms were small and expected partners to perform product, recruiting, financing, and board work themselves. Perkins's HP adjacency was unusually relevant to minicomputers, lasers, and electronic instrumentation; biotechnology required scientists, contract laboratories, and pharmaceutical partners beyond his personal competence. Genentech S-1, business and risk sections
- When to use — Use during technical diligence, founder references, roadmap review, manufacturing plans, and board crises where the investor has current relevant competence. Do not use old operator status to overrule specialists, cross into unrelated political/historical judgment, or replace independent validation.
- How to apply it — Draw the causal mechanism; ask what observation would refute it; inspect raw test and customer evidence; identify the expert whose knowledge dominates yours; write the boundary of your competence; and record whether the founder changes the model when evidence changes. [researcher inference · strong]
- Examples —
- University Laboratories — researcher-applied inference (documented formative behavior): [researcher inference · moderate] Perkins reports conceiving a lower-cost laser, obtaining Packard's permission, using household capital, and supervising the company; the amount and merger economics conflict or remain unavailable. CHM, pp.5–6 Bancroft, printed pp.18–21
- Tandem — researcher-applied inference: [researcher inference · strong] Perkins had relevant HP computer experience, but Treybig framed the market and Green/Katzman designed the system; the decision-near record prevents technical credit from collapsing into the financier. BusinessWeek, pp.64–65
- Acuson — documented application (investor-stated): [investor-stated · strong] Perkins disregarded the plan's large projections and questioned Maslak about acoustic diffraction until he understood the claimed imaging difference. CHM, pp.20–21
- 2014 wealth letter — researcher-applied inference (documented domain-transfer counterexample): [researcher inference · strong] a provocative historical analogy and tax-weighted-voting remarks show that technical or board success does not validate reasoning in an unrelated political domain. New Yorker, paragraphs beginning “On Saturday” TIME, voting and retraction paragraphs
- Evidence — Independent placements include Tandem's decision-near account, Genentech's S-1, CHM on University Laboratories and Acuson, Bancroft on HP experience, and contemporaneous 2014 controls.
- Limits, failure modes, and exception ledger — Fluency can become arrogance, pattern overreach, or anchoring to the technology of a prior era. Self-reported interrogation does not show reject accuracy; one favorable Acuson case does not establish a base rate; and the public record lacks ordinary losing deals where Perkins's technical confidence was wrong. CHM, p.18 and pp.20–21
- Evolution and contradictions — Early hands-on operating work plausibly preceded the style, while later public accounts make that origin increasingly coherent. Genentech shows good boundary use through external scientists; Apple shows category knowledge becoming a reason not to inspect; the 2014 episode supplies the clearest unrelated-domain boundary. [researcher inference · strong] Genentech founders Apple lab
- Modern VC translation (as of 2026-08-23) — Transfers: interrogate the mechanism and write the boundary of investor competence. Changed assumptions / what fails: prior operating success does not validate judgment in a new technical, regulatory, or political domain. Observable signals: raw evidence, a falsifier, an independent specialist, documented expert dissent, and an explicit competence boundary. Likely misuse: authority substitution—treating operator status as proof. Adaptation: require independent validation before technical conviction enters IC. Analyst-created hypothetical: a former application-software operator reviewing a foundation-model company relies on reproducible model and security evaluations rather than intuition about training science. [researcher inference · provisional] Historical premise: CHM, pp.5–21 Domain-transfer control
Make board service crisis work, with explicit formal and shadow decision rights
Rule — Treat equity as payment for recruiting, financing, strategy, and crisis response, but map formal board authority separately from the shadow control created by follow-on signaling, information asymmetry, reputation, and financing dependence. Make bad news travel fast and preserve independent escalation. [researcher inference · strong]
Evidence status — Perkins's board-as-product and follow-on-control claims are [investor-stated · strong]; Tandem and Genentech establish long formal roles, while Acuson and HP expose disagreement and process limits. Founder/employee satisfaction and ordinary-board outcomes remain [researcher inference · insufficient public record]. Red Herring, role and hardest-part answers
Provenance and origin — In 1993 Perkins said: “Once we sign the check, we’re on the team.” [investor-stated · strong] Red Herring, “competitive-advantage” answer Later accounts add the follow-on-signaling control claim. Bancroft/NVCA, printed pp.62–65 CHM, pp.19–20
Mechanism and reasoning — A director who knows the company and can mobilize capital, executives, or partners may shorten crisis response. The same credible threat to withhold follow-on support can coerce founders or minority holders even when the term sheet gives no explicit control, so helpful involvement and concentrated power are inseparable design questions. [researcher inference · strong]
Historical and vehicle context — Thin early management benches, scarce private capital, concentrated boards, and long pre-liquidity periods made partner labor unusually consequential. Public-company boards add disclosure, investigation, independence, and shareholder duties that cannot be managed as an informal founder relationship. SEC HP order, ¶¶1–15
When to use — Use before taking a board seat, during financing gaps, leadership changes, product-roadmap disputes, investigations, and exits. Do not equate availability with competence or permit “help” to bypass recusal, minutes, counsel, reserved matters, or independent review.
How to apply it — Map ownership, votes, reserved matters, information rights, follow-on alternatives, and reputational leverage; define bad-news and whistleblower routes; pre-agree crisis cadence and recusals; record dissent and decision owner; and test whether the company can finance if the lead abstains. [researcher inference · strong]
Examples —
- Tandem — documented application: [documented behavior · strong] the 1991 proxy verifies Perkins as director and chair from 1974, while decision-near reporting and participant accounts attribute recruiting, financing, and strategy across Perkins, Treybig, the technical team, Byers, and Johnson. Tandem proxy, printed pp.2–5 BusinessWeek, pp.64–65
- Genentech — documented application: [documented behavior · strong] the filing and oral histories support Perkins's chair/financing role but preserve Swanson, Boyer, scientists, management, licensees, and partners as separate decision centers. Genentech S-1, management and business sections Perkins Genentech oral history, printed pp.16–23
- Acuson — documented application (investor-stated adverse case): [investor-stated · moderate] Perkins opposed a five-to-seven-year product gap and resigned when Maslak rejected incremental integration; missing minutes prevent the later sale from proving his preferred roadmap. CHM, p.21 Acuson case boundary
- HP — researcher-applied inference (contemporaneous governance counterexample): [researcher inference · strong] the SEC record shows Perkins first preferred private treatment of a director leak, resigned after the board vote, and only later investigated pretexting; costly dissent does not establish procedural correctness. HP corrective 8-K, Item 8.01 SEC order, ¶¶6–15
Evidence — The model has independent company, regulatory, decision-near, and direct-voice controls: Tandem proxy, Genentech S-1, HP 8-K, SEC order, and Perkins's 1993 role statement.
Limits, failure modes, and exception ledger — The record has no founder/employee denominator, ordinary board evaluations, reserved-matter schedules, or financing-counterfactual series. Shadow control can make reporting fast or suppress dissent; long tenure can create knowledge or entrenchment; resignation can be principled or impulsive. Model-specific routes attempted include issuer proxies/filings, HP's 8-K and SEC order, CHM/Bancroft oral histories, and deal archives; early minutes, term sheets, founder surveys, and follow-on decision logs remain unavailable.
Evolution and contradictions — Early company-formation stories celebrate informality and intervention; Acuson shows strategy dissent; HP shows that public-company procedure, investigation, disclosure, and proportionality can conflict. Perkins's private-resolution preference at HP is narrower than a later simple whistleblower story. [researcher inference · strong] Charlie Rose, 21:42–26:08 SEC order, ¶¶4–9
Modern VC translation (as of 2026-08-23) — Transfers: treat governance, recruiting, financing, and crisis response as investment work while mapping shadow power. Changed assumptions / what fails: formal votes alone do not capture leverage created by financing dependence, information, and reputation. Observable signals: credible financing alternatives, independent review, minutes, recusals, escalation routes, and documented dissent. Likely misuse: renaming coercion or shadow management “active governance.” Adaptation: put shadow-control analysis beside the cap table before crisis. Analyst-created hypothetical: before leading a bridge, the investor records whether non-participation would collapse the company and assigns the decision to independent directors. [researcher inference · provisional] Historical premise: Bancroft/NVCA, printed pp.62–65 Governance control: SEC order, paragraphs 1–15
Use referral networks as filters, then audit access and source concentration
- Rule — Use trusted founders, executives, partners, and co-investors to reduce search and reference costs, but track every source route and test whether the network is excluding unfamiliar people or reinforcing a category prior. A referral is a lead, not diligence. [researcher inference · strong]
- Evidence status — Perkins repeatedly describes referral sourcing and firm-network compounding; Qume, Hybritech, VLSI, and Compaq independently show multi-person routes. The sourcing mechanism is [documented behavior · strong]; superiority to open intake is [researcher inference · insufficient public record] because source-by-source conversion and outcome data are absent. Red Herring, sourcing answer
- Provenance and origin — The earliest public direct formulation recovered is the 1993 interview. Draper's CHM oral history adds a concrete early-network episode: he says he brought Qume to K&P after three weak referrals and David Lee founded it. Draper oral history, PDF p.15 / printed p.16
- Mechanism and reasoning — A credible referrer supplies context, reputation, and cheaper references; repeated portfolio relationships can surface opportunities before a broad process. But homophily, reciprocity, prestige, and category fatigue can turn the filter into an access moat that mistakes familiarity for merit. [researcher inference · strong] Wharton, closing Q&A
- Historical and vehicle context — Early Silicon Valley venture firms were few and shared deals, diligence, and boards because single funds were small. Later Perkins firm counts span multiple vehicles and generations; they evidence a large institutional network, not his personal sourcing denominator. Draper oral history, PDF pp.15–16 Firm/count reconciliation
- When to use — Use for discovery, references, specialist access, and syndication. Do not use a trusted source to skip product evidence, founder meetings, conflict checks, or cold-inbound review.
- How to apply it — Tag referrer, relationship, incentive, route, response time, meeting, diligence, decision, and outcome; require one independent evidence route; compare warm and cold conversion; audit demographic/geographic/category concentration; and preserve a cheap exception-review gate. [researcher inference · strong]
- Examples —
- Qume — documented application: [retrospective witness · strong] Draper says he brought David Lee's Qume to K&P and that it later sold to ITT at a large gain; the account assigns sourcing outside Perkins. Draper oral history, PDF p.15
- Hybritech — documented application: [retrospective witness · strong] Byers followed Ivor Royston for two months, brought the idea to Perkins, received encouragement, and then led the company with a shared board. Byers oral history, PDF p.37
- Compaq — documented application: [documented behavior · strong] Kleiner Perkins's own history says Sevin Rosen approached the firm and John Doerr sponsored the investment; Perkins's supported contribution belongs to later board/M&A work. Kleiner Perkins, “Compaq,” paragraphs 1–5
- Apple — researcher-applied inference (documented intake counterexample): [researcher inference · strong] Perkins says category fatigue prevented him from taking the meeting; the source route and exception signals are unknown, but the intake filter destroyed the option to learn. CHM panel, 01:08:03–01:08:43
- Evidence — Independent participant and firm records provide more than four placements: Draper on Qume, Byers on Hybritech, Kleiner Perkins on Compaq, CHM's VLSI founder history, and Perkins on referral sourcing at Wharton.
- Limits, failure modes, and exception ledger — The record lacks a sourced-opportunity denominator, cold-inbound log, declined-deal ledger, source-quality score, and founder-access comparison. Qume's “large gain” is not audited economics; the firm Compaq page is promotional; and Apple lacks a referral record. A network may create access without selection skill, exclude outsiders, or pass along correlated errors.
- Evolution and contradictions — Early syndication was explicitly collaborative; later firm scale increased the claimed network while weakening Perkins-personal attribution. Perkins's meritocracy language conflicts with the closed referral mechanism unless open-route outcomes are measured. [researcher inference · strong] Draper, PDF pp.15–16 Wharton, closing Q&A
- Modern VC translation (as of 2026-08-23) — Transfers: use trusted referrals for discovery and context, then run independent diligence. Changed assumptions / what fails: network prestige does not establish channel superiority and can hide access concentration. Observable signals: source-tagged conversion, response time, progression, founder experience, geography, institution, category, and outcome. Likely misuse: calling a closed referral funnel meritocratic because its known winners are strong. Adaptation: compare warm, cold, outbound, scout, and university cohorts and preserve an exception gate. Analyst-created hypothetical: an AI fund audits every channel and expands outreach where non-network talent is systematically excluded. [researcher inference · provisional] Historical premise: Draper oral history, PDF pp.15–16 Perkins network statement: Wharton closing Q&A
Pursue asymmetric outcomes while preserving the loss, pass, luck, and ordinary denominator
- Rule — Build for a few outcomes large enough to dominate a venture portfolio, cap learning cost when proof fails, and preserve every loss, pass, ordinary result, and lucky rescue so later storytelling cannot turn a skewed distribution into a clean personal hit rate. [researcher inference · strong]
- Evidence status — Perkins explicitly describes a small number of exceptional outcomes and cheap failure; Fund I recollections consistently foreground Tandem and Genentech. The asymmetry is [investor-stated · strong], while exact concentration, IRR, hit rate, and Perkins-personal alpha are [researcher inference · insufficient public record]. CBS, two-in-ten passage Bancroft/NVCA, printed pp.32–38
- Provenance and origin — The earliest mapped public formulation is the 1993 Red Herring interview, where Perkins paired large-scale ambition with acknowledged failures. Later WBUR and Bancroft accounts make Tandem and Genentech the first fund's dominant favorable cases while admitting ordinary positions and failure. Red Herring, hits/failures answers WBUR, 00:02:55–00:03:56
- Mechanism and reasoning — Limited liability caps a company's loss while successful ownership can compound; staging further caps cash at risk. The cognitive danger is selection on visibility: outliers create archives and repeated stories, while ordinary results, insurance, timing, and team contribution disappear. [researcher inference · strong]
- Historical and vehicle context — Fund I is variously reported as 17 or 18 companies; Perkins questioned the higher count. He recalled $8 million committed, roughly $4 million deployed, and seven or eight non-outliers near break-even, but no audited capital calls or distributions are public. Bancroft/NVCA, printed pp.31–40 Count reconciliation
- When to use — Use in portfolio construction, reserves, follow-ons, fund attribution, postmortems, and performance claims. Do not use “power law” to excuse an undisciplined check, ignore ownership/price, or claim a numeric 80/20 split without a defensible denominator.
- How to apply it — Define vehicle and inclusion rule; list every company and follow-on; classify realized, unrealized, written off, passed, and excluded; record cash flows and ownership where available; separate skill from external rescue; rank economic, historical, and learning value separately; and rerun results without the dominant outlier. [researcher inference · strong]
- Examples —
- Genentech and Tandem — researcher-applied inference (documented company outcomes): [researcher inference · strong] company records establish scientific/product milestones and public outcomes, but no public fund ledger proves their exact share of Fund I value. Genentech S-1 Tandem 1979 annual report, stock-price section
- Snow Job — documented application (failure boundary): [investor-stated · moderate] Perkins calls the company a failure and later says its successful ceiling was too small; a secondary table reports zero 1984 value but cannot produce a realized multiple. WBUR, 00:12:35–00:15:43 Snow Job lab
- Apple — researcher-applied inference (documented pass): [researcher inference · strong] the firm did not collect enough information to evaluate the founders; no counterfactual ownership or return is admissible. Bancroft/NVCA, printed p.61
- Unnamed satellite — researcher-applied inference (investor-stated luck case): [researcher inference · provisional] a launch failure allegedly triggered insurance and reversed a large expected loss, but anonymity and missing cash flows prohibit ranking it. WBUR, 00:31:51–00:33:08
- Evidence — The breadth includes Genentech's issuer record, Tandem's issuer record, Bancroft's Fund I/pass account, WBUR's failure/luck segments, and the uniform deal ledger.
- Limits, failure modes, and exception ledger — Model-specific attempted routes included Bancroft/CHM oral histories, Fund I count variants, 1984 testimony, company filings, firm archives, later portfolio tables, and named-deal searches. Still missing are the executed partnership, audited fund statements, complete portfolio schedule, capital calls, ownership/dilution, distributions, carry, and ordinary-company identities. Therefore no personal IRR, TVPI, DPI, hit rate, loss ratio, or literal 80/20 share is supportable. Retrieval ledger
- Evolution and contradictions — Perkins's later accounts repeat the outliers but vary company and firm counts. He candidly calls the satellite outcome luck and Apple a mistake, which strengthens the need for a ledger; repeated hero stories still cannot show whether the method predicted unseen cases. [researcher inference · strong] Talks denominator analysis
- Modern VC translation (as of 2026-08-23) — Transfers: pursue asymmetric payoff while preserving the complete decision and cash-flow denominator. Changed assumptions / what fails: company valuation, firm participation, or one outlier cannot substitute for vehicle cash flows and role attribution. Observable signals: contributions, distributions, ownership, dilution, write-offs, passes, luck cases, and results with the top company removed. Likely misuse: invoking power law to excuse weak underwriting or erase ordinary outcomes. Adaptation: freeze the ledger beside narrative memos and report gross and net results by vehicle. Analyst-created hypothetical: an emerging manager shows results excluding its AI outlier plus false-negative and written-off cohorts before claiming repeatability. [researcher inference · provisional] Historical premise: CBS two-in-ten passage Searched denominator
Define the fund opportunity perimeter before quality is visible
- Rule — Decide which fund, account, partner, SPV, or affiliate owns an opportunity before its quality is known; prohibit preferential personal purchases and after-the-fact cherry-picking; audit and disclose every residual allocation conflict. [researcher inference · strong]
- Evidence status — Perkins's purported 1972 “Investor's Bill of Rights” is available only through his 2009 oral description. The rule is [investor-stated · moderate]; original text, contractual implementation, compliance, and industry causal impact are [researcher inference · insufficient public record]. Bancroft/NVCA, printed pp.32–35
- Provenance and origin — Perkins says he wrote the original business plan and that he and Kleiner rejected cheap pre-fund founder purchases, opportunity diversion, unaudited reporting, and selective personal co-investment. He also says the plan is lost, so this is a recollection of a text rather than an inspected clause. Bancroft/NVCA, printed pp.32–35, especially p.34
- Mechanism and reasoning — A GP sees opportunities before LPs and can observe quality before allocating. Ex-ante scope and equivalent economics reduce the option to privatize winners and leave weaker exposure with the fund; audits and independent review make deviations observable. [researcher inference · strong]
- Historical and vehicle context — The 1972 limited partnership combined LP capital, GP discretion, private information, carry, and illiquid securities. Perkins says he and Kleiner were also LPs, but common exposure does not prove equal fees, carry, tax, information, timing, or later-vehicle treatment. Bancroft/NVCA, printed p.40
- When to use — Use before fundraising and before sourcing across overlapping funds, personal accounts, scouts, opportunity funds, SPVs, continuation vehicles, and strategic affiliates. Do not substitute a same-headline-price claim for a review of timing, information, governance, fees, liquidity, and follow-on rights.
- How to apply it — Write eligibility and priority rules; time-stamp first presentation; compare security and all economics; disclose conflicts and recusals; require independent approval for deviations; audit annually; and preserve declined or reallocated opportunities. [researcher inference · strong]
- Examples —
- Purported K&P personal-investment ban — documented application (investor-stated; contract unverified): [investor-stated · moderate] Perkins says neither partner made a personal investment the fund could have made and that their own LP commitment provided participation; no agreement or audit tests compliance. Bancroft/NVCA, printed pp.34–35 and 40
- Cetus/Genentech — documented application (retrospective witnesses): [retrospective witness · moderate] K&P already held Cetus while Swanson explored recombinant DNA; Perkins and Byers say the Cetus position was sold as Genentech created a conflict. Security, price, approval, and proceeds remain unknown. Perkins Genentech oral history, printed pp.3–4 Byers oral history, printed pp.12–13
- Genentech R&D partnerships — researcher-applied inference (investor-stated adverse case): [researcher inference · moderate] Perkins describes personal participation with disclosed conflict, creating tension with a simple no-personal-investment rule; the public record lacks the approvals, allocation policy, and participant economics needed to resolve it. Bancroft/NVCA, printed pp.56–57
- Evidence — The model falls below the ideal three independently documented implementations. It has one rule source, a separate Perkins Genentech oral history, Byers's independent Cetus account, and Genentech's issuer filing, but no executed allocation policy or audit. [researcher inference · provisional]
- Limits, failure modes, and exception ledger — Model-specific routes attempted: Bancroft and CHM plan accounts; Genentech/Byers/Cetus histories; Genentech S-1; partnership/vehicle testimony; public portfolio and filing searches. Not found: the plan, partnership agreements, subscription papers, audit reports, allocation logs, personal-account records, R&D partnership agreements, conflict approvals, or enforcement history. “Anything the fund might invest in” can be vague, and formal equality can conceal better information or liquidity.
- Evolution and contradictions — Perkins claimed the rule became an industry standard, but no adoption series supports causality. Later personal participation in Genentech-related partnerships creates an unresolved boundary: it may have been disclosed and permitted, or may show that the original rule was narrower than the later story. [researcher inference · moderate] Writings synthesis
- Modern VC translation (as of 2026-08-23) — Transfers: define opportunity allocation before quality is visible. Changed assumptions / what fails: overlapping funds, SPVs, scouts, employee accounts, and continuation vehicles create conflicts that a same-headline-price test cannot resolve. Observable signals: time-stamped presentation, eligibility rule, security and economics comparison, recusal, independent approval, and audit trail. Likely misuse: writing a waiver after the winner becomes apparent. Adaptation: preassign priority and publish the deviation process to affected stakeholders. Analyst-created hypothetical: an AI fund permits an SPV only after its flagship declines under an independently documented process using identical information. [researcher inference · provisional] Historical premise: Bancroft/NVCA, printed pp.32–40 Conflict control: Byers, printed pp.12–13
Attribute every outcome role by role before judging repeatability
- Rule — Before crediting or blaming an investor, map who originated, sourced, sponsored, diligenced, approved, financed, invented, operated, recruited, governed, authorized follow-ons, and influenced exit. Firm brand, narration, and a later board seat are not substitutes for a role. [researcher inference · strong]
- Evidence status — Repeated conflicts across participant, firm, company, filing, and decision-near records make the attribution control [researcher inference · strong]. It is an analyst-reconstructed model rather than a rule Perkins is documented as teaching, and precise causal percentages remain impossible.
- Provenance and origin — The need emerges from source conflicts: Perkins's broad “we” narration on Compaq and biotechnology, participant accounts assigning Qume and Hybritech elsewhere, and company records preserving founders/scientists/operators. The model was reconstructed from the full note and deal ledger rather than retrofitted to one winner. Attribution controls
- Mechanism and reasoning — Correct role assignment distinguishes repeatable investor skill from founder execution, institutional access, scientific infrastructure, co-investor judgment, vehicle capital, market regime, and luck. Without it, a famous firm's outcome becomes evidence for whichever senior partner tells the story. [researcher inference · strong]
- Historical and vehicle context — Early K&P was a partnership, later KPCB vehicles added Caufield, Byers, Doerr, and others, and syndicated financings shared work across firms. Later portfolio-company and firm counts cannot be back-assigned to a retired co-founder. 1984 JEC hearing, p.166 Vehicle map
- When to use — Use in deal memos, partner reviews, track-record portability, succession, fundraising attribution, board narratives, postmortems, and reference calls. Do not convert a role map into false causal percentages when counterfactual contribution is unknowable.
- How to apply it — Create a dated row for each role and decision; name the person and vehicle; attach a decision-near source; separate original investment from later governance; record disputes; distinguish company outcome from fund cash flow; and downgrade unsupported “we,” “founded,” “led,” or “returned” claims. [researcher inference · strong]
- Examples —
- Hybritech — researcher-applied inference (documented attribution correction): [researcher inference · strong] Byers followed Royston, brought the idea, became founding CEO, and led it; Perkins encouraged, mentored, approved, and joined the board. Byers oral history, PDF pp.27 and 37–39
- Qume — researcher-applied inference (documented attribution correction): [researcher inference · strong] Draper says he brought David Lee's company to K&P; Perkins's pleasure at the outcome is not sourcing credit. Draper oral history, PDF p.15
- Compaq — researcher-applied inference (documented attribution correction): [researcher inference · strong] the firm credits Sevin Rosen with the referral and Doerr with sponsorship; Perkins's supported role is later director/M&A work. Kleiner Perkins, “Compaq”
- Genentech — researcher-applied inference (documented multi-party record): [researcher inference · strong] Swanson and Boyer are founders; Cohen–Boyer science, laboratories, institutions, management, partners, and K&P financing/board work remain distinct. Genentech founders Genentech S-1, business/management sections
- Evidence — Independent sources exceed the default breadth: Byers on Hybritech, Draper on Qume, Kleiner Perkins on Compaq, Genentech's issuer record, and Tandem's decision-near account.
- Limits, failure modes, and exception ledger — A role list does not reveal quality, effort, counterfactual necessity, or economic share. Many checks, votes, board minutes, and allocation records are missing; participants may remember self-servingly; firm pages are promotional; and founders/operators can also understate investor work. Preserve unresolved roles rather than force symmetry.
- Evolution and contradictions — Perkins's later narration sometimes expands “we” to field-building or Compaq; independent records narrow those claims without erasing his genuine Tandem, Genentech, Acuson, and later Compaq governance work. Hybritech and Qume show why attribution became more important as the partnership added people. WBUR, 00:37:22–00:38:53 Compaq control
- Modern VC translation (as of 2026-08-23) — Transfers: assign credit and blame role by role at decision time. Changed assumptions / what fails: firm brand, later board work, and multi-stage platform participation do not prove origination or early selection. Observable signals: named introducer, sponsor, diligence owner, approver, negotiator, board member, recruiter, follow-on decision-maker, vehicle, and narrator. Likely misuse: turning a role ledger into invented causal percentages. Adaptation: preserve dated sources and disclose disputed or unknown roles. Analyst-created hypothetical: a partner joining an AI company after Series C receives governance credit for a sale but not seed sourcing, invention, or the early-fund selection record. [researcher inference · provisional] Historical premise and role conflicts Vehicle control
Evidence gaps and candidate models
The inventory reconciled every distinct rule or philosophy signal in the accessible w-*, t-*, and d-* notes plus writings.md, talks.md, and investments.md. Nine models survived. Closely related signals were merged rather than multiplied: idea first, risk up front, cheap proof, and valuation after evidence feed the decisive-risk and capability-sequencing models; board as product, bad news, long duration, and follow-on signaling feed the governance model; power law, fail cheap, and denominator discipline feed one portfolio model.
| Candidate not promoted to a standalone model | Why it was merged or excluded | Exact evidence still needed |
|---|---|---|
| Individual decision ownership / anti-committee | Wharton records the claim, and HP shows costly personal dissent, but no comparative decision set shows individual ownership outperforming a well-run committee. It remains a governance diagnostic, not a proven standalone rule. Wharton, HP/corporate-VC paragraphs | Dated IC records with named owners, dissent, decisions, and matched outcomes. |
| Follow-on signaling as control | The mechanism is important but comes primarily from Perkins's later statements; it is therefore included as a boundary inside the board model rather than presented as independently validated efficacy. CHM, p.19 | Term sheets, financing alternatives, follow-on votes, founder/minority accounts, and company outcomes. |
| Stay on boards beyond IPO | Perkins supplies a learning/crisis rationale but admits that hundreds of firm companies made indefinite involvement impossible. Selection criteria and comparative outcomes are absent. CHM, pp.19–20 | Board-tenure ledger, exit dates, crisis interventions, independent director/founder assessments, and outcome controls. |
| Incremental product cadence | Acuson supplies one disputed roadmap case and no contemporaneous minutes; it remains a governance example, not a general Perkins rule. CHM, p.21 | Acuson board materials, alternative product plans, customer evidence, and additional documented applications. |
| Founder integrity separate from idea quality | TechCrunch records the distinction, but the corpus has no documented application set connecting character evidence to invest/pass decisions and outcomes. TechCrunch, idea/character discussion | Reference memos, integrity-related passes, board interventions, and counterexamples. |
| Entry-price discipline / excess capital | Perkins forecast excess venture capital and price inflation, but supplied no defined vintage, metric, horizon, or scored deal set. Charlie Rose, 09:52–10:40 WBUR, 00:34:31–00:35:15 | Entry valuations, ownership, dilution, reserves, exits, fund cash flows, predefined forecast series, and scoring date. |
| Small-IPO underpricing | The public record contains a recollected pricing argument around Genentech, not three grounded applications or an issuer-to-fund sale ledger. | Prospectus/pricing memos, allocation and lockup records, selling decisions, and comparable K&P IPO cases. |
| Confidence as an investing edge | Perkins acknowledges that confidence can appear as arrogance; the record has no reliable calibration dataset. The 2014 letter is a warning against globalizing confidence. CHM, p.18 | Decision forecasts with probabilities, errors, updates, dissent, and outcomes across wins, ordinary deals, and losses. |
Reasonable public routes were exhausted for these candidates through the Bancroft and CHM transcripts, processed public audio/video, HP/SEC records, issuer filings, decision-near press, partner/founder oral histories, firm archives, and the deal/denominator searches recorded in sources.md. Missing private partnership, IC, board, allocation, and fund-cash-flow artifacts remain capable of changing the inventory; an honest candidate ledger is preferable to turning slogans or one successful outcome into extra models.
What the evidence supports
The defensible thesis is narrower than the legend. From the early 1970s through the early 1990s, Perkins appears to have been useful in technically uncertain company formation: interrogate the mechanism, name the decisive risk, fund a small proof, then recruit, finance, and govern around the evidence. Tandem and Genentech are the strongest cases; Acuson is a weaker retrospective replication. This supports process, not measured personal alpha. [researcher inference · moderate] thesis labs
HP and University Laboratories plausibly supplied the operator-to-investor bridge, but no contemporary file proves a Doriot or HP lesson caused a deal. The biography labels remembered influence and causal inference correctly. [researcher inference · moderate] Bancroft, printed pp. 5–27 CHM, pp. 5–7
Substantial shared board work is supported at Tandem and Genentech. The package preserves founders, scientists, operators, partners, co-investors, and vehicles instead of turning every firm result into Perkins authorship. Economically, Tandem and Genentech dominate the visible favorable Fund I story, but complete cash flows are absent; excluding both leaves no basis for calling the residual vehicle exceptional. [researcher inference · strong] people map Pareto analysis
What could falsify the thesis
The thesis predicts decision-date records with a named bottleneck, cheap proof, low burn before proof, explicit roles, and identifiable work after proof. Original files or matched ordinary cases without that sequence would weaken it. A complete portfolio showing no advantage after controls for founder, sector, stage, vehicle, ownership, access, and syndicate would overturn its performance implication. [researcher inference · strong] predictions and falsifiers
Apple already shows a screen blocking information collection; Snow-Job shows working product with weak market/shock analysis; the satellite story is unaudited luck; HP shows board confidence is not procedural correctness. [researcher inference · strong] Apple Snow-Job satellite SEC order, paragraphs 4–15
The strongest rival explanation is access, scarce early capital, exceptional founders, partner labor, public science, exit markets, two outliers, and outcome-selected memory. Complete cash flows, pass logs, original IC files, role-stamped board evidence, independent founder/employee testimony, and matched peers would discriminate skill from legend. [researcher inference · strong] skeptical case
Attribution and denominator audit
Accounts conflict between 17 companies and 18 investments; the ledger instead tracks 15 paths—13 investments/leads, one missed intake, and one later-governance control. Dividing 15 by 17 or 18 would mix categories, so coverage is correctly “not computable.” [researcher inference · strong] denominator
Five material corrections survive: Qume came through Sutter Hill/Draper; Hybritech is principally Byers-sponsored; Compaq originated with Sevin Rosen/Doerr while Perkins's role is later governance; Collagen has a Perkins association but unknown terms; Tandem separates founder, technical, diligence, sponsor, syndication, and board roles. [researcher inference · strong] Draper, pp. 15–16 Byers, printed pp. 15–21 Byers, printed p. 28 Compaq Tandem
Reported $8 million committed, about $4 million deployed, later 20%-to-30% carry, and the purported Investor's Bill of Rights are participant accounts, not audited Fund I terms. Genentech's 14.5% pre-offering holding belongs to K&P, not Perkins. No personal IRR, TVPI, DPI, hit rate, loss ratio, outlier percentage, or Apple counterfactual is supportable. [researcher inference · strong] Bancroft, pp. 31–40 Genentech prospectus, pp. 137–138
Transfer limits for modern VCs
What transfers is discipline: define maximum payoff; name decisive uncertainty; buy decision-changing evidence before scale; map capabilities; separate board rights from financing leverage; assign roles; preserve losses, passes, ordinary outcomes, and luck. [researcher inference · moderate] dated translation
Where software iteration is cheap, AI benchmarks are volatile, safety or regulatory risk is intertwined, or a vehicle uses multistage funds, SPVs, global sourcing, and larger evidence trails, literal replay may be inappropriate. Technical proof may not resolve adoption, distribution, rights, manufacturing, regulation, or economics; networks may reproduce access bias; follow-on leverage may become shadow control. These are analyst translations, not Perkins claims about 2026. [researcher inference · provisional] model 2 vehicle fit
Signals are a ranked hypothesis, rival explanation, pre-registered threshold, deadline, next action, source-tagged funnel, role/vehicle ledger, reserve plan, independent escalation, and results without the top outlier. Misuse is milestone theater: testing the easy risk, calling a benchmark a business, treating firm participation as personal selection, or using power law to excuse weak underwriting. [researcher inference · strong] field guide
Unresolved evidence
Highest-value gaps are the partnership agreement; original plan; audited accounts, roster, and distributions; cap tables; investment/pass memos; votes; board minutes; independent founder/employee accounts; raw Something Ventured interview; lawful full memoir; and satellite identity/insurance record. [researcher inference · strong] retrieval ledger archive boundary
Conflicts remain conflicts: 17/18 companies; $100k/$200k/$250k Genentech seed accounts; stage-specific Tandem amounts; changing firm counts; and whether later rules held in ordinary cases. The package supports “shared technical company-builder with a testable staging method,” not “audited top-ranked personal investor.” [researcher inference · strong] source boundary
High-risk central-claim audit trail
“Pass—bounded” means the claim survives only with its stated limit.
| Category | Document/section; source/locator | Result |
|---|---|---|
| Biography and causal chain | _profile.md opening/causal audit; Bancroft, pp. 5–27, CHM, pp. 5–7 | Pass—bounded: chronology holds; causality retrospective/inferred. |
| Edge and falsifiers | _profile.md — “A falsifiable thesis”; investmentphilosophy.md — §§1,16; WBUR, 23:23–31:38, Tandem, pp. 64–65 | Pass—bounded: mechanism, period, predictions, controls and overturning evidence explicit; matched performance untested. |
| Pareto/economics | investments.md first H2/return ledger; Genentech S-1, pp. 137–138, Tandem report | Pass—bounded: top tier unordered; no percentage or realized-return rank. |
| Counts, capital, vehicle | investments.md denominator/vehicle; Bancroft, pp. 31–40 | Pass—bounded: $8m/$4m self-report; 17/18 disputed; 30% not assigned to Fund I. |
| Tandem rationale/terms | investments.md lab 1; BusinessWeek, pp. 64–65 | Pass—bounded: $50k and $1m staged; $500k later syndicate; security/ownership unknown. |
| Genentech rationale/terms | investments.md lab 2; S-1, pp. 107–121, oral, pp. 4–9 | Pass—bounded: contracted proof; seed figures conflict; 14.5% entity-level. |
| Acuson | investments.md lab 3; CHM, pp. 20–21, Siemens filing | Pass—bounded: interrogation/exit hold; early terms retrospective, proceeds unknown. |
| Apple/Snow/satellite | investments.md labs 4–5/postmortems; Apple, p. 61, Snow, 12:35–15:43, satellite, 31:51–33:08 | Pass—bounded: intake miss, incomplete failure, provisional luck; no counterfactual/ranking. |
| Team attribution | profile/investments people maps; Byers, printed pp. 15–21, printed p. 28, Draper, Johnson, Compaq | Pass: Tandem, Qume, Hybritech, Collagen, Compaq roles separated; unknowns stay unknown. |
| HP record | writings 2006; HP 8-K, Item 8.01, SEC, paragraphs 4–15 | Pass: no written reason, private preference, later inquiry, HP respondent remain distinct. |
| 2014 record | writings 2014; New Yorker, “On Saturday”/“Soon after”, TIME | Pass—bounded: authenticated fragments, not full letter; voting seriousness ambiguous. |
| Exact quotations | mental models 2,5; BusinessWeek, p. 64, Red Herring, risk/advantage answers | Pass: “We try to isolate whatever the biggest risk is in a given deal.” and “Once we sign the check, we’re on the team.” have exact locators and are under 25 words. |
| Peer comparisons | profile comparison/philosophy §15; Rock, Johnson, Crisp | Pass: all index-complete; dimension-matched, qualitative, no return rank. |
| Modern premises | philosophy §14; nine model translations; field guide | Pass—bounded: dated 2026-08-23; transfer/failure/signals/misuse/adaptation and cited premises explicit. |
Reproducible ordinary-claim audit
Rule fixed before results: one ordinary claim from each core file's third H2, cycling attribution, source quality, governance, process, base rate, sourcing, case example, and review correction; then fifth-H2 controls in profile, investments, talks, and philosophy. Column 2 fixes the rerunnable target.
| # | Document/claim | Source/locator | Result |
|---|---|---|---|
| 1 | _profile.md H2 3: Tandem roles divided | Byers, pp. 9–11; Johnson, pp. 39–45 | Pass. |
| 2 | sources.md H2 3: Compaq sponsorship elsewhere | KP narrative | Pass—promotional, firm attribution only. |
| 3 | writings.md H2 3: plan targeted four GP-LP conflicts | Bancroft, pp. 34–35,40 | Pass—oral description. |
| 4 | talks.md H2 3: risk must be decomposable | WBUR, 06:58–07:57,23:23–31:38 | Pass—efficacy unmeasured. |
| 5 | investments.md H2 3: 1981 source said roughly half written off | TIME, Aug. 10, 1981 | Pass—era rate, not Fund I. |
| 6 | investmentphilosophy.md H2 3: Qume came via Draper/Sutter Hill | Draper, pp. 15–16 | Pass—economics unknown. |
| 7 | mental-models.md H2 3: Hybritech roles center Byers/Royston | Byers, printed pp. 15–21; printed p. 28 | Pass—retrospective. |
| 8 | review.md H2 3: Compaq excluded from Perkins selections | KP narrative | Pass—later governance retained. |
| 9 | _profile.md H2 5: K&P/KPCB I–IV distinct; three partnerships reported 86 companies | Tandem proxy; JEC, p. 166 | Pass—no personal attribution. |
| 10 | investments.md H2 5: Hybritech principally Byers's deal | Byers, printed pp. 15–21; printed p. 28 | Pass—Perkins roles retained. |
| 11 | talks.md H2 5: Doriot/HP sequence is self-described | Bancroft, pp. 5–25 | Pass—investor-stated, not causal proof. |
| 12 | investmentphilosophy.md H2 5: initial capital buys the next discriminating proof | Genentech oral, pp. 4–7 | Pass—case-grounded, not portfolio-proven. |
Twenty-point acceptance evidence
| # | Pass/fail | Exact document/section inspected | Evidence/correction or continuation |
|---|---|---|---|
| 1 | PASS | _profile.md chronology/“Sourced chronology” | Formation, process, action, outcome, causal labels substantive. |
| 2 | PASS | _profile.md “A falsifiable thesis”; investmentphilosophy.md §§1,16 | Edge, period, predictions, controls, alternatives, overturning evidence explicit. |
| 3 | PASS | investments.md “No-hindsight decision lab” 1–5 | Known/believed/unknown precede outcome; unavailable fields stay unknown. |
| 4 | PASS | investments.md first H2 | Three ranks separate; outlier test; no unsupported percentage. |
| 5 | PASS | _profile.md; sources.md; writings.md; talks.md; investments.md; investmentphilosophy.md; mental-models.md; review.md — H1→H2/bodies | All substantive; index stayed active during review, so no premature claim here. |
| 6 | PASS | investments.md denominator/ledger | Known through excluded reconciled; “not computable” justified. |
| 7 | PASS | writings.md coverage/four families | Seven routes collapse transparently; access boundaries explicit. |
| 8 | PASS | mental-models.md nine models | Each has 12 bullets, cases, counterexample, citations, exception ledger where needed. |
| 9 | PASS | writings.md translation; investmentphilosophy.md §14; mental-models.md modern bullets | Dated; transfer, failure, signals, misuse, adaptation, labeled hypothetical covered. |
| 10 | PASS | investments.md labs 4–5/postmortems; d-apple-pass.md; d-snow-job.md | Miss/failure fields covered to evidence boundary; missing process change disclosed. |
| 11 | PASS | _profile.md/investments.md people maps; d-attribution-ordinary-luck.md matrix | Sourcing through narration named or unknown. |
| 12 | PASS | investments.md vehicle section; investmentphilosophy.md §6 | Capital, duration, fundraising, carry, governance, geography, liquidity, regulation, reserves, succession covered. |
| 13 | PASS | investments.md historical context; _profile.md context; investmentphilosophy.md §13 | Decision-near market, macro, regulation, science/base rates separated from hindsight. |
| 14 | PASS | sources.md coverage; writings.md/talks.md retrieval limits | Direct voice, filings, archives, catalogs processed or barrier logged; checkboxes untouched here. |
| 15 | PASS | review.md central/ordinary audit tables | Adjacent links and available locators; absent schemes disclosed. |
| 16 | PASS | _profile.md; sources.md; writings.md; talks.md; investments.md; investmentphilosophy.md; mental-models.md; review.md — material claims | Canonical attribution/support labels distinguish statement, record, behavior, witness, inference. |
| 17 | PASS | _profile.md skeptical case; investmentphilosophy.md §16; review.md falsification | Luck, access, team/vehicle, denominator, survivorship, economics, era tested. |
| 18 | PASS | _profile.md comparison; investmentphilosophy.md §15; _index.md 006/008/010 | Three peers complete, dimension-matched, underlying evidence linked. |
| 19 | PASS | investments.md labs 1–5; _profile.md/investmentphilosophy.md paths | Four questions precede outcomes; Apple/Snow include miss/failure. |
| 20 | PASS | _profile.md learning path; investmentphilosophy.md §17; maps/checklists | Chronology, reading order, links, glossary, checklists, Markdown maps work. |
Content-level acceptance is PASS on all 20 gates, with evidence limits preserved. This report makes no queue-bookkeeping, index, themes, remote, merge, or publication claim/change.
Writings and direct written record
- ★★★ | Valley Boy: The Education of Tom Perkins | https://books.google.com/books/about/Valley_Boy.html?id=AwkVP65ZTiIC | 2007 | lawful limited preview plus publisher record; reliability B for the participant memoir, access partial | Core authored retrospective; visible pages and independently reproduced passages only, never the inaccessible body.
- ★★ | Valley Boy publisher record | https://www.penguinrandomhouse.com/books/301214/valley-boy-by-tom-perkins/9781101216989/ | 2007 | publisher metadata; reliability A for edition/B promotional description | Edition and scope control; no substitute for text.
- ★★ | Valley Boy Hachette audiobook record | https://www.hachetteaudio.com/titles/tom-perkins/valley-boy/9781600240614/ | 2007 | publisher metadata; reliability A for edition and narrator | Routes a lawful audio edition; no transcript exposed.
- ★★★ | HP resignation and disclosure record | https://www.sec.gov/Archives/edgar/data/47217/000110465906059471/a06-19246_18k.htm | 2006-09-06 filing; resignation 2006-05-18 | full issuer-authored corrective 8-K; reliability A; Item 8.01 | Decision-near record that explicitly says Perkins supplied no written reason at resignation; control with the SEC order.
- ★★ | Sex and the Single Zillionaire | https://books.google.com/books/about/Sex_and_the_Single_Zillionaire.html?id=UQdSPgAACAAJ | 2006 | bibliographic metadata only; reliability A for identity, access unavailable | Authored novel, low relevance to investment method; body excluded.
- ★★ | Wall Street Journal letter on “progressive Kristallnacht” | https://www.newyorker.com/business/currency/tom-perkins-and-schadenfreude-in-silicon-valley | 2014-01-24 | full original inaccessible; exact passage reproduced in contemporaneous coverage; reliability B | Public-judgment evidence; distinguish authenticated fragments, reported retraction, and later voting comments.
- ★★ | Perkins oral description of a purported “Investor’s Bill of Rights” | https://digicoll.lib.berkeley.edu/record/218580/files/perkins_tom.pdf | stated origin circa 1972; described in 2009 oral history | full participant transcript, missing original plan; reliability A/B; printed pp. 32–35 and 40 | GP-LP allocation, audit and co-investment conflict rules; not generic board rights and not an inspected artifact.
Oral histories, talks, video, and interviews
- ★★★ | Tom Perkins — Bancroft/NVCA oral history | https://digicoll.lib.berkeley.edu/record/218580/files/perkins_tom.pdf | interviews 2009; published 2010 | full participant transcript; reliability A for custody/B for memory; printed-page locators | Main life, Fund I, Tandem, Genentech, governance, failures, and philosophy spine; quotation restrictions apply.
- ★★★ | Kleiner & Perkins oral history interview on Genentech | https://digicoll.lib.berkeley.edu/record/54368/files/klienerperkins00perkrich.pdf | interviews 2001; published 2002 | full participant transcript; reliability A/B | Deep Genentech financing, board, partnership, licensing, and attribution record.
- ★★★ | Oral History of Tom Perkins — Computer History Museum | https://archive.computerhistory.org/resources/access/text/2012/03/102714336-05-01-acc.pdf | recorded 2011-07-22 | full official transcript; reliability A/B | University Labs, decision method, staging, Acuson, investment count, and later reflection.
- ★★★ | “Venture Pioneer Tom Perkins Unveils Some Secrets” | https://cryptoniteventures.substack.com/p/venture-pioneer-tom-perkins-rip-unveils | interview originally published 1993 | full interview mirror by original interviewer; reliability B | Earlier direct record for Fund I, boards, Snow-Job, sourcing, and 180-deal denominator.
- ★★★ | “Pioneers of Venture Capital” | https://computerhistory.org/events/pioneers-venture-capital/ | event 2002-09-30; official video uploaded 2017-11-21 | official CHM event/video record; reliability A for custody/date and B for speaker-bounded machine-caption paraphrase | Perkins with William Draper III, Pitch Johnson, Burton McMurtry, Arthur Rock, and Donald Valentine; full caption route searched, with three moderator-bounded Perkins turns processed.
- ★★★ | Silicon Valley Deconstructed with Tom Perkins and Don Valentine | https://techcrunch.com/video/silicon-valley-deconstructed-with-tom-perkins-and-don-valentine/ | 2013-09-11 appearance | mapped hosted-video route did not expose a playable body, captions, or transcript in this run | Processed to the public access boundary; no timestamped speech inferred, and the associated edited report is the content record.
- ★★ | TechCrunch report on Perkins–Valentine panel | https://techcrunch.com/2013/09/11/vc-titans-tom-perkins-and-don-valentine-articulate-what-makes-a-good-vc/ | 2013-09-11 | contemporaneous edited event report with attributed fragments; reliability B | Full report processed; source is a partial selection, not a replacement for inaccessible full video context.
- ★★★ | “Tom Perkins: The Captain of Capitalism” | https://www.cbsnews.com/news/tom-perkins-the-captain-of-capitalism/ | page 2007-11-01; broadcast 2007-11-04; updated 2008-05-30 | 13:26 hosted video plus CBS-edited narrative transcript/report; reliability B | Full HTML body processed; memoir-era claims, Genentech seed-number conflict, governance, and personality, with no caption or shot-level transcript.
- ★★★ | Thomas Perkins — Charlie Rose | https://charlierose.com/videos/11352 | 2007-12-05 | full official speaker-labeled, timestamped transcript plus hosted player; reliability A/B | Capital-supply forecast, green/biotech/China predictions, partner attribution, referral network, missing-plan ethics claim and HP private-resolution chronology.
- ★★★ | “If There Is No Risk, You Have Already Missed the Boat” | https://knowledge.wharton.upenn.edu/article/venture-capitalist-tom-perkins-if-there-is-no-risk-you-have-already-missed-the-boat/ | published 2012-12-05; exact recent-visit date unreported | institutional edited talk report with direct answers; reliability B | Risk selection, board work, ethics, and sector views; no recording or transcript surfaced.
- ★★ | Creative Capital interview, “Valley Boy,” part 1 | https://creativecapital.wordpress.com/2008/01/17/valley-boy-part-1-of-an-interview-with-vc-pioneer-thomas-j-perkins/ | 2008-01-17; interview conducted 2007-11 | full edited Q&A; reliability B for attributed speech | Fail-cheap and power-law claims, bubbles, entry price/liquidity, MySpace board advice, firm succession, and green/KPCB XIII/China forecasts; promised part 2 not located.
- ★★ | Venture Voice interview | https://venturevoice.com/tom-perkins-kleiner-perkins-caufield-byers | 2006 route | current slug resolves to the generic Venture Voice page; historical guest roster authenticates Perkins but the episode body/audio was not recoverable | Processed to the public access boundary; no speech or quotation inferred.
- ★★ | WBUR “On Point” — Silicon Valley’s Tom Perkins | https://www.wbur.org/onpoint/2007/11/28/silicon-valleys-tom-perkins | 2007-11-28 | official 45:59 MP3; reliability B for interview speech | Full official audio transcribed with OpenAI
whisper-1; timestamps preserved and machine wording paraphrased. - ★★ | Commonwealth Club appearance | https://digitalcollections.hoover.org/objects/4419/thomas-j-perkins-speech | Hoover catalog date 2007-11-14; ABC7 says recorded 2007-10-28 | official 59:27 catalog metadata; reliability A for custody/date fields, body unavailable | Conflicting dates and access boundary preserved; no speech inferred.
- ★★ | Risk & Reward interview | https://www.riskandreward.org/episode/tom-perkins | date unavailable | canonical route returned a 114-byte empty body in this run; reliability C | Processed to the public access boundary; identity/date/media remain unauthenticated.
- ★★ | Something Ventured Tom Perkins interview | https://americanhistory.si.edu/collections/archival-item/sova-nmah-ac-1467-ref12 | recorded in person 2008-10-08 | Smithsonian item metadata; reliability A for custody/session identity, raw interview not public | Exact archival item processed; edited film does not substitute for raw speech.
Firm, vehicle, company, deal, and team records
- ★★★ | Tandem annual reports and 1991 proxy compilation | https://archive.computerhistory.org/resources/access/text/2023/07/102659458-05-01-acc.pdf | company records 1970s–1990s | contemporaneous issuer material; reliability A | Director chronology, partnership-vehicle dates, governance, and founder boundary.
- ★★★ | Genentech 1980 prospectus in Fred Middleton oral-history appendix | https://digicoll.lib.berkeley.edu/record/54361/files/genentechoralhi00middrich.pdf | 1980 prospectus; later archival compilation | contemporaneous offering record; reliability A | Entity ownership and capitalization control; no automatic personal-Perkins economics.
- ★★★ | Genentech founders page | https://www.gene.com/about-us/leadership/our-founders | current official history | issuer retrospective; reliability B/promotional | Protects Swanson/Boyer founder credit and scientific origins.
- ★★★ | Kleiner Perkins history — Genentech | https://www.kleinerperkins.com/perspectives/genentech/ | current firm retrospective | reliability B/promotional | Firm-side narrative; requires participant and contemporaneous controls.
- ★★★ | Kleiner Perkins history — Compaq | https://www.kleinerperkins.com/perspectives/compaq/ | current firm retrospective | reliability B/promotional | Attributes original sponsorship to John Doerr/Sevin Rosen; confines Perkins to later governance/M&A.
- ★★★ | Compaq 2001 proxy | https://www.sec.gov/Archives/edgar/data/714154/000089155401501345/d25040_def14a.htm | 2001 | SEC-filed issuer record; reliability A | Board chronology and holdings; does not establish original sourcing.
- ★★★ | HP corrective Form 8-K | https://www.sec.gov/Archives/edgar/data/47217/000110465906059471/a06-19246_18k.htm | 2006-09-06 | SEC-filed issuer record; reliability A | Board-leak chronology, resignation-letter disclosure, and governance dispute.
- ★★★ | SEC Order, Hewlett-Packard Company | https://www.sec.gov/files/litigation/admin/2007/34-55801.pdf | 2007-05-23 | official administrative order; reliability A | Establishes disclosure findings against HP, not personal misconduct by Perkins.
- ★★★ | House hearing on HP pretexting | https://www.congress.gov/109/chrg/CHRG-109hhrg31472/CHRG-109hhrg31472.pdf | 2006-09-28 | official congressional hearing; reliability A | Sworn/public testimony and documentary chronology; separate allegation from adjudication.
- ★★★ | Brook Byers oral history | https://digicoll.lib.berkeley.edu/record/218191/files/byers_brook.pdf | interviews 2002–2005 | full participant transcript; reliability A/B | Tandem diligence, Hybritech lead credit, Collagen, and Perkins mentorship.
- ★★★ | Pitch Johnson oral history | https://digicoll.lib.berkeley.edu/record/103604/files/johnson_franklin.pdf | interviews 2008 | full participant transcript; reliability A/B | Tandem syndication, board attribution, and independent peer cross-check.
- ★★★ | William Draper oral history | https://archive.computerhistory.org/resources/access/text/2013/04/102743096-05-01-acc.pdf | participant interview | full CHM transcript; reliability A/B | Qume sourcing through Draper/Sutter Hill and early-network context.
- ★★ | Fred Coury oral history | https://archive.computerhistory.org/resources/access/text/2019/08/102738262-05-01-acc.pdf | CHM participant interview | full transcript; reliability A/B | Semiconductor/company witness route; role and identity must be bounded.
- ★★★ | Siemens acquisition of Acuson | https://www.sec.gov/Archives/edgar/data/94063/000094787100000492/0000947871-00-000492-0001.htm | 2000 | SEC-filed transaction record; reliability A | Exit-value control; no Perkins check, ownership, dilution, carry, or realized proceeds.
- ★★ | LSI Logic oral-history panel | https://archive.computerhistory.org/resources/access/text/2013/05/102658037-05-01-acc.pdf | participant panel | full CHM transcript; reliability A/B | Shared company-building and board attribution; no person-level cash-flow chain.
- ★★ | VLSI Technology oral-history panel | https://archive.computerhistory.org/resources/access/text/2013/05/102658035-05-01-acc.pdf | participant panel | full CHM transcript; reliability A/B | Semiconductor financing context and team controls.
- ★★★ | TIME, “Boom Time in Venture Capital” | https://content.time.com/time/subscriber/article/0,33009,954903-1,00.html | 1981-08-10 | contemporaneous press; reliability B | Near-decision-date venture-market and Genentech/Tandem valuation context.
- ★★★ | Washington Post, “Designer Genes Are Snapped Up” | https://www.washingtonpost.com/archive/business/1980/10/15/designer-genes-are-snapped-up/a4253927-0edf-4b4b-bb13-1a061e3371e1/ | 1980-10-15 | contemporaneous IPO report; reliability B | Genentech offering demand and value; not realized LP returns.
- ★★ | News Corp 2013 Form 10-Q | https://www.sec.gov/Archives/edgar/data/1564708/000119312513432748/d608992d10q.htm | 2013 | SEC-filed issuer record; reliability A | Later public-board role and governance-control route.
- ★★ | News Corp poison-pill litigation opinion | https://law.justia.com/cases/delaware/court-of-chancery/2014/ca-6285-vcp.html | 2014 | Delaware Chancery opinion; reliability A | Adjudicated governance context; attribute findings to the court, not press gloss.
Books, scholarship, press, and adverse controls
- ★★★ | Tom Nicholas, VC: An American History — first-fund chapter | https://www.perlego.com/book/1148153/vc-an-american-history-pdf | 2019 | scholarly history; reliability B; licensed access | Key 17-company, loss-rate, gross-return, and outlier claims; distinguish author calculations from audited fund returns.
- ★★ | Sally Smith Hughes, review of Nicholas | https://thebiotechconsultant.com/wp-content/uploads/2020/05/Nicholas-Review-Hughes.pdf | 2020 | scholarly review; reliability B | Challenges historical framing and biotechnology detail; use as critique, not substitute evidence.
- ★★ | Sebastian Mallaby, The Power Law | https://www.penguinrandomhouse.com/books/580010/the-power-law-by-sebastian-mallaby/ | 2022 | secondary history; reliability B; publisher metadata/legitimate edition | Broad firm and era synthesis; exact claims require accessible page locators.
- ★★★ | Bob Phelps, first-fund reconstruction | https://www.ianhathaway.org/blog/2019/7/31/vc-an-american-history | 2019-07-31 | analytical review of Nicholas; reliability C/B | Routes conflicting 17-company and gross-return figures; never present as audited LP return.
- ★★ | FundingUniverse — Kleiner Perkins history | https://www.fundinguniverse.com/company-histories/kleiner-perkins-caufield-byers-history/ | undated company-history compilation | tertiary; reliability C | Conflicting 18-investment/41.5% account and portfolio-name leads only.
- ★★ | FundingUniverse — Tandem history | https://www.fundinguniverse.com/company-histories/tandem-computers-inc-history/ | undated | tertiary; reliability C | Company chronology lead to resolve against issuer and participant records.
- ★★ | New Yorker, “The Kona Files” | https://www.newyorker.com/magazine/2006/10/16/the-kona-files | 2006-10-09 | reported reconstruction; reliability B | HP board-conflict chronology and narrative tension.
- ★★ | Fortune on SEC action against HP | https://fortune.com/2007/05/23/sec-hp-pretexting/ | 2007-05-23 | contemporaneous press; reliability B | Cross-checks the SEC order; order remains controlling.
- ★★ | Washington Post on Fiorina revision | https://www.washingtonpost.com/news/wonk/wp/2015/10/02/why-this-silicon-valley-bigshot-voted-to-fire-carly-fiorina-but-wants-her-to-be-president/ | 2015-10-02 | reported interview/reconstruction; reliability B | Tests retrospective revision about firing Fiorina.
- ★★ | Independent report on 1995 yacht collision | https://www.independent.co.uk/news/world/yachtsman-gets-suspended-sentence-for-collision-1578766.html | 1995-10-23 | contemporaneous court report; reliability B | Personal-judgment record; no investment inference.
- ★★ | New Yorker on wealth letter and reported retraction | https://www.newyorker.com/business/currency/tom-perkins-and-schadenfreude-in-silicon-valley | 2014-01-27 | contemporaneous commentary with authenticated fragments; reliability B | Exact rhetoric, defense, firm response and public reaction; TIME controls the reported retraction.
- ★★ | TIME on tax-weighted voting comments | https://time.com/8466/tom-perkins-taxes/ | 2014-02-14 | contemporaneous report/video summary; reliability B | Separate later voting claim from original letter.
- ★★ | Wired review of Valley Boy | https://www.wired.com/2007/07/tom-perkins-valley-boy/ | 2007 | contemporaneous book review; reliability C/B | Memoir content lead and reputation context; no substitute for book body.
- ★★ | Publishers Weekly review of Valley Boy | https://www.publishersweekly.com/9781592403134 | 2007 | trade review; reliability C/B | Scope and self-presentation control; not investment evidence.
- ★★ | Fortune obituary | https://fortune.com/2016/06/09/tom-perkins-death/ | 2016-06-09 | obituary; reliability B | Death/date and reputation; firm-logo portfolio lists cannot prove person-level credit.
- ★★ | Silicon Valley “Something Ventured” retrospective | https://www.sfgate.com/business/article/Something-Ventured-looks-at-Silicon-Valley-2309156.php | 2011-11-04 | film/profile report; reliability B | Edited-history context and participant identification.
- ★★ | Don Valentine oral history | https://archive.computerhistory.org/resources/access/text/2014/11/102658321-05-01-acc.pdf | CHM participant interview | full transcript; reliability A/B | Rival early-firm model and Perkins-panel cross-check; Valentine profile itself is unfinished.
Archives, policy, and unresolved-record routes
- ★★ | Stanford collection — Something Ventured | https://purl.stanford.edu/mb678nw9491 | interview circa 2009 | official archive metadata; reliability A for custody; access restricted | Raw interview/transcript route processed to public boundary.
- ★★ | Smithsonian Something Ventured collection | https://sova.si.edu/record/NMAH.AC.1467 | 2008–2010 | official finding aid; reliability A | Custody and access control; duplicate manifestations collapse into one interview family.
- ★★ | CHM collections search — Tom Perkins | https://www.computerhistory.org/collections/search/?q=%22Tom%20Perkins%22 | live catalog | official discovery route; reliability A for metadata | Enumerates transcript, event, and company records; deduplicate manifestations.
- ★★ | Berkeley Bancroft oral-history catalog | https://digicoll.lib.berkeley.edu/search?ln=en&p=Tom+Perkins | live catalog | official discovery route; reliability A | Rights, custody, and related-interview control.
- ★★ | 1978 Congressional Record — capital-gains debate | https://www.congress.gov/95/crecb/1978/05/26/GPO-CRECB-1978-pt12-3-1.pdf | 1978-05-26 | official contemporaneous record; reliability A | Policy/era context; does not establish Perkins's individual causation.
- ★★ | Federal Reserve, economics of private equity | https://www.federalreserve.gov/pubs/staffstudies/168/ss168.pdf | 1995 | official staff study; reliability A/B | Industry institutions, information, and return-comparison context; not a KPCB audit.
- ★★ | Martin Kenney, venture capital and the US innovation system | https://kenney.faculty.ucdavis.edu/wp-content/uploads/sites/332/2018/03/how-venture-capital-became-a-component-of.pdf | scholarly history | reliability B | Public research, policy, network, and era controls against lone-pioneer causation.
- ★ | CourtListener exact-name litigation sweep | https://www.courtlistener.com/?q=%22Tom%20Perkins%22&type=r | searched 2026-08-23 | discovery route; reliability C for negative-search boundary | No verified person-level investment sanction surfaced; bounded absence is not proof.
Coverage assessment
Mechanical coverage
The map contains 68 retrieval routes: 28 ★★★, 39 ★★, and 1 ★ across 7 writings/direct texts, 16 oral-history/talk/interview routes, 20 firm/vehicle/deal/team records, 17 books/press/adverse controls, and 8 archival/policy routes. Catalog/transcript, event/video, and Something Ventured manifestations must be collapsed when counting underlying works. Checkboxes remain open at SCOUT; each later phase closes a route only after full processing or an explicit access-boundary note.
Identity, chronology, and direct-voice boundary
The target is Thomas James “Tom” Perkins (1932–2016), the MIT/HBS-trained HP operator, University Laboratories founder, Kleiner & Perkins co-founder, Tandem/Genentech chair, and later HP/Compaq/News Corp director. Searches use Tom Perkins, Thomas J. Perkins, Thomas James Perkins, Kleiner & Perkins/K&P/KPCB, University Laboratories, and named companies/partners. They exclude unrelated Tom/Thomas Perkins identities and distinguish Perkins's personal company, finite partnership vehicles, successor KPCB funds, and public-company directorships. The direct spine is unusually rich but overwhelmingly retrospective; the 1993 interview, 2006 HP issuer/regulatory filings, 1980 prospectus/press, and issuer proxies provide decision-nearer controls.
Writings and access boundary
Seven priority written-record routes surfaced: Valley Boy and two edition manifestations; HP’s issuer-authored corrective 8-K; Sex and the Single Zillionaire metadata; the 2014 wealth letter; and Perkins’s oral description of a purported “Investor’s Bill of Rights” in K&P’s missing original plan. They collapse to three Perkins-authored or Perkins-described writing families plus one decision-near written governance control. The memoir and 2014 original are only partially accessible; the novel and edition routes are metadata-only; the missing plan is not treated as inspected text.
Vehicle, investment, and attribution boundary
Public sources conflict on whether 1972 Fund I contained 17 or 18 companies. Perkins reports $8 million committed, about $4 million invested, and the rest returned; no public partnership agreement, audited schedule, capital account, dilution history, carried-interest allocation, or net-LP cash flow was found. Candidate paths include University Laboratories, Qume, Tandem, Genentech, Snow-Job/Advanced Recreation Equipment, American Athletic Shoe/Tread-Two, Collagen, Acuson, LSI Logic, VLSI Technology, Cetus, Andros, Hybritech, and later governance at Compaq. This is a discovery roster, not a complete portfolio. Genentech seed accounts of $100,000, $200,000, and $250,000 are left unreconciled by date/tranche/entity; Tandem amounts also vary by financing stage. Swanson/Boyer, Treybig/Loustaunou, Kleiner, Byers, Johnson, scientists, operators, co-investors, LPs, and later KPCB partners retain their roles.
Adverse and comparison coverage
The adverse record includes Snow-Job and ordinary/failure candidates, Apple and other passes, two-outlier concentration, changing 180/450/525/500-plus/1,000-plus firm-level denominators, HP pretexting/disclosure, Fiorina narrative revision, News Corp governance, the 1995 yacht collision, and 2014 wealth/voting rhetoric. SEC findings against HP are not personal Perkins investment sanctions. Completed comparison peers are Arthur Rock, Pitch Johnson, and Peter Crisp; unfinished Eugene Kleiner and Don Valentine packages are evidence sources, not final comparison profiles.
Unresolved after public-route exhaustion
- Fund I's governing documents, complete 17/18-company roster, cash flows, carried interest, and net LP returns remain private or unfound.
- Public evidence cannot reconcile each Genentech/Tandem funding amount into a dated security and entity chain.
- No complete Perkins-attributable portfolio denominator or person-level hit rate exists; later firm totals change definitions and include other partners.
- Valley Boy, the novel, raw Something Ventured interview, and several audio appearances lack a lawful public full-text body.
- Ordinary outcomes—Andros, American Athletic Shoe, Collagen, Acuson, LSI/VLSI, and early failures—remain less documented than Tandem and Genentech.
- No archive, company, firm, or person was contacted; missing private records are disclosed rather than guessed.
Source notes
[d] Acuson — a differentiated device in an apparently saturated market
- Decision dates: 1979–81 founder-funded development; January 1982 institutional financing; October 1982 prototype bridge; 1983 product launch; 1986 IPO; undated pre-1995 roadmap dispute; 2000 Siemens acquisition.
- Core sources: Smithsonian Institution Archives, Acuson Ultrasound Machines Videohistory Interviews, 1997, collection summary/finding aid (institutional custody; 12-hour/203-page multi-employee oral-history collection, but transcript is not publicly downloadable); Smithsonian ACUSON 128 object record (institutional artifact metadata); HBS working paper, “Ultrasound Imaging—Cheap, Versatile, and Safe,” pp. 8–11, especially “Acuson's Path to Product Launch” (independent secondary synthesis citing the Smithsonian transcript); Tom Perkins CHM oral history, pp. 20–21 (retrospective investor account); Tandem 1991 proxy, Perkins biography (issuer record of Perkins's Acuson chair role); SFGate, June 19, 1995 (contemporaneous press on chair succession); SEC v. Dugan complaint, ¶¶1–4 and Siemens annual report filed December 6, 2002, acquisitions section (transaction controls).
- Access: public HTML/PDF routes inspected 2026-08-23. The Smithsonian transcript itself is repository-access only; no speech is inferred from the finding aid. The business plan, 1982 term sheet, cap table, loan agreement, board minutes, product-roadmap papers, and fund ledger were not found publicly.
- Reliability: A for Smithsonian custody/artifact metadata, formal board-role records, and SEC/Siemens transaction facts; B for the HBS synthesis and contemporaneous press; B/C for Perkins's retrospective decision and causation claims.
Case boundary and why confidence is lower
Samuel Maslak, Robert Younge, and Amin Hanafy supplied the technical formation and company building. Perkins's supported role is investor, technical questioner, recruiter, and chair—not inventor or operator. The Smithsonian's institutional history identifies the founder-engineers and computer-controlled ultrasound advance; Perkins's own account likewise centers Maslak's plan and technology. [researcher inference · strong] Smithsonian object record CHM, pp. 20–21
“Founded” dates differ by convention: the HBS/Smithsonian-derived chronology begins the effort in 1979, while the Smithsonian collection summary describes Acuson Corporation as founded in 1982. The safe chronology is founder-funded work from 1979, institutional financing in January 1982, and product launch in 1983. [researcher inference · strong] HBS, p. 10 Smithsonian collection summary
This case has a credible independent chronology but not a public decision-date term sheet or full Smithsonian transcript. Initial pricing, ownership, and formal investor protections therefore remain unresolved.
Decision 1 — January 1982: finance the full prototype after founder-funded work
Frozen information set
| State at the decision date | Reconstruction |
|---|---|
| Known | Maslak and Younge had worked at HP; Hanafy added ultrasound-transducer specialization. The founders had spent roughly two years using savings, second mortgages, consulting income, and a reported $100,000 angel investment before the institutional round. [retrospective witness · moderate] HBS, “Acuson's Path to Product Launch,” p. 10 |
| Known | Ultrasound was already an established diagnostic category. By the early 1980s, many US hospitals owned multiple systems and an industry forecaster expected revenue declines; the market looked saturated, while numerous prior entrants had failed to win meaningful share. [researcher inference · strong] HBS, pp. 7–9 |
| Believed | Computer-controlled image formation with 128 channels could produce materially sharper images than incumbent 64-channel systems and command adoption despite market saturation. Perkins says his diligence focused on the physical mechanism—acoustical diffraction—rather than the plan's length or forecast. [retrospective witness · moderate] CHM, pp. 20–21 Smithsonian object record |
| Known | Maslak had no prior management record of running an organization. Perkins says this was explicit at investment and that leadership replacement remained possible. [investor-stated · strong] CHM, pp. 20–21 |
| Unknown | Whether the architecture would work as an integrated prototype; image quality in clinical hands; manufacturing yield/reliability; regulatory treatment; sales-force cost; willingness of radiologists to switch; incumbent response; service burden; and whether Maslak could recruit/manage a scaled team. [researcher inference · strong] HBS, pp. 8–11 |
| Disputed | Perkins presents the investment as driven by his mechanism-level technical understanding; the HBS/Smithsonian chronology says the first four VCs approached offered more than requested. Scarcity of capital therefore was not the documented edge, and investor selection may have reflected fit rather than rescue. [researcher inference · moderate] CHM, pp. 20–21 HBS, p. 10 |
Alternatives and market/financing context
- Continue founder/angel funding; license the architecture to an incumbent; choose another VC; raise the requested amount from a syndicate; or take KPCB capital and board help. The HBS reconstruction reports that Acuson chose KPCB after four firms expressed immediate interest. [retrospective witness · moderate] HBS, p. 10
- The reported January 1982 financing was $2.5 million, half from the existing angel and half from KPCB. The source does not disclose security, valuation, ownership, liquidation preference, option pool, board composition, reserves, or founder vesting. [retrospective witness · moderate] HBS, p. 10
- The market case was non-consensus in a precise way: the category was real but apparently saturated, and the product had to create replacement demand through image quality rather than depend on greenfield adoption. [researcher inference · strong] HBS, pp. 8–10
Risks, kill criteria, terms, and milestones
- Technical kill: inability to turn the 128-channel/computer-controlled thesis into a stable integrated prototype with clinically superior images. [researcher inference · strong] Smithsonian object record HBS, pp. 9–10
- Clinical/market kill: expert radiologists do not judge the image improvement meaningful enough to switch or pay; sales/service economics cannot support a direct organization; incumbents match the advance before launch. [researcher inference · strong] HBS, pp. 9–10
- People kill: Maslak cannot recruit, delegate, or accept missing operating functions. Perkins says the possibility of replacing him was discussed, but no founder-side or board-record confirmation was found. [investor-stated · moderate] CHM, pp. 20–21
- Milestones: integrated prototype → evaluations by leading radiologists → manufacturing preparation → first shipments → repeatable sales/direct support. This chain is independently reconstructed, though not preserved as a signed financing schedule. [researcher inference · strong] HBS, p. 10
Decision 2 — October 1982: bridge an over-budget prototype to clinical and manufacturing proof
Frozen information set
| State at the decision date | Reconstruction |
|---|---|
| Known | Ten months after the institutional round, Acuson had a working prototype but had spent more than expected. [researcher inference · strong] HBS, p. 10 |
| Believed | A bounded bridge could finance expert clinical tests and manufacturing preparation, the next two observable uncertainties. [researcher inference · strong] HBS, p. 10 |
| Unknown | Whether tests would be favorable; whether the prototype could become a producible/serviceable system; cash required to launch; and pricing/volume after launch. [researcher inference · strong] HBS, pp. 9–10 |
| Disputed/absent | HBS describes KPCB's bridge as a loan. Principal, maturity, interest, security, covenants, conversion rights, and whether the angel participated are not public. [researcher inference · insufficient public record] HBS, p. 10 |
The disciplined choice would be to finance only through independent clinical evaluation and manufacturing readiness, not to treat sunk cost or a working engineering prototype as product-market proof. The historical loan appears consistent with that logic, but no contemporaneous credit memo states the condition. [researcher inference · moderate] HBS, p. 10
Board interventions, follow-ons, and the roadmap dispute
- Perkins says he recruited the key management team and that Maslak developed into an exceptional manager. Without names, hiring dates, or other witness accounts in the public record, preserve this as investor attribution rather than documented sole causation. [investor-stated · moderate] CHM, pp. 20–21
- The Tandem 1991 proxy identifies Perkins as Acuson's chairman, providing a primary role control. It does not show his ownership, committee authority, attendance, interventions, or KPCB's fund assignment. [contemporaneous record · strong] Tandem 1991 proxy, Perkins biography
- Perkins later says he opposed committing the company to a five-to-seven-year architectural leap without nearer releases, proposed incremental integration, resigned when Maslak refused, and believed the resulting product gap forced a sale. The public transcript supplies no date, board vote, roadmap, competing forecast, or stop criteria. [investor-stated · moderate] CHM, p. 21
- Contemporary press records that Maslak replaced Perkins as chair in 1995; Siemens acquired Acuson in November 2000. That chronology does not prove the alleged five-to-seven-year gap caused the sale. Market conditions, product performance, and other strategic alternatives remain confounders. [researcher inference · strong] SFGate, June 19, 1995 Siemens annual report, acquisitions section
- Governance asymmetry matters: product cadence may protect the installed franchise, while board pressure for incremental releases can also underfund a true architectural leap. With no board records or founder response, the lesson is an unresolved tradeoff, not “Perkins was right.” [researcher inference · strong] CHM, p. 21
What the case can and cannot teach
- The strongest decision mechanism is an evidence ladder: founder-funded exploration → institutional prototype capital → bridge for independent clinical/manufacturing proof → launch. [researcher inference · strong] HBS, p. 10
- A mature or apparently saturated market can still support replacement demand if a product creates a clinically visible step-change, but Acuson's exceptional outcome is not a base rate for the many ultrasound entrants that failed. [researcher inference · strong] HBS, pp. 8–11
- Backing an inexperienced manager worked here, according to Perkins, but required explicit succession risk and recruiting. One favorable case cannot validate founder-replacement threats generally. [researcher inference · strong] CHM, pp. 20–21
- The case cannot establish initial ownership, terms, person-level proceeds, or that the disputed product roadmap caused the sale. Those claims require the term sheet, cap table, board minutes, product plans, and transaction/shareholder records. [researcher inference · strong] CHM, p. 21
No-hindsight reader exercise
Stop here. Use only the January 1982 facts, or separately the October 1982 prototype facts. Do not use launch, market share, IPO, or Siemens outcomes.
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome — reveal after the exercise
The HBS/Smithsonian-derived chronology reports a $2.5 million January 1982 round split evenly between the existing angel and KPCB. After an over-budget prototype, KPCB supplied an October loan for leading-radiologist tests and manufacturing preparation. Favorable tests preceded 1983 sales and shipments. [retrospective witness · moderate] HBS, “Acuson's Path to Product Launch,” p. 10
Acuson's 128 system reached market in 1983; the institutional artifact record describes it as the first ultrasound system run by a software program. HBS reports that Acuson led US ultrasound sales by 1985 and went public in 1986. These are company/product outcomes, not proof of Perkins's individual alpha. [researcher inference · strong] Smithsonian object record HBS, pp. 9–10
Siemens acquired Acuson in November 2000 for approximately $700 million; an SEC complaint concerning pre-announcement trading fixes the September 27 announcement and $23-per-share cash tender terms. Neither source reports Perkins/KPCB ownership, dilution, distributions, or proceeds. [contemporaneous record · strong] Siemens annual report, acquisitions section SEC v. Dugan, ¶¶1–4
[d] Apple — missed meeting / anti-portfolio decision lab
Evidence boundary
- URL: Computer History Museum, “Venture Capital: The Early Years” | Event: 2002-09-30 | Format: panel video | Reliability: A for Perkins's retrospective account | Access: full | Accessed: 2026-08-23 | Locator: 01:08:03–01:08:43
- URL: Bancroft Library, Thomas J. Perkins: Kleiner Perkins Caufield & Byers: Venture Capital, Entrepreneurship, and the Commercialization of Technology | Interviews: 2007–2008; published 2009 | Format: oral history | Reliability: A for Perkins's retrospective account | Access: full; quotation restricted, paraphrase only | Accessed: 2026-08-23 | Locator: printed p.61, PDF page 76
- URL: TechCrunch, “VC Titans Tom Perkins And Don Valentine Articulate What Makes A Good VC” | Published: 2013-09-11 | Format: event report | Reliability: B | Access: full | Accessed: 2026-08-23 | Locator: closing paragraph
- URL: Apple Computer, Inc., 1980 prospectus | Published: 1980-12 | Format: contemporaneous securities prospectus scan hosted by a third party | Reliability: A for the scanned filing; hosting is unofficial | Access: full | Accessed: 2026-08-23 | Locator: printed pp.27–28, “Principal Shareholders”
- URL: Apple Computer, “The First Ten Years” marketing/history document | Published: 1982 | Format: contemporaneous company-produced history | Reliability: A/B; primary but promotional | Access: full | Accessed: 2026-08-23 | Locator: PDF p.30, financing and December 1980 offering paragraphs
- URL: TIME, “The Seeds of Success” | Published: 1982-02-15 | Format: near-contemporaneous magazine report | Reliability: B | Access: full | Accessed: 2026-08-23 | Locator: paragraphs on Markkula's financing and Apple sales
This was not a conventional evaluated “pass” in the surviving record. Perkins says he did not take a scheduled meeting because he categorized Apple with computer kits he had already disliked. No pitch deck, referral note, meeting date, investment memo, proposed price, security, allocation, partnership vote, or rejection letter survives publicly. The lab therefore tests intake and category-screening discipline, not pricing or formal IC judgment.
What decision actually occurred?
- In the 2002 CHM panel, Perkins said he never met Steve Jobs and did not take a scheduled opportunity after seeing several computer kits he considered unattractive. [investor-stated · moderate] CHM panel, 01:08:03–01:08:43
- In the 2009 Bancroft oral history he again described rejecting Apple without meeting Jobs, reasoning that the apparent technical risk was low and marketing risk was difficult to stage. This is a later explanation, not proof that these words or framework existed at the time. [investor-stated · moderate] Bancroft oral history, printed p.61 / PDF p.76
- In 2013 he summarized the error publicly: “We didn't even look at Steve Wozniak. Big mistake.” This corroborates ownership of the miss but supplies no decision-date detail. [investor-stated · moderate] TechCrunch, closing paragraph
Case classification: researcher inference; strong. The record supports missed/delegated intake screen, not diligenced pass. Calling it a priced pass would invent a level of engagement and information the sources explicitly deny.
Ex-ante information set — freeze before outcome
What Perkins says was known at the screen
- An Apple-related meeting was scheduled or available.
- He had recently seen several computer-kit opportunities and judged that category unattractive.
- He grouped the new opportunity with those prior kits and did not meet Jobs; his later account says Wozniak was not evaluated either. CHM panel, 01:08:03–01:08:43 TechCrunch, closing paragraph
What the public record does not establish was known at that moment
| Question | Status at the missed meeting |
|---|---|
| Exact date and route to K&P | Unknown. Do not merge this episode with Don Valentine's separate referral chain unless a source connects them. |
| Whether both Jobs and Wozniak would attend | Unknown; Perkins's later references establish whom he says he failed to evaluate, not the invitation roster. |
| Which product/prototype was available | Unknown. The record does not establish whether the meeting concerned Apple I, Apple II, a kit, a finished system, or only the founders. |
| Traction, orders, revenue, margins, channel evidence | Unknown. |
| Founder background and division of labor | Unknown to the surviving K&P screen record. |
| Capital sought, proposed valuation/security/ownership | Unknown; no offer terms have been found. |
| K&P partner involvement or delegated decision rights | Unknown. Perkins's non-attendance is clear; who scheduled, screened, or could override him is not. |
| Dissent or alternatives considered | Unknown. No partner memo or meeting record survives. |
Retrospective information barred from the decision
Apple's later financing, public offering, sales, market position, and founder wealth must not enter the screen. Even Apple’s later success does not prove every home-computer kit was investable or that K&P would have obtained a meaningful allocation. It proves only that the cost of failing to inspect this particular exception was enormous in hindsight. [researcher inference · strong]
Team and decision-rights map
| Actor | Supported role in this episode | Unsupported extension to avoid |
|---|---|---|
| Tom Perkins | He says he did not take the meeting and later owned the mistake. CHM, 01:08:03–01:08:43 | Sole firm vote, negotiated rejection, or knowledge of terms |
| Eugene Kleiner / other K&P personnel | Public record reviewed here is silent. | Assuming approval, dissent, or even awareness |
| Steve Jobs | Opportunity/founder Perkins says he did not meet. | Assuming a full pitch was delivered to K&P |
| Steve Wozniak | Technical founder Perkins later says K&P did not evaluate. TechCrunch, closing paragraph | Assuming he was scheduled to attend or that his work was shown |
| Mike Markkula, Venrock, Arthur Rock, Capital Management | Later financing participants according to Apple-produced history and the 1980 prospectus. Apple history, PDF p.30 Apple prospectus, printed pp.27–28 | Treating their later diligence or terms as information K&P possessed |
No-hindsight underwriting frame
No contemporaneous memo exists. The following is an analyst-created process for deciding whether to spend diligence time—not a reconstruction of Apple-specific economics.
| Screen question | Minimum evidence before rejecting | Why it matters |
|---|---|---|
| Is this merely another kit? | Inspect the product architecture, integration, intended user, and distribution plan | A category label can conceal a different product boundary or customer |
| Is there real technical differentiation? | Put the technical founder in front of an evaluator; compare demonstrated capability with existing kits | Perkins later acknowledged Wozniak was not evaluated |
| Can marketing risk be staged? | Test paid orders, dealer interest, user setup time, support load, and a tightly scoped launch | “Marketing risk” need not imply an all-or-nothing check |
| Is founder complementarity unusual? | Meet both technical and commercial founders; map responsibilities and missing hires | A missed meeting supplies no founder evidence |
| Is the category prior still valid? | Write the prior and one observable falsifier before screening | Prevents fatigue from becoming an untestable veto |
Most defensible process failure: researcher inference; strong. The error was not a demonstrably wrong valuation or market forecast; it was failing to buy information cheaply enough to learn whether the new case falsified an adverse category prior.
Potential alternative explanation: researcher inference; moderate. Given many undifferentiated kit pitches, declining a meeting may have been rational time allocation ex ante. A robust critique therefore depends on how cheaply K&P could have inspected the founders/product, how the opportunity reached the firm, and whether there were visible exception signals. Those facts are missing.
Terms, milestones, and kill criteria
- Historical terms: insufficient public record. No K&P security, price, check, ownership, milestone, reserve, or board term was proposed in the recovered evidence.
- Analyst-created first step: take the meeting and run a bounded technical/product inspection; this is an information-gathering decision, not yet an investment.
- Analyst-created kill criteria: no meaningful product differentiation from existing kits; no evidence that non-hobbyists could use the system; inability to demonstrate a plausible channel or service model; founders unable to divide technical/product/commercial work; or capital needs too high for staged validation.
- Analyst-created milestones: demonstrated integrated prototype; paid demand or credible channel commitments; gross-margin/service model; manufacturing plan; and a clear use for the next financing tranche.
All four bullets are researcher inference; insufficient public record as descriptions of K&P's actual Apple process.
Counterevidence and retrieval gaps
- Perkins's repeated admission is the only recovered direct evidence of K&P's screen. No independent witness or decision-date firm record fixes the meeting date, referral route, materials, or who else participated.
- His later explanation that technical risk was low while marketing risk could not be staged may be a genuine memory, a compressed lesson, or hindsight rationalization; no contemporaneous note adjudicates among them.
- The public record does not support a counterfactual K&P ownership percentage, return, “loss,” or claim that the firm would have won an allocation after meeting.
- Search routes attempted: Perkins CHM appearances, Bancroft oral history, TechCrunch event coverage, CHM Apple/Markkula archives, Apple company history, 1980 prospectus/cap table, and near-contemporaneous press. No K&P diary, calendar, referral letter, pitch materials, IC record, offered terms, or Apple corroboration of the K&P meeting was found.
Reader decision lab — answer before revealing the historical outcome
The prompts are analyst-created. Use only the sparse screen record above, not Apple's later financiers, IPO, sales, or fame.
- Would you invest then? First decide the narrower question: would you spend partner/technical-review time on the meeting? If yes, define what evidence would move you from meeting to diligence to a staged check.
- What would kill the deal? State observable product, user, channel, team, and capital-intensity failures—not the label “computer kit.”
- What ownership/terms fit the risk? If diligence earns an investment, choose terms that fit pre-scale hardware, manufacturing, and channel risk; explicitly acknowledge that no historical offer price or allocation is known.
- What evidence unlocks the next check? Specify the prototype demonstration, paid-user behavior, manufacturability, support load, channel commitment, and founder-role evidence required for each tranche.
Actual decision and sourced outcome — reveal after the exercise
Perkins did not take the meeting, and K&P did not evaluate Wozniak or invest according to Perkins's own later accounts. [investor-stated · moderate] CHM panel, 01:08:03–01:08:43 Bancroft oral history, printed p.61 / PDF p.76 Apple instead reports initial outside financing associated with Markkula and later capital from Venrock, Arthur Rock and Capital Management; it sold 4.5 million shares to the public in December 1980. [retrospective witness · moderate] Apple history, PDF p.30 The 1980 prospectus lists Jobs, Markkula, Wozniak, Scott and named outside holders/directors; K&P is not among the listed principal holders. That table is not a complete early financing ledger, so its absence is only a control, not the proof of the miss. [contemporaneous record · strong] Apple prospectus, printed pp.27–28
Near the offering period, TIME reported Apple's rapid sales growth and Markkula's early capital and operating role. Those are sourced outcomes, not evidence the K&P screen could have known. [contemporaneous record · moderate] TIME, paragraphs on Markkula and sales
[d] Ordinary deals, luck, and attribution controls
Why attribution changes the record
The public Perkins narrative mixes at least five units that must not be collapsed: Perkins's personal pre-K&P venture; a K&P fund investment he may have led; a firm investment sourced or led by another partner; a syndicate decision shared with other firms; and a later board/M&A role after the original investment. [researcher inference · strong] The matrix below uses unknown rather than turning later narration or a famous board seat into sourcing credit.
Coverage here is deliberately non-comprehensive. It tests named ordinary/failure/luck cases from the assigned workstream and does not establish a portfolio denominator or economic ranking. Checks, ownership, dilution, reserves, realized cash flows, and vehicle identities are missing for most cases.
Unnamed satellite insurance case — luck at the evidence boundary
- URL: WBUR, “Silicon Valley's Tom Perkins” | Published: 2007-11-28 | Format: radio interview/audio | Reliability: A for Perkins's retrospective account | Access: full audio | Accessed: 2026-08-23 | Locator: 00:31:51–00:33:08
Perkins recalled a roughly $20 million satellite-capacity position that seemed headed toward a roughly $15 million write-off before a launch failure triggered insurance coverage; he explicitly characterized the favorable reversal as luck. [investor-stated · provisional] WBUR, 00:31:51–00:33:08
That is the defensible boundary. The company, fund/vehicle, year, launch provider, satellite, insurer, policy language, premium, beneficiary, exact carrying value, impairment basis, recovery timing, and net cash outcome are all unnamed. “We” does not identify Perkins as source, sponsor, director, or sole decision maker. No independent corroboration was recovered. It would be circular to search candidate satellite failures and choose the one that resembles the anecdote.
Luck interpretation: investor-stated; provisional. The anecdote is valuable because Perkins himself distinguishes outcome from skill. It does not prove that the original underwriting was good, that insurance was deliberately structured as downside protection, or that the reported recovery was a fund return rather than an operating-company balance-sheet event.
Reader discipline: do not construct a full decision lab from this evidence. There is no date-specific information set, company identity, security, terms, alternatives, or outcome record to freeze. The next admissible evidence would be a named portfolio/company record, board or investment memo, launch/insurance filing, or a second participant account.
Attribution matrix
| Candidate | Entity / vehicle boundary | Sourcing and sponsorship | Governance / operating attribution | Outcome evidence | Perkins-credit conclusion |
|---|---|---|---|---|---|
| Snow Job / Advanced Recreation Equipment | Later secondary table places it in KP First Fund; security and ownership unknown. Nebraska report, “KP First Fund (1972–1984)” table | Unknown; Perkins later narrates it. | Unknown. | Perkins calls it a failure; secondary table reports zero 1984 value. | K&P/Fund I case; Perkins-specific decision rights unresolved. See failure lab. |
| Apple | No K&P investment; a missed intake meeting, not an executed vehicle position. | Perkins owns the non-meeting; other K&P roles unknown. CHM, 01:08:03–01:08:43 | None established. | Apple later financed elsewhere and went public. | Perkins anti-portfolio admission, not a Perkins deal. See miss lab. |
| American Athletic Shoe / TRED 2 | K&P/Fund I attribution appears only in retrospective portfolio accounts; contemporaneous ads establish the company/product, not the investor. Lebret, Perkins email reproduced in post | Person-level source/sponsor unknown. | Founder, K&P board seat, follow-ons, and stop decision unknown. | Perkins later called it a failure in an email reproduced by a secondary researcher; no audited economics found. | Firm/fund case; do not assign to Perkins personally. |
| Qume | K&P investment, syndicated/source relationship with Sutter Hill; vehicle/check unknown here. | William Draper III says he brought Qume to K&P after earlier weak referrals; founder David Lee originated the company. Draper oral history, PDF p.15 / printed p.16 | Perkins role beyond receiving/participating is not specified in this account. | Draper says Qume was sold to ITT at a large gain; firm archive lists 1974 early-stage partnership and ITT acquisition. Kleiner Perkins partnership archive, Qume entry | Not a Perkins-originated deal; credit Draper/Lee and K&P team unless more records emerge. |
| Collagen | K&P investment; vehicle/check/ownership not recovered. | Brook Byers attributes the investment to Perkins in the same early period as Genentech. Byers oral history, PDF p.38 | Stanford doctors/scientists originated the company; Howard Palefsky later ran it, per Byers. | Byers says it became public; no offering filing/economics reviewed in this slice. | Person-level Perkins attribution supported by one participant witness; economics/governance remain unresolved. |
| Hybritech | KPCB Fund I; Byers says initial capital was $300,000 with a six-month technical milestone. Byers oral history, PDF p.27 | Brook Byers followed Ivor Royston, brought the opportunity, became founding CEO, and led it; Perkins encouraged him and approved/supportively governed. Byers oral history, PDF p.37–39 | Byers and Perkins both joined the board; Ted Greene later operated the company; Pitch Johnson co-invested, according to the participant accounts. Perkins oral history, printed pp.58–59 / PDF p.73 | Later public/exit outcome requires separate filing reconstruction. | Byers-led deal with meaningful Perkins approval, mentorship, and board support; not a sole Perkins deal. |
| LSI Logic | First round was a $6 million syndicate including Sequoia, K&P, IVP and others; founder says roughly 50% was sold. LSI oral-history panel, PDF p.14 | Wilf Corrigan and founder team originated; Don Valentine, Reid Dennis, Perkins and others supplied syndicated capital/board governance. | Founder account says the company pivoted technology without clearly informing VCs; later Perkins and Valentine negotiated a contingent second-round valuation step-up. LSI panel, PDF pp.32–35 | Company later went public; full economics not reconstructed here. | Shared syndicate/board case with a direct Perkins governance role; not sole sourcing or sole value creation. |
| VLSI Technology | Dataquest archive lists a December 1980 $10 million initial financing from six institutions including K&P. Dataquest “I.C. Start-Ups,” VLSI profile, printed p.388 | Founders Jack Balletto, Dan Floyd, Gunnar Wetlesen and Douglas Fairbairn; contemporary/reconstructed histories identify other institutional leads. CHM Fairchildren, VLSI entry | Reviewed board list does not establish a Perkins seat; absence from one snapshot is not proof of no involvement. | Philips later acquired VLSI according to CHM; investor economics unknown. | K&P firm/vehicle exposure supported; no person-level Perkins sourcing or governance credit established. |
| Andros | Contemporary TIME says K&P had backed Andros since 1972; exact vehicle and capitalization are not supplied. TIME, “Some Thoughtful Gambling,” p.4 | An HBS retrospective bio says Eugene Kleiner recruited George Schapiro to become CEO. HBS event biography, George Schapiro paragraph | Schapiro is credited with transforming the instrumentation company; individual K&P board roles are not recovered. | HBS bio reports growth from about $1 million to $25 million and public-company status; promotional bio, not audited returns. | Firm case with stronger recovered Kleiner/Schapiro attribution; not presently a Perkins deal. |
| University Laboratories | Perkins's personal capital/side venture while employed at HP; it predates K&P and is not a venture-fund investment. | Perkins says he conceived the laser product, obtained David Packard's permission, and supplied money/supervision; exact personal amount conflicts across tellings. CHM oral history, PDF pp.5–6 | Operating employees and Packard's permission were essential; names/decision rights are not fully recovered here. | Perkins says company merged into Spectra-Physics; transaction economics need independent control. | Direct Perkins entrepreneurial/investing precursor, not a K&P portfolio deal. |
| Compaq | K&P participated after Sevin Rosen approached the firm; vehicle/check/ownership not supplied on firm page. | Founders Rod Canion, Bill Murto and Jim Harris originated; Sevin Rosen sourced K&P; John Doerr sponsored internally. Kleiner Perkins, “Compaq: Building a Major New Computer Company” | Firm page credits recruiting Sparky Sparks and says Perkins later served as director and helped with the Tandem and HP combinations. | Historic company outcome is clear, but individual return contribution is not reconstructed. | Not a Perkins original deal; later Perkins governance/M&A role is distinct and supportable. |
| Acuson | K&P investment; check/ownership/vehicle not specified in recovered interview. | Perkins says he interrogated Sam Maslak on acoustical diffraction and committed to invest. CHM Perkins oral history, PDF pp.20–21 | Perkins says K&P recruited key managers and that he later resigned after disagreeing over product timing. | He links resignation to later sale, but the causal link and economics need independent filings. | Direct Perkins deal/governance case; exit causation remains retrospective and uncorroborated. |
| Segway | Perkins's oral history says he was no longer involved with KPCB when the opportunity arose. Bancroft oral history, printed p.61 / PDF p.76 | Other KPCB personnel, not Perkins; exact sponsor outside this slice. | None for Perkins established. | Not analyzed here. | Explicit exclusion from Perkins's record despite possible firm association. |
Case controls and adverse evidence
American Athletic Shoe / TRED 2
Contemporaneous running-magazine advertisements establish a real mail-in resoling service under the spelling TRED 2, operated by American Athletic Shoe Corp. The September–October 1974 ad offered resoling for named running-shoe brands at $9.95; later 1975 ads show $11.95. The PDF scans do not provide a stable printed-page locator for the full-page ads, so the issue and named artifact are the locator. NorCal Running Review, issue 50, Sep–Oct 1974, full-page TRED 2 ad Issue 56, Sep–Oct 1975, full-page TRED 2 ad
The advertising controls product, company, period, channel, and price. It does not establish K&P ownership, founder/team, sales, gross margin, customer retention, failure date, or cause. A secondary research post reproduces an email in which Perkins identifies Advanced Recreation Equipment and American Athletic Shoe as early failures and also names Qume; the post is useful as a direct-email fragment but not an audited portfolio record. [investor-stated · provisional] Hervé Lebret, “About Kleiner Perkins' first fund, episode 2,” Perkins email reproduced in post
Adverse control: researcher inference; moderate. TRED 2 and Snow Job weaken a clean story that K&P began with a fully formed high-technology-only filter. They do not prove Perkins personally sourced either investment or that their failures alone caused a later strategy change.
Qume
William Draper III's CHM oral history is unusually useful because it reallocates sourcing credit: after three weak referrals to K&P, he says he brought Qume, founded by David Lee, and that its sale to ITT pleased Perkins. [retrospective witness · moderate] Draper oral history, PDF p.15 / printed p.16
The firm archive independently confirms a 1974 early-stage K&P partnership and ITT acquisition but is promotional and gives no check, ownership, partner, or return. Kleiner Perkins partnership archive, Qume entry This is an ordinary attribution warning: Perkins's satisfaction with an outcome is not sourcing credit.
Collagen and Hybritech
Brook Byers distinguishes the cases. He attributes Collagen's investment to Perkins, while identifying Stanford physicians/scientists as company originators and Howard Palefsky as later president. [retrospective witness · moderate] Byers oral history, PDF p.38 This supports a Perkins investment association, not check, ownership, board interventions, or return contribution.
For Hybritech, Byers followed Ivor Royston, brought the company, took an operating role, and describes $300,000 of initial financing tied to a six-month scientific milestone. Perkins encouraged and mentored Byers and joined the board. [retrospective witness · moderate] Byers oral history, PDF pp.27 and 37–39 Perkins oral history, printed pp.58–59 / PDF p.73 Hero-worship control: Perkins's approval and board contribution matter, but the record does not support reassigning Byers's sourcing and hands-on company building to Perkins.
LSI Logic and VLSI Technology
The LSI oral-history panel documents collective financing and contested governance. Founders raised a $6 million first round from Sequoia, K&P, IVP and others; the founder recalled selling roughly half the company. The team then changed its technology plan after customer calls without clearly telling the VCs. LSI panel, PDF pp.12–14 Later, Perkins and Valentine resisted an eightfold second-round valuation increase and negotiated a conditional participation structure, while the founder had alternative capital lined up. [retrospective witness · moderate] LSI panel, PDF pp.32–35
This is adverse governance evidence, not merely a winner story: founders could change the technical path before investors understood it, and follow-on pricing involved bargaining power and external financing alternatives. No one investor should receive sole credit.
VLSI is different. A Dataquest archival company profile lists K&P among six institutional sources in a $10 million initial financing, while CHM's founder history names the four founders and identifies other financing leads. Dataquest VLSI profile, printed p.388 CHM Fairchildren, VLSI entry No reviewed source assigns sourcing, sponsorship, diligence, or a board seat to Perkins. Conclusion: documented firm exposure; insufficient public record for person-level Perkins credit.
Andros and University Laboratories
TIME's 1981 profile places Andros within K&P's backed companies, but an HBS retrospective bio says Eugene Kleiner recruited CEO George Schapiro. TIME, “Some Thoughtful Gambling,” p.4 HBS biography, George Schapiro paragraph Attribution conclusion: contemporaneous firm association plus retrospective Kleiner/Schapiro operating credit; insufficient public record for Perkins-specific contribution.
University Laboratories belongs in Perkins's formative record but outside the K&P fund ledger. Perkins says he conceived a lower-cost laser product while at HP, obtained David Packard's permission, and supplied personal capital and supervision before a merger with Spectra-Physics. CHM Perkins oral history, PDF pp.5–6 His accounts conflict on whether the initial personal investment was $10,000 or $15,000. Bancroft oral history, printed pp.18–21 / PDF p.33 The merger economics have not been independently recovered. Conclusion: investor-stated personal venture; moderate, with amount and outcome economics unresolved.
Compaq and Acuson
Kleiner Perkins's own Compaq history credits Sevin Rosen with bringing the opportunity to K&P and John Doerr with sponsoring it, while identifying Rod Canion, Bill Murto, and Jim Harris as founders. It credits Perkins only later as a director involved in the Tandem and HP transactions. Kleiner Perkins, “Compaq” Attribution conclusion: retrospective witness (firm account); strong for internal division of labor, though promotional and ideally controlled by board/filing records. “We started Compaq” is therefore too broad when attributed personally to Perkins.
Acuson is one of the stronger direct Perkins cases: he describes technical diligence with Sam Maslak, a commitment, management recruiting, and a later board resignation over product sequencing. CHM Perkins oral history, PDF pp.20–21 Yet the claim that the company was sold because of or soon after his resignation is retrospective causal narration; an independent chronology, board record, and acquisition filing are required before treating it as established causation. [investor-stated · moderate]
Cross-case decision-rights map
| Function | Strongest supported example | Attribution hazard exposed |
|---|---|---|
| Founder/originator | Lee at Qume; Byers/Royston at Hybritech; Corrigan/team at LSI; four founders at VLSI; Compaq founding trio | Investors do not originate a company merely by financing it. |
| External source/referral | Draper brought Qume; Sevin Rosen brought Compaq to K&P | A K&P outcome is not automatically a Perkins-sourced deal. |
| Internal sponsor | Byers at Hybritech; Doerr at Compaq; Perkins at Acuson | Partnership brand and seniority can erase the actual sponsor. |
| Approval/mentorship | Perkins at Hybritech | Approval and coaching are valuable but distinct from sourcing and company building. |
| Syndicate negotiation | Perkins and Valentine at LSI second round | Terms can be collective and shaped by founder alternatives. |
| Operating work | Byers at Hybritech; Schapiro at Andros; Compaq founders/executives | Investment narratives often under-credit operators. |
| Later governance/M&A | Perkins at Compaq; Perkins at Acuson | Later board influence must not be back-projected into original sourcing. |
| Narration | Perkins on Snow Job and satellite case | Story ownership does not establish decision ownership. |
Failure, pass, luck, and ordinary-outcome controls
- Failure: Snow Job and American Athletic Shoe provide evidence that Fund I included unsuccessful consumer/recreation bets. Only Snow Job has a partially reconstructable causal environment; TRED 2 lacks operating and shutdown records.
- Pass: Apple is a missed meeting/intake failure. It cannot support claims about K&P price discipline, ownership selection, or a partnership vote because no offer or diligence record was recovered.
- Luck: the unnamed satellite anecdote is valuable only as a self-attributed lucky outcome; its anonymity prevents a deal-level return or process claim.
- Ordinary/team-attributed cases: Qume, Collagen, Hybritech, LSI, VLSI, Andros, University Laboratories, Compaq, and Acuson show that
K&P-backed,Perkins narrated, andPerkins dealare not synonyms. - Missing denominator: fame-selected cases cannot establish frequency, repeatability, loss rate, gross/net performance, or the record excluding Genentech. [researcher inference · strong] A defensible portfolio conclusion requires complete Fund I–vehicle schedules and cash flows, not anecdotes.
Uncertainty and retrieval ledger
| Gap | Routes attempted | Evidence needed to resolve |
|---|---|---|
| Satellite identity/economics | WBUR direct-voice segment; candidate identification deliberately not inferred | Named company/vehicle plus investment, launch, insurance, and cash records |
| Snow Job and TRED 2 ownership/loss | Product/company aliases, ad scans, participant accounts, later portfolio table | Fund schedule, cap table, board minutes, shutdown/liquidation records |
| Qume person-level K&P role | Draper CHM oral history; K&P archive | Qume financing filing, board list, K&P memo, Lee interview |
| Collagen terms/governance | Byers oral history | Offering prospectus, early cap table, board records, Palefsky/Perkins account |
| Hybritech exact economics | Byers and Perkins oral histories | Financing/IPO/acquisition filings reconciled to fund cash flows |
| LSI ownership and partner return | Founder panel | Prospectus ownership table, partnership allocation, follow-on cash flows |
| VLSI Perkins role | CHM founder history, Dataquest company profile | Board filings, K&P records, founder/partner attribution |
| Andros chronology and role | TIME, HBS bio | Early financing/corporate filings and board history |
| University Laboratories amount/outcome | Perkins CHM and Bancroft accounts | HP permission record, incorporation/cap table, Spectra-Physics transaction record |
| Compaq original and later roles | K&P firm history | SEC prospectus/proxies and transaction documents; independent participant testimony |
| Acuson causal exit claim | Perkins CHM oral history | Board chronology, resignation record, acquisition filing, independent director/operator account |
Overall attribution judgment: researcher inference; strong for the boundary, moderate for case completeness. The defensible Perkins record contains direct cases, shared cases, externally sourced firm cases, later governance roles, a personal pre-fund venture, an anti-portfolio miss, failures, and at least one self-described lucky reversal. Treating all of them as a single personal track record would systematically overstate both skill and failure while hiding how the partnership actually worked.
[d] Tom Perkins — denominator, vehicles, and return boundaries
- Scope: public reconstruction of Tom Perkins, Kleiner & Perkins/KPCB vehicles, portfolio-count claims, and reported performance through the end of Perkins's active investing period, with later firm totals used only as denominator controls.
- Accessed: 2026-08-23.
- Evidence rule: every material statement carries an attribution type and confidence. “Strong” means the public record is sufficient for the narrow claim, not that an unaudited economic claim has become audited.
Bottom line
- The first partnership's $8 million commitment and roughly $4 million deployed are Perkins's late recollections, not an inspected capital-account schedule. His claim that the balance was returned is plausible but does not tell us whether the $4 million excludes fees, reserves, recycling, or later calls. [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–35 WBUR, 00:02:55–00:03:56
- The public denominator is not defensible enough for a coverage fraction. Secondary sources conflict between 17 companies and 18 investments; Perkins himself later said 17 sounded too high, while his Bancroft memory accounts for only Tandem, Genentech, and roughly seven or eight other deals. No public partnership ledger resolves whether rounds, companies, write-offs, side vehicles, or pre-fund positions are counted. [researcher inference · strong] Nebraska report, p. 7 FundingUniverse, “Origins,” paras. 3–5 Perkins email reproduced by startup-book.com Bancroft/NVCA, printed pp. 38–40
- Public sources support a two-outlier shape, not an audited Fund I return: Perkins consistently identifies Tandem and Genentech as the exceptional outcomes and remembers the other early deals as roughly break-even in aggregate. The size, timing, ownership, dilution, fees, carry, and distributions needed to calculate TVPI, DPI, IRR, or a person-level multiple are absent. [investor-stated · moderate] Red Herring, first-fund and return answers Bancroft/NVCA, printed p. 38
Count reconciliation
These claims are not points on one cumulative series. They change speaker, date, unit, period, inclusion rule, and sometimes source family.
| As-of / claim | What the source actually supports | Boundary and disposition | Evidence |
|---|---|---|---|
| Fund I: 17 companies | A 2014 state-commissioned secondary report describes just over $7 million invested in 17 companies during 1973–77; another educational article repeats a reported 17-company count, reported 31x, and reported 51% IRR. | Neither is a partnership record. The $7 million conflicts with Perkins's roughly $4 million recollection, so the count is a secondary hypothesis, not the denominator. | [researcher inference · provisional] Nebraska report, p. 7 EIX, paragraph beginning “The first venture portfolio” |
| Fund I: 18 investments | A 2003 company-history compilation reports 18 investments, eight losers, reported 41.5% over ten years, reported $1.6 million in Tandem, and reported 200:1 on Genentech. | It cites press/book sources but supplies no fund schedule. “Investments” may count differently from “companies”; all economics remain lead-quality. | [researcher inference · provisional] FundingUniverse, “Origins,” paras. 3–5 |
| Fund I: Perkins's memory | Perkins later said 17 seemed too many; in the full Bancroft interview he recalls Tandem, Genentech, and roughly seven or eight other first-partnership deals that were only barely profitable. | The email is reproduced by a third party, and the oral history expressly relies on memory. It disproves false precision but cannot supply a replacement denominator. | [retrospective witness · moderate] startup-book.com reproduction Bancroft/NVCA, printed p. 38 |
| 1984: 86 companies | Frank Caufield told Congress that KPCB and two successor partnerships had invested in 86 companies as of his testimony. | This is the cleanest dated early firm count, but it spans three KPCB partnerships after Caufield joined. His nearby “approximately 20” sentence describes Oak Grove Ventures, not K&P, and must not be imported into Fund I. | [contemporaneous record · strong] JEC hearing, p. 166 |
| 1993: 180 deals | Perkins said the firm had done 180 deals by the original Red Herring interview. | “Deals” is undefined and may mean companies, financings, or firm-created companies; it is not Perkins-personal attribution. | [investor-stated · moderate] Red Herring, deal-sourcing answer |
| Circa 2018 publication: 836 investments / 450+ companies | A later academic-book chapter reports database-style totals of 836 investments in more than 450 companies. | This is a secondary snapshot, not a claim made by Perkins or KPCB and not tied to a reproducible extraction date. It is retained only to explain the “450” number. | [researcher inference · provisional] Accelerators in Silicon Valley, p. 48 |
| 2007–11: 525 companies | Perkins used more than 525 as the size of the firm's company/referral network and again described 525 KPCB companies when explaining why indefinite board service was impossible. | Repetition strengthens speaker attribution, not auditability. The number is rounded, firmwide, and definition-incomplete. | [investor-stated · moderate] WBUR, 00:25:51–00:26:58 CHM, pp. 19–20 |
| 2012: 500+ companies | A Wharton-edited report attributes “more than 500” to Perkins in a referral-network answer. | The lower rounded number after 525 shows why these are rhetorical magnitudes, not a monotonic investment ledger. | [investor-stated · provisional] Wharton, closing Q&A |
| 2022: 1,000+ companies | KP's current partnership says it had been an early backer of more than 1,000 companies over five decades and reported more than $35 billion returned to LPs. | This includes later partners, 20 generations of early funds, select/growth vehicles, and decades after Perkins's active tenure. It is an official promotional firm claim, not Perkins's record. | [contemporaneous record · moderate] KP20/KP Select II, paras. 2–4 |
Search denominator and exclusions
| Gate field | Result | Evidence / reason |
|---|---|---|
| Publicly claimed / known | 17 or 18 Fund I companies/investments; 86 firm companies through three partnerships by 1984; later 180, 450+, 500+/525, and 1,000+ firm counts. | [researcher inference · strong] count table |
| Candidate records searched | 15 candidate company paths, including investments, one pass, and one later-governance attribution control. | [documented behavior · strong] uniform candidate ledger |
| Candidate records researched | 15/15 to at least one public source; several terminate at an honest access or attribution boundary. | [documented behavior · strong] uniform candidate ledger |
| Sufficiently reconstructed for outcome attribution | University Laboratories, Tandem, Genentech, Hybritech, Acuson, and the Apple pass; only Tandem, Genentech, Acuson, and Apple have sufficiently detailed public decision-process evidence for learning analysis. | [researcher inference · moderate] three Pareto views |
| Unresolved | Exact Fund I roster; ordinary-company identities; all complete fund cash flows; most vehicle assignments; check/security/ownership/dilution for non-signature deals. | [researcher inference · strong] missing cash-flow fields |
| Excluded | HP and News Corp public-board episodes; later firm-logo wins such as Amazon, Google, and Netscape; Compaq's original selection as a Perkins deal. | A board seat is not an investment, and the original Compaq investment is attributed by the firm to Sevin Rosen's referral and John Doerr's sponsorship. [researcher inference · strong] KP Compaq Compaq proxy |
| Coverage fraction | Not computable. | A numerator of named public cases divided by 17 or 18 would combine a research candidate list with an unverified fund denominator. [researcher inference · strong] 17/18 conflict |
Vehicle, firm, and person boundary
| Layer | Publicly supportable statement | What remains unknown | Evidence |
|---|---|---|---|
| University Laboratories | Perkins says he used household capital and supervised an operating company while employed at HP with Packard's permission; it later merged into Spectra-Physics. | Capital, outside financing, equity split, consideration, taxes, and personal proceeds. It is not a K&P fund investment. | [investor-stated · moderate] CHM, pp. 5–6 |
| Kleiner & Perkins / first partnership | Perkins and Eugene Kleiner formed the 1972 partnership; Perkins reports $8 million committed, roughly $4 million deployed, and the balance returned. The partnership, not Perkins personally, owned investment interests. | Executed agreement, LP commitments, GP commitment, fee base, calls, reserves/recycling, audited financials, distribution waterfall, and final liquidation. | [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–35 |
| KPCB successor partnerships | Tandem's 1991 proxy identifies Perkins as GP of K&P (1972), KPCB I (1978), II (1980), III (1982), and IV (1986). These are separate legal/economic vintages, not one evergreen pool. | Commitments, cash flows, allocation rules, ownership, partner carry shares, and which vehicle funded most named companies. | [contemporaneous record · strong] Tandem 1991 proxy, Perkins biography |
| Firm brand | K&P became KPCB after Caufield and Byers joined, then added later partners and vehicles. Firm totals aggregate decisions by many sponsors. | A firm-logo portfolio list cannot allocate sourcing, selection, governance, or economics to Perkins. | [researcher inference · strong] JEC hearing, p. 166 KP 2022 |
| Company / portfolio security | Issuer records can verify boards, IPOs, acquisitions, and sometimes entity holdings; they do not automatically expose underlying LP or partner economics. | Purchase dates, security classes, transfers, dilution, lockups, distributions in kind, sale timing, fees, taxes, and carry. | [researcher inference · strong] Genentech prospectus Tandem issuer records |
Vehicle economics and incentives
- Capital source and form: the public record describes a finite limited partnership backed by LPs including institutional/family interests; the Hillman and Rockefeller-linked fundraising memories are retrospective. The executed subscription and partnership documents are not public. [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–33
- Deployment and reserves: Perkins's $8 million / $4 million / balance-returned account implies unusually low deployment, but without calls, expenses, recycling, reserves, and closing statements it cannot be treated as invested-cost data. [researcher inference · strong] Bancroft/NVCA, printed pp. 31–35
- Carry: Perkins recalls that the profit share changed from 20% to 30% in connection with Henry Hillman's offer to anchor a later, intended $100 million fund. The effective vehicle, hurdle, catch-up, clawback, vesting, and partner allocation are not in the transcript; applying 30% to Fund I would be false precision. [investor-stated · provisional] Bancroft/NVCA, printed pp. 36–37
- Governance and conflicts: Perkins says an early “Investor's Bill of Rights” barred personal portfolio investments, required immediate profit distribution and independent audits, and separated new funds from follow-ons. The underlying artifact was missing in 2009 and reportedly recovered by 2011 but still was not reproduced, so these remain participant descriptions rather than enforceable terms inspected here. [investor-stated · moderate] Bancroft/NVCA, printed pp. 33–35 and 40 CHM, pp. 17–18
- Decision rights: K&P used close board involvement and, in Perkins's description, first-investor follow-on signaling as practical control even at low ownership. That mechanism can support troubled companies but also creates founder dependence; formal protective provisions are unavailable. [researcher inference · strong] CHM, pp. 19–20
- Liquidity and succession: Perkins says profits were distributed rather than warehoused and later reduced his board load; public records establish Tandem and Genentech IPOs, but not the dates or prices at which fund shares were distributed or sold. [researcher inference · strong] Red Herring Tandem issuer records Genentech prospectus
Performance and return-claim ledger
| Claim | Narrow interpretation | Why no audited return follows | Evidence |
|---|---|---|---|
| $8 million committed; roughly $4 million invested; remainder returned | Fundraising/deployment memory. | No call, fee, reserve, recycling, or distribution schedule. | [investor-stated · moderate] Bancroft/NVCA, printed pp. 31–35 |
| reported: Fund I returned “hundreds of millions” | Perkins's 1993 order-of-magnitude recollection. | Gross/net, valuation/distribution, dates, and recipient base undefined. | [investor-stated · provisional] Red Herring, first-fund answer |
| reported: Fund I 51% annualized; 16% without Tandem and Genentech | A secondary review's summary of Tom Nicholas's historical reconstruction. | Underlying cash-flow series and calculation convention are not public here; do not call it audited LP IRR. | [researcher inference · moderate] Hathaway review, Fund I discussion |
| reported: Fund I 41.5% over ten years; 18 investments | A tertiary company-history compilation. | Conflicts with the 17/51% account and supplies no source table or cash flows. | [researcher inference · provisional] FundingUniverse, “Origins” |
| reported: Tandem $1.4 million to $12.5 million at IPO / $152 million if held to 1984; Genentech $200,000 to $300+ million at IPO | Secondary marked-value examples illustrating outlier scale. | “If held,” IPO value, and entity value are not realized fund distributions; ownership, dilution, follow-ons, and timing are incomplete. | [researcher inference · provisional] Hathaway review, Fund I discussion |
| reported: K&P approximately 40% compounded annually over almost 30 years | Perkins's 2001 firm-level self-assessment. | Period endpoints, net/gross convention, vehicle set, cash flows, fees, and survivorship are unspecified. | [investor-stated · provisional] Genentech oral history, printed p. 2 |
| reported: roughly $30 billion divided among investors and management by 1993 | Perkins's firmwide cumulative distribution claim. | “Divided” is undefined; no fund, LP, partner, or date schedule. | [investor-stated · provisional] Red Herring, firm-economics answer |
| reported: more than $35 billion returned to LPs over five decades | Current firm's promotional cumulative distribution claim. | Includes many later vehicles and partners; no Perkins-period slice or audit attachment. | [contemporaneous record · moderate] KP 2022, para. 4 |
No multiple is calculated from any entry and exit valuation in this package. At most, the public record supports a directional statement: Tandem and Genentech were very large positive outliers; several ordinary positions were near break-even or losses; the share of Fund I economics attributable to each is unknown. [researcher inference · strong] Bancroft/NVCA, printed pp. 38–40
Missing cash-flow fields
For Fund I and every company-level return claim, a defensible calculation still needs:
- Legal vehicle and closing; commitment by LP; GP commitment; call date and amount; fee/expense allocation; recycled proceeds; reserves; and final liquidation. [researcher inference · strong] Bancroft/NVCA
- Company, round, date, check, security, price, fully diluted ownership, pro rata, follow-ons, transfers, option-pool dilution, stock splits, lockups, distributions in kind, sale dates/prices, dividends, write-offs, and taxes. [researcher inference · strong] Genentech prospectus
- Gross and net cash-flow series; NAV methodology; carried-interest rate and base; management fees; hurdle/catch-up/clawback; and partner-level carry allocation. [researcher inference · strong] Bancroft/NVCA carry account
- A contemporaneous portfolio schedule mapping each company to its vehicle. Until these fields are recovered, TVPI, DPI, RVPI, gross/net IRR, PME, loss ratio, and Perkins-personal return are not computable. [researcher inference · strong] 17/18 conflict
Historical context
- K&P launched when institutional venture was small and the limited-partnership form offered pass-through taxation, flexible profit sharing, and less public reporting than a closed-end investment company. Those advantages also explain today's evidence gap: a private LP was not required to publish company-level cash flows. [researcher inference · moderate] Hathaway review, limited-partnership discussion
- The 1974 market/oil shock matters to both capital supply and Snow-Job's distribution story. Later changes to capital-gains treatment and pension-fund rules enlarged the pool only after the earliest K&P decisions; later-scale economics should not be projected backward onto Fund I. [researcher inference · moderate] 1978 Congressional Record, p. 15697 Red Herring, Snow-Job
- Genentech was not finance acting alone: federally supported university science, Cohen–Boyer recombinant-DNA work, external laboratories, licensing institutions, regulatory pathways, and pharmaceutical partners were part of the production system. [researcher inference · strong] Genentech founders Genentech oral, printed pp. 1–29
- By 1984 KPCB and two successor partnerships had raised about $220 million and invested in 86 companies, showing how rapidly the institution outgrew the $8 million founding vehicle. That scale transition is firm history, not proof that Perkins individually chose or generated each outcome. [contemporaneous record · strong] JEC hearing, p. 166
Retrieval ledger and remaining records
- Processed full public transcripts for Bancroft/NVCA, Genentech, and CHM, plus the full 1993 Red Herring interview and WBUR official audio-derived note. [documented behavior · strong] Bancroft/NVCA Genentech oral CHM Red Herring WBUR
- Processed issuer/official controls: Tandem annual-report/proxy compilation, Genentech prospectus appendix, Acuson transaction filing, Compaq proxy, 1984 JEC testimony, and current KP announcement. [documented behavior · strong] Tandem Genentech Acuson Compaq JEC KP
- Highest-value missing artifacts: executed 1972 partnership agreement and subscriptions; audited statements and capital accounts; complete portfolio schedule; company cap tables and transfer ledgers; distribution notices; partner carry allocations; and the recovered Hillman copy of the original plan. [researcher inference · strong] CHM, pp. 17–18
[d] Genentech — prove the science before financing an integrated drug company
- Decision dates: January–April 1976 formation and seed authorization; 1977 somatostatin proof; 1978 insulin/commercial-partner and financing decisions; late 1979–October 1980 public-financing decision.
- Core sources: Genentech Amendment No. 2 to Form S-1, filed October 14, 1980, reproduced at appendix E, printed pp. 102–152 (contemporaneous issuer record); Genentech official founders page, founder sections (issuer retrospective); Tom Perkins Genentech oral history, printed pp. 3–13 and 20–24 (retrospective participant; Bancroft publication rights require paraphrase); Tom Perkins Bancroft/NVCA oral history, printed pp. 45–57 (retrospective participant; rights restriction); Brook Byers oral history, printed pp. 19–24 (retrospective participant); Fred Middleton oral history, printed pp. 13–18 and 26–38 (retrospective executive, bundled with the S-1); Washington Post, October 15, 1980 (decision-near independent press).
- Access: full public PDF/HTML routes inspected 2026-08-23. No transcript quotation is reproduced. The 1976 term sheet, initial cap table, research contracts, board minutes, and Fund I ledger were not found publicly.
- Reliability: A for S-1 formal roles, ownership, risks, milestones, and use of proceeds; B for decision-near press and cross-checked participant recollections; C for recalled causal credit and unreconciled seed amounts.
Case boundary and attribution
Robert Swanson and Herbert Boyer founded Genentech in 1976. Boyer and Stanley Cohen's recombinant-DNA work supplied the scientific foundation; Swanson recognized the commercial possibility, recruited Boyer, and operated the company. K&P supplied early capital and governance; Perkins served as chair and participated in staging, financing, licensing, and the IPO decision. Treating Perkins or K&P as the scientific or sole corporate founder is false. [contemporaneous record · strong] S-1, “Certain Transactions,” printed p. 136 Genentech founders page
The S-1 fixes the formal boundary: Swanson was president/CEO/director from April 1976; Boyer was vice president/director; Perkins was chair and a director from April 1976. The early board was therefore founder/scientist/investor governance, not investor-only control. [contemporaneous record · strong] S-1, “Management,” printed pp. 132–133
Decision 1 — early 1976: finance a falsifiable molecular proof
Frozen information set
| State at the decision date | Reconstruction |
|---|---|
| Known | Cohen and Boyer had demonstrated recombinant-DNA techniques, but no Genentech product, laboratory, production system, clinical program, regulatory approval, or pharmaceutical distribution existed. [contemporaneous record · strong] Genentech founders page S-1, “The Company” and “Business,” printed pp. 107 and 115–121 |
| Known | Swanson and Boyer formed a January 1976 partnership to investigate feasibility; Genentech incorporated April 7. The 1980 S-1 records each founder's formation contribution as $500 cash plus half of that partnership interest. [contemporaneous record · strong] S-1, “Certain Transactions,” printed p. 136 |
| Believed | Engineered microorganisms could express human proteins with commercial value; somatostatin offered a shorter, cheaper proof target than immediately building an integrated insulin company. The experiment could be contracted across existing laboratories rather than requiring a full internal build. [retrospective witness · strong] Perkins Genentech oral history, printed pp. 4–7 |
| Unknown | Whether the gene could be synthesized, inserted, expressed, detected, and reproduced; whether university/contract rights would be enforceable; future NIH/FDA rules; patent breadth; scale-up; clinical safety/efficacy; market acceptance; and capital needed to reach a marketed drug. [researcher inference · strong] S-1, “Risk Factors,” printed pp. 107–109 |
| Disputed | Retrospective accounts cite $100,000, $150,000, $200,000, or about $250,000 at different apparent stages. The public record does not reconcile date, tranche, entity, or security; the defensible fact is a small staged program, not one clean seed number. [researcher inference · insufficient public record] Perkins NVCA oral history, printed pp. 47–49 Middleton, printed p. 13 |
Alternatives and financing context
- Reject as too early; license the university science to an incumbent; finance the founders' proposed integrated laboratory/company; fund only a contracted proof; or retain the opportunity inside K&P. Participant accounts support the contracted-proof choice, but no 1976 board memo ranks these alternatives. [retrospective witness · moderate] Perkins Genentech oral history, printed pp. 4–7
- The choice transferred cash and fixed-cost risk to UCSF/City of Hope/Caltech, but created coordination and intellectual-property risk. Later rights disputes mean contracting did not eliminate risk; it changed its form. [researcher inference · strong] Perkins Genentech oral history, printed pp. 5–9 S-1, “Research and Development,” printed pp. 119–120
- Byers's account independently preserves the same logic: apply modest capital to the most important scientific risk through external laboratory projects. It remains a participant recollection, not a decision-date document. [retrospective witness · moderate] Byers, printed pp. 22–24
Risks, kill criteria, terms, and milestones
- The implicit scientific kill criterion was failure to express and identify the selected human protein through the contracted sequence. Equipment, steps, measurement, time, and success criteria were the core diligence questions Perkins says he asked Boyer. [investor-stated · strong] Perkins Genentech oral history, printed pp. 4–5
- Rights were a coequal kill criterion: absent contracts assigning usable commercial rights from the institutions and scientists, technical success would not create financeable company value. The S-1 later disclosed royalty obligations to UCSF and City of Hope and acknowledged unknown third-party patent/licensing costs. [contemporaneous record · strong] S-1, “Risk Factors—Proprietary Technology,” printed p. 108 S-1, “Research and Development,” printed pp. 119–120
- Milestones: form founder partnership/company → contract the experiment → express somatostatin in 1977 → move to insulin → demonstrate human insulin in 1978 → find scale/manufacturing/distribution partner. [contemporaneous record · strong] S-1, “Business” and “Products,” printed pp. 115–121
- Initial security, K&P ownership, preference, board vetoes, founder vesting, and reserve commitments remain unknown. Do not infer them from the 1980 table. [researcher inference · strong] S-1, “Certain Shareholders,” printed pp. 137–138
Decision 2 — 1977–79: commercial partnership versus full integration
Frozen information after somatostatin proof
| State at the decision date | Reconstruction |
|---|---|
| Known | Genentech produced somatostatin in 1977 and human insulin in 1978; those were technical milestones, not marketed-drug approvals. [contemporaneous record · strong] S-1, “Products,” printed pp. 120–121 |
| Known | In August 1978 Genentech granted Lilly exclusive worldwide manufacturing/marketing rights for human insulin in exchange for royalties; Lilly then funded/handled scale-up and clinical work. [contemporaneous record · strong] S-1, “Human Insulin and Proinsulin,” printed pp. 120–121 |
| Known | Wilmington bought the equivalent of 250,000 post-recapitalization common shares for $500,000 in May 1978; Lubrizol bought the equivalent of 1,000,000 shares for $10 million in September 1979 and received a board-representation covenant while it held the threshold stake. [contemporaneous record · strong] S-1, “Certain Transactions,” printed p. 136 |
| Believed | Pharmaceutical partners could fund development, scale, regulatory work, and distribution while Genentech retained cash, royalties, and learning. The firm still intended eventually to develop manufacturing and direct sales capability. [contemporaneous record · strong] S-1, “The Company,” printed p. 107 |
| Unknown | Approval timing, ultimate royalty revenue, contract cancellation, manufacturing economics, patent scope, clinical outcomes, and whether Genentech could retain talent/control while depending on incumbents. [contemporaneous record · strong] S-1, “Risk Factors,” printed pp. 107–109 and “Marketing,” printed pp. 124–126 |
Alternatives, dissent, and board intervention
- Alternatives were to sell or license each program; build a fully integrated pharmaceutical company immediately; mix partner-funded products with independently retained programs; or sell the company to a major pharmaceutical firm. The observed strategy was a portfolio of licenses and contract funding while building internal capability. [researcher inference · strong] S-1, “Products,” “Marketing,” and “Use of Proceeds,” printed pp. 111 and 120–126
- The S-1 says major customer contracts were cancellable on short notice in early research stages and benchmark-dependent; even completed benchmarks did not assure commercialization or royalties. A rational stop rule therefore required both scientific evidence and partner willingness to continue. [contemporaneous record · strong] S-1, “Marketing,” printed pp. 124–126
- Perkins claims a role in Lilly negotiations and financing design, but credits Swanson with major territory/licensing work; Middleton identifies CFO/legal/negotiating contributions. Attribution belongs to the board, management, scientists, counsel, and partners, not Perkins alone. [retrospective witness · strong] Perkins Genentech oral history, printed pp. 8–10 and 22–24 Middleton, printed pp. 16–18
Decision 3 — late 1979 to October 1980: IPO before product sales
Frozen information set from the S-1
| State at the decision date | Reconstruction |
|---|---|
| Known | Genentech had contract-development revenue but no product sales to end users; it had a $691,000 accumulated deficit at June 30, 1980 and expected further losses. [contemporaneous record · strong] S-1, “Risk Factors—Early Development Stage,” printed p. 107 |
| Known | The offering proposed 1,000,000 new shares at $35, leaving 7,472,102 outstanding; estimated net proceeds were $32.315 million, with about $18 million for facilities/equipment and the balance for R&D, clinical work, inventories, receivables, and working capital. [contemporaneous record · strong] S-1, “Prospectus Summary,” “Dilution,” and “Use of Proceeds,” printed pp. 105 and 109–111 |
| Known | New investors would pay $35 per share versus an average $1.85 paid by existing shareholders, with $29.18 immediate dilution from offering price to pro-forma net tangible book value. [contemporaneous record · strong] S-1, “Dilution,” printed pp. 109–110 |
| Known | As of July 31, 1980 Swanson and Boyer each held 14.3%, Lubrizol 24.0%, and K&P 14.5%. Perkins was identified as K&P's general partner; the table did not assign K&P's stake to Perkins personally. [contemporaneous record · strong] S-1, “Certain Shareholders,” printed pp. 137–138 |
| Believed | Going public first could finance an integrated build, establish category leadership, attract talent, and preserve independence. Perkins and Middleton differ on whether K&P liquidity/fundraising was an important motive. [retrospective witness · moderate] Perkins Genentech oral history, printed pp. 12–13 Middleton, printed pp. 27–30 |
| Unknown | FDA timing, product efficacy, patent scope, capital sufficiency beyond 18 months, competitor entry, scientist retention, public-market volatility, and whether contract customers would commercialize successful programs. [contemporaneous record · strong] S-1, “Risk Factors,” printed pp. 107–109 |
Risks, kill criteria, terms, and dissent
- The issuer's own kill set was explicit: inability to turn development contracts into profitable products; insufficient capital; adverse regulation or failed FDA approval; unusable/blocked IP; superior incumbent competition; loss of technical edge/talent; and public-price volatility. [contemporaneous record · strong] S-1, “Risk Factors,” printed pp. 107–109
- Alternatives included more private strategic equity, sale to big pharma, debt/lease financing, delay, or IPO. Middleton says sale discussions did not produce an acceptable transaction and management concluded by May/June 1980 that public financing was the best route. This is executive recollection, not board minutes. [retrospective witness · moderate] Middleton, printed pp. 29–30
- Swanson reportedly resisted public-company burden; Perkins pressed category leadership. The S-1's risk disclosure shows why either position was rational ex ante. [retrospective witness · moderate] Perkins Genentech oral history, printed pp. 12–13
- No reviewed record establishes K&P's selling shares in the IPO. The prospectus states all 1,000,000 offered shares were sold by the company; K&P disclosed an intended post-offering distribution to its partners, which is not the same as company proceeds or Perkins's personal sale. [contemporaneous record · strong] S-1 cover and “Certain Shareholders,” printed pp. 104 and 137
What the case can and cannot teach
- The strongest portable mechanism is not “bet on biotech”; it is to convert an expensive integrated-company proposal into a rights-secured, falsifiable proof, then add capital after evidence. [researcher inference · strong] Perkins Genentech oral history, printed pp. 4–9
- Licensing and partner funding reduced manufacturing/distribution exposure but introduced dependence, cancellation, and rights risks. [researcher inference · strong] S-1, “Marketing,” printed pp. 124–126
- The S-1 is a powerful no-hindsight control precisely because it records no product sales, severe scientific/regulatory uncertainty, and substantial new-investor dilution immediately before the celebrated offering. [researcher inference · strong] S-1, “Risk Factors” and “Dilution,” printed pp. 107–110
- This case cannot establish person-level return, exact seed ownership, or that Perkins rather than Swanson, Boyer, Middleton, scientists, counsel, partners, and the board caused the outcome. [researcher inference · strong] S-1, “Management,” “Certain Transactions,” and “Certain Shareholders,” printed pp. 132–138
No-hindsight reader exercise
Stop here. Use only the frozen 1976 information for the seed decision, or the S-1 information for the 1980 financing decision. Do not use the first-day trade or Roche outcomes.
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome — reveal after the exercise
K&P financed the founders' staged research program; Genentech used outside laboratories to prove somatostatin in 1977, advanced to human insulin in 1978, and partnered with Lilly for manufacturing, clinical work, and distribution. The exact K&P seed amount and security remain unreconciled. [documented behavior · strong] S-1, “Business” and “Products,” printed pp. 115–121 Perkins Genentech oral history, printed pp. 4–9
The October 1980 IPO sold 1,000,000 company shares at $35. Contemporary reporting records exceptional demand and a rapid first-day price increase, but that market reaction is not evidence that the ex-ante scientific/regulatory risks had disappeared. [contemporaneous record · strong] S-1 cover and summary, printed pp. 104–105 Washington Post, October 15, 1980
Roche began its ownership relationship in 1990 and acquired the remaining public interest in March 2009 at $95 cash per share. Those later company outcomes do not reveal K&P dilution, distributions, carry, or Perkins's personal realized proceeds. [contemporaneous record · strong] Genentech/Roche 2009 Form 8-K SEC tender completion filing
[d] Snow Job / Advanced Recreation Equipment — failure decision lab
Evidence boundary
- URL: Computer History Museum, “Venture Capital: The Early Years” | Event: 2002-09-30 | Format: panel video | Reliability: A for Perkins's retrospective account | Access: full | Accessed: 2026-08-23 | Locator: 00:51:23–00:51:49
- URL: WBUR, “Silicon Valley's Tom Perkins” | Published: 2007-11-28 | Format: radio interview/audio | Reliability: A for Perkins's retrospective account | Access: full audio | Accessed: 2026-08-23 | Locator: 00:12:35–00:15:43
- URL: Zeitgeist Films, Something Ventured study guide | Published: undated, documentary released 2011 | Format: official distributor study guide | Reliability: B | Access: full | Accessed: 2026-08-23 | Locator: documentary sequence 00:21:31:20–00:23:40
- URL: Norbert Haupt, Nebraska Investment Council venture-capital report | Published: 2014 | Format: state-hosted secondary report | Reliability: B/C for the reconstructed Kleiner Perkins table; not a fund audit | Access: full | Accessed: 2026-08-23 | Locator: table headed “KP First Fund (1972–1984),” row “Advanced Recreation Equipment”
- URL: Federal Reserve History, “Oil Shock of 1973–74” | Published: 2013-11-22, updated 2023-11-22 | Format: official historical essay | Reliability: B | Access: full | Accessed: 2026-08-23 | Locator: sections “Oil Embargo Begins” and “The Embargo Ends”
- URL: Federal Reserve Bank of Minneapolis, December 1973 Beige Book | Published: 1973-12-12 | Format: contemporaneous Federal Reserve conditions report | Reliability: A/B | Access: full | Accessed: 2026-08-23 | Locator: paragraph reporting declining recreational-vehicle activity and fuel-shortage employment effects
- URL: The Kneeslider, “SnowJob Twin Tracked Motorcycle Conversion Kit” | Published: undated | Format: hobbyist article reproducing a 1973 advertisement | Reliability: C for commentary; B for the visible ad artifact | Access: full | Accessed: 2026-08-23 | Locator: reproduced advertisement headed “Snow Job,” Advanced Recreation Equipment, Mountain View; original magazine title/page not supplied
This reconstruction has one direct narrator—Perkins, decades after the event—and no surviving investment memo, partnership minutes, capitalization table, board record, operating statements, or liquidation record. Accordingly, the historical claims below distinguish his retrospective account from independent controls. The decision rationale, team roles, terms, and causal lesson remain incompletely documented.
Attribution and identity
- Perkins describes Snow Job as one of Kleiner & Perkins's early investments and as a motorcycle-to-snowmobile conversion. [investor-stated · moderate] The accounts are internally consistent across the 2002 CHM panel and 2007 WBUR interview, but both are retrospective and derive from the same witness. CHM panel, 00:51:23–00:51:49 WBUR, 00:12:35–00:15:43
- A reproduced 1973 advertisement identifies the seller as Advanced Recreation Equipment of Mountain View and depicts the Snow Job conversion kit. This independently controls product/company identity, not investment ownership, product performance, demand, safety, or economics. [contemporaneous record · moderate] Reproduced 1973 ad, no original page supplied
- A later state-hosted portfolio reconstruction places Advanced Recreation Equipment in the “KP First Fund (1972–1984)” with a 1973 investment, reported cost of $229,200, and zero value at June 30, 1984. This is not an audited K&P statement, so both the vehicle and economics remain reported rather than verified. [researcher inference · provisional] Nebraska report, “KP First Fund (1972–1984)” table
Ex-ante information set — freeze before outcome
What a 1973 decision maker could support from the surviving record
- Product concept: a removable conversion intended to turn a motorcycle into a tracked snow vehicle; a 1973 advertisement establishes that the product was being marketed, not that it had reached repeatable commercial production. Reproduced 1973 ad, named artifact
- Claimed technical state: Perkins later recalled that the machine worked on both pavement and snow. That statement is useful evidence of what he says the partnership believed, but no independent test report, certification, warranty history, or customer trial was found. [investor-stated · provisional] WBUR, 00:12:35–00:15:43
- Market proposition: the advertisement offered a consumer conversion product from a Mountain View company. Surviving evidence does not establish target-unit volume, gross margin, seasonality assumptions, dealer commitments, service burden, geographic concentration, or whether buyers preferred purpose-built snowmobiles. Reproduced 1973 ad, named artifact
- Macro state at the likely investment date: the precise check date is not public. The OAPEC embargo began in October 1973; therefore it cannot be assumed to have been known if the investment preceded it. Federal Reserve History, “Oil Embargo Begins”
Unknown or disputed at the decision date
| Question | Surviving evidence |
|---|---|
| Who sourced and sponsored the deal? | Unknown. Perkins later narrated it, but narration does not prove sourcing or sole decision authority. |
| Which partner approved, negotiated, or joined the board? | Unknown. No partnership or board record found. |
| Vehicle, check, security, ownership, price | Vehicle and cost appear only in a later secondary table; security, ownership, valuation, and cash-flow timing are unknown. |
| Founder and operating team | Not established in the reviewed primary record. |
| Customer proof and alternatives | Unknown; no orders, pilots, dealer contracts, market study, or competing products were recovered. |
| Product readiness and safety | Perkins's “worked” recollection is not an engineering, manufacturing, or safety control. |
| Follow-on plan, reserves, or stop milestones | Unknown. |
| Dissent and rejected alternatives | Unknown. |
Information that must not leak into the ex-ante case
The later oil embargo, fuel shortage, reported zero residual value, and Perkins's conclusion that the product lacked venture-scale importance are outcome/postmortem evidence. They cannot be used to make a 1973 decision appear obviously wrong. The exact timing matters: the embargo may have been unforeseeable at the commitment date, while winter seasonality, discretionary demand, distribution concentration, and sensitivity to fuel availability were foreseeable categories of risk even without foreknowledge of the shock. [researcher inference · moderate] Federal Reserve History, “Oil Embargo Begins”
Reconstructed underwriting frame
No historical IC memo survives, so the following is an analyst-created decision frame, not Perkins's documented process.
| Dimension | Evidence available then | Appropriate unresolved test |
|---|---|---|
| Technical | Claimed dual-surface functionality; advertised product | Independent durability, safety, installation-time, and cold-weather tests |
| Demand | A novel recreational use case | Paid orders across more than one snow region; repeat dealer sell-through; cancellation data |
| Unit economics | No surviving evidence | Bill of materials, labor, warranty reserve, shipping, dealer margin, and seasonal working capital |
| Competition/substitution | No surviving comparison | Compare conversion cost/performance with used and new purpose-built snowmobiles |
| Distribution | No surviving evidence | Named dealers, geographic concentration, inventory ownership, and service obligations |
| External risk | Fuel sensitivity and recreational/discretionary demand were relevant risk categories | Scenario test for fuel-price spike, shortage, poor snowfall, and one lost winter season |
| Venture scale | No evidence of a path to a large company | Adjacent-product roadmap or a market large enough to return a concentrated early fund |
Initial thesis reconstruction: researcher inference; insufficient public record. A plausible thesis was that a lower-cost, adaptable conversion could open snow recreation to motorcycle owners without buying a dedicated vehicle. The record does not establish that Perkins or the partnership actually wrote, quantified, or relied on that thesis.
Kill criteria reconstruction: researcher inference; insufficient public record. Failure of independent reliability/safety tests; weak paid demand after a full winter; dealer concentration; warranty/service costs erasing gross margin; inability to finance seasonal inventory; or no credible venture-scale adjacency should have stopped or staged capital. None is documented as the historical K&P criterion.
Terms and milestones reconstruction: researcher inference; insufficient public record. Given unproven manufacturing and seasonal demand, an analyst might stage capital against third-party tests, paid pre-orders, dealer sell-through, and a capped inventory build, with protective board information rights. The actual security, ownership, board rights, tranches, reserves, and follow-ons are unknown.
Macro shock and postmortem evidence
- By December 1973 the Minneapolis Fed reported that recreational-vehicle activity was declining and fuel shortages were affecting employment; this independently supports a severe sector shock. Minneapolis Fed, 1973-12-12 conditions report
- The embargo ran from October 1973 into March 1974, and the oil-price shock was large. This supports a hostile demand and distribution environment but does not prove the company's exact sales path. Federal Reserve History, “Oil Embargo Begins” and “The Embargo Ends”
- Perkins later said gasoline restrictions disrupted the product and described Snow Job as a failure. Searches found allocation and fuel-shortage evidence, but not a federal rule specifically making gasoline sales for this product or “sports vehicles” illegal. His sharper legal formulation therefore remains uncorroborated and may compress allocation rules, service-station practices, and collapsing discretionary demand into one story. [investor-stated · provisional] WBUR, 00:12:35–00:15:43
- Perkins also said that even success would not have created a company of Genentech-like importance. That is a later portfolio-scale judgment, not demonstrated ex-ante underwriting. [investor-stated · moderate] WBUR, 00:12:35–00:15:43
- The secondary portfolio table reports $229,200 of cost and zero value at June 30, 1984. It does not show write-off date, follow-ons, recoveries, fees, or cash flows, so it cannot support a precise loss multiple. [researcher inference · provisional] Nebraska report, named portfolio table
Most defensible lesson: researcher inference; moderate. Technical functionality does not remove distribution, cyclicality, regulation, fuel availability, working-capital, or market-size risk. The stronger claim—that K&P learned a documented rule and changed its process because of Snow Job—lacks a contemporaneous memo or repeated before/after evidence.
Contradictions, counterevidence, and retrieval gaps
- The product's identity and 1973 marketing are independently visible; actual technical reliability, safety, manufacturing yield, sales, and customer satisfaction are not.
- Perkins's “gasoline could not legally be sold” formulation is stronger than the official evidence recovered. Federal Reserve sources establish an embargo, allocation pressure, and recreational-vehicle contraction—not a product-specific nationwide ban.
- The reported cost and zero value come from a 2014 secondary reconstruction, not fund books. No realized cash-flow schedule supports a return calculation.
- Search routes attempted: company/product aliases (
Snow Job,SnowJob,Advanced Recreation Equipment), 1973 magazine-ad reproductions, K&P Fund I tables, documentary guides, CHM and WBUR direct voice, and federal 1973 fuel-allocation/recreational-vehicle records. No public investment memo, cap table, board minutes, audited fund schedule, company financials, shutdown notice, product-test report, or legal text matching Perkins's specific ban claim was found.
Reader decision lab — answer before revealing the historical outcome
The prompts are analyst-created. Use only the frozen 1973 record above; do not use the embargo, write-off, or later K&P winners.
- Would you invest then? Choose invest, staged invest, or pass. State whether an advertised and reportedly functioning conversion is enough to fund manufacturing before paid multi-region demand exists.
- What would kill the deal? Name measurable thresholds for reliability, safety, dealer sell-through, gross margin after warranty/service, seasonal inventory exposure, and addressable market.
- What ownership/terms fit the risk? Specify a security and governance package consistent with hardware, inventory, seasonality, and an unproven consumer channel; do not assume the historical terms are known.
- What evidence unlocks the next check? Define the minimum independent product test, paid-order cohort, dealer data, unit-economics proof, and stress scenario required before releasing another tranche.
Actual decision and sourced outcome — reveal after the exercise
Kleiner & Perkins invested in Advanced Recreation Equipment/Snow Job according to Perkins's repeated retrospective accounts; the precise partner decision, terms, ownership, and follow-on history remain unknown. He described the investment as a failure after the 1973–74 fuel shock impaired the recreational-vehicle market. [investor-stated · moderate] CHM panel, 00:51:23–00:51:49 WBUR, 00:12:35–00:15:43 The later secondary table reports zero value at June 30, 1984, but is not sufficient to calculate a realized multiple. [researcher inference · provisional] Nebraska report, “KP First Fund (1972–1984)” table
[d] Tandem Computers — staged formation under technical and financing uncertainty
- Decision dates: mid-1974 seed/recruiting authorization; 1975 proof-dependent first institutional financing and December 1975 syndication; 1977 public financing; 1997 merger outcome.
- Core sources: December 8, 1975 BusinessWeek article preserved in the Computer History Museum press archive, article pp. 64–65 (decision-near independent reporting; OCR is imperfect); Tom Perkins Bancroft/NVCA oral history, printed pp. 38–42 (retrospective participant; Bancroft publication rights require paraphrase); Brook Byers oral history, printed pp. 9–10 (retrospective participant; Bancroft publication rights require paraphrase); Pitch Johnson oral history, printed pp. 38–40 (retrospective co-investor); Tandem 1979 annual report, stock-price section (issuer record of the 1977 offering); Tandem 1991 proxy, election and beneficial-ownership sections, printed proxy pp. 2–5 (issuer record); Compaq 1999 Form 10-K, note 2 (later acquirer filing).
- Access: full public PDF/HTML routes inspected 2026-08-23. No direct transcript quotation is reproduced. The 1974 business plan, original term sheets, cap tables, board minutes, customer contract, and Fund I ledger were not found publicly.
- Reliability: A for issuer/acquirer dates and formal roles; B for the decision-near press account and independently attributed participant recollections; C for unsupported causal or return claims.
Case boundary
This is a reconstruction of three different decisions, not a single hindsight-compressed “Tandem bet”: pay to make the concept testable; finance after a technical design existed; then finance manufacturing and market entry. The decision-near article was published while the Model 16 remained an engineering prototype and before customer acceptance, so it is the principal ex-ante control. [researcher inference · strong] BusinessWeek, article pp. 64–65
Perkins was sponsor, plan collaborator, initial capital advocate, recruiter, and chair. James “Jimmy” Treybig supplied the market concept and operated the company; Michael Green and James Katzman resolved the software/hardware architecture; Jack Loustaunou supplied finance; Eugene Kleiner approved partnership exposure; Brook Byers did outside diligence; Pitch Johnson and others supplied independent capital and board validation. “Perkins founded Tandem” without those boundaries is prohibited compression. [researcher inference · strong] Perkins, printed pp. 38–42 Byers, printed pp. 9–10 Johnson, printed pp. 38–40
Decision 1 — mid-1974: buy the design proof, not the finished company
Frozen information set
| State at the decision date | Reconstruction |
|---|---|
| Known | Treybig had identified continuous online transaction processing as a customer problem and had relevant HP minicomputer marketing experience. A complete fault-tolerant product did not exist. [contemporaneous record · strong] BusinessWeek, article p. 64 |
| Known | Green had worked on HP time-sharing software, Katzman on HP/Amdahl computer architecture, and Loustaunou in HP finance/cost accounting. The differentiated product therefore still depended on a multi-function team rather than one investor's technical insight. [contemporaneous record · strong] BusinessWeek, article pp. 64–65 |
| Believed | A modular multiprocessor with redundant communication and peripherals could keep transaction systems operating through component failures at a tolerable premium to ordinary minicomputers. This was a design thesis, not yet customer-validated fact. [contemporaneous record · moderate] BusinessWeek, article pp. 64–65 |
| Unknown | Whether hardware and Guardian-like software could coordinate failure recovery; whether a production unit could meet the reliability claim; whether banks would entrust critical operations to a startup; manufacturing and service cost; sales-cycle length; and warranty exposure. [researcher inference · strong] BusinessWeek, article pp. 64–65 |
| Disputed | Later accounts variously call Perkins/K&P the founder and Treybig the founder. The safer decision-date description is incubation inside K&P around Treybig's concept, with Green/Katzman producing the technical design. [researcher inference · strong] Perkins, printed pp. 38–40 BusinessWeek, article p. 64 |
Alternatives and context
- Do nothing or pursue other K&P opportunities; license the concept to an incumbent; fund a conventional paired-primary/backup system; or pay only for a small architecture/recruiting study before company-scale capital. Only the last route is documented as chosen. [researcher inference · moderate] Perkins, printed pp. 38–40
- The contemporaneous market had expensive custom backup systems, serious downtime costs, and no proof that a standard modular product could displace incumbent or user-built solutions. The article also described venture capital as scarce in late 1975. [contemporaneous record · strong] BusinessWeek, article pp. 64–65
- Perkins later recalls incumbent competition, product complexity, weak public markets, and LP skepticism as objections. These are plausible but retrospective; no rejection letters or LP minutes were found. [retrospective witness · moderate] Perkins, printed pp. 39–41
Risk, kill criteria, and first terms
- The implicit first kill criterion was architectural: if Green and Katzman could not specify a workable hardware/software combination that removed single-point failures, do not fund the full build. [researcher inference · strong] BusinessWeek, article p. 64
- The decision-near article reports $50,000 from Kleiner and Perkins in mid-1974 to recruit Green and Katzman, followed by $1 million after the design work impressed them. That sequence supports staging, but does not disclose security, pre/post-money valuation, liquidation preference, board vetoes, founder vesting, or exact ownership. [contemporaneous record · strong] BusinessWeek, article p. 64
- A later secondary history reports 40% ownership for the $1 million round, but no primary term sheet or cap table was found; that number is not promoted to fact here. [researcher inference · insufficient public record] BusinessWeek provides the amount but not ownership
Decision 2 — 1975: finance prototype completion, manufacturing, and market entry
Frozen information set
| State at the decision date | Reconstruction |
|---|---|
| Known | By December 1975 Tandem had an engineering prototype, a named technical/finance team, and a product announcement; it still had little to show customers beyond the prototype and team. [contemporaneous record · strong] BusinessWeek, article p. 64 |
| Known | A second financing brought $2 million from E.M. Warburg Pincus, Data Science Ventures, and K&P, whose reported additional contribution was $500,000. [contemporaneous record · strong] BusinessWeek, article p. 64 |
| Believed | The Tandem 16 could ship in 1976, approach continuous availability, and offer a lower-cost alternative to custom redundant installations. Management's very high reliability claims were forecasts, not verified field data. [contemporaneous record · moderate] BusinessWeek, article pp. 64–65 |
| Unknown | Production yield, field failure rates, customer willingness to rely on the startup, service burden, installed price, gross margin, and whether a first reference customer would convert into repeatable demand. [researcher inference · strong] BusinessWeek, article pp. 64–65 |
| Disputed | Byers says he tested technical viability and sought Asset Management participation; Johnson says Byers did most of the outside work; Perkins emphasizes his own plan, recruiting, and financing role. These accounts are complementary on activity but do not allocate causal percentages. [retrospective witness · strong] Byers, printed pp. 9–10 Johnson, printed pp. 38–40 |
Alternatives, dissent, and financing fit
- The live alternatives were to stop after the engineering prototype; delay until a customer prepaid; rely only on K&P; or syndicate to add capital, independent diligence, and board capacity. The historical record supports the syndicate choice. [researcher inference · strong] BusinessWeek, article p. 64 Johnson, printed pp. 38–40
- Johnson recalls that Perkins explicitly wanted outside validation and that Byers inspected the team, market, and technology. That is useful dissent-seeking evidence, although Johnson and Byers were eventual investors rather than disinterested observers. [retrospective witness · moderate] Johnson, printed pp. 38–40
- Kill criteria now should have included failure to produce a production-ready unit, inability to win a reference customer on bounded commercial exposure, or evidence that redundancy cost/service needs erased customer value. Only the production/customer sequence is visible; no board-approved kill memo survives publicly. [researcher inference · moderate] BusinessWeek, article p. 65 Perkins, printed pp. 40–42
Milestones, follow-ons, and governance
- Design team recruited → workable architecture → engineering prototype → production unit/manufacturing capability → first bounded-risk customer acceptance → repeat sales. That is the defensible milestone chain; exact dates and acceptance terms beyond the December 1975 article remain unresolved. [researcher inference · strong] BusinessWeek, article pp. 64–65 Perkins, printed pp. 40–42
- Perkins later reports spending about 60% of his time on recruiting, financing, and strategy. Treat this as self-attributed workload, not a causal allocation of Tandem's outcome. [investor-stated · moderate] Perkins, printed pp. 39–42
- The 1991 issuer proxy establishes that Perkins and Treybig had served as directors since 1974 and Perkins as chair since 1974; Pitch Johnson had served since 1975. It cannot prove what each did in a specific early board meeting. [contemporaneous record · strong] Tandem 1991 proxy, printed pp. 2–5
- The same proxy reports 1990 beneficial holdings, not initial ownership or realized investment returns. Do not reverse-engineer Fund I economics from that later table. [researcher inference · strong] Tandem 1991 proxy, beneficial-ownership table
Decision 3 — 1977 public financing and later stop/exit
- Tandem's issuer record says its common stock was offered publicly on December 14, 1977 at $11.50 per share. Public evidence reviewed here does not reconstruct the S-1, selling holders, lockups, or K&P distribution, so the listing is an outcome marker rather than a complete underwriting decision lab. [contemporaneous record · strong] Tandem 1979 annual report, stock-price section
- Compaq's filing records that it completed a stock-for-stock merger with Tandem on August 29, 1997, issued 126 million Compaq shares at a 1.05-for-one exchange ratio, and accounted for the transaction as a pooling of interests. This is company outcome, not evidence of Perkins's personal proceeds. [contemporaneous record · strong] Compaq 1999 Form 10-K, note 2
- No public ledger establishes initial K&P ownership, dilution, distributions, carry, or net LP cash flows. Any “Fund I return” number derived from deal value is therefore incomplete. [researcher inference · strong] Tandem 1991 proxy Compaq 1999 Form 10-K
What the case can and cannot teach
- The documented mechanism is conditional staging: pay first for team/architecture evidence, then for prototype/production, then seek market proof. It does not show that every deep-technology uncertainty can be isolated cheaply. [researcher inference · strong] BusinessWeek, article pp. 64–65
- Outside syndication added independent diligence and governance, but it also diluted K&P's sole-control narrative. [researcher inference · moderate] Byers, printed pp. 9–10 Johnson, printed pp. 38–40
- The case cannot establish person-level alpha, initial ownership, realized proceeds, or a general base rate. Those require term sheets, cap tables, fund ledgers, and comparable failed/ordinary cases. [researcher inference · strong] Tandem 1991 proxy
No-hindsight reader exercise
Stop here. Use only the frozen 1974–75 information above; do not use the IPO, 1990 holdings, or Compaq outcome.
- Would you invest then?
- What would kill the deal?
- What ownership/terms fit the risk?
- What evidence unlocks the next check?
Actual decision and outcome — reveal after the exercise
K&P reportedly funded a $50,000 recruiting/design step in mid-1974, then $1 million after Green and Katzman produced a persuasive architecture; in late 1975 K&P added a reported $500,000 in a $2 million syndicate while the Model 16 remained an engineering prototype. [contemporaneous record · strong] BusinessWeek, article p. 64
Tandem converted that sequence into a production business, offered common stock publicly on December 14, 1977, and later merged into Compaq in August 1997. No reviewed public record converts those corporate outcomes into audited Perkins, K&P, or LP returns. [contemporaneous record · strong] Tandem 1979 annual report, stock-price section Compaq 1999 Form 10-K, note 2
[t] Early Bay Area Venture Capitalists: Thomas J. Perkins
- URL: https://digicoll.lib.berkeley.edu/record/218580/files/perkins_tom.pdf | Recorded: 2009-09-18 and 2009-09-23 | Published: 2010 | Format: videotaped oral history with interview-edited transcript | Reliability: A for Perkins's attributed statements; B for retrospective chronology and self-attribution; C for unsupported industry-wide causation or performance claims | Access: full 70-page transcript; quotation restricted by Bancroft publication-rights notice | Accessed: 2026-08-23 | Locator scheme: printed page plus audio-file timestamp; PDF page is printed page + 15
Takeaways
- Perkins describes a distinct operating model: enter before a complete management team exists, work beside the entrepreneur, and finance the smallest experiment that can remove the decisive risk before committing full capital. [investor-stated · strong] printed pp. 2–3, 01-00:04:48–08:49
- The model begins with opportunity scale rather than low apparent risk. He argues that every venture contains risk; an opportunity without a meaningful barrier attracts competitors and cannot sustain scale. [investor-stated · strong] printed pp. 60–61, 04-00:12:35–17:05
- Fund I was reported as $8 million of commitments with only $4 million invested and the balance returned. Perkins says seven or eight non-outlier deals were only barely profitable, making Tandem and Genentech the remembered economic engines. This establishes concentration, not an audited return decomposition. [investor-stated · moderate] printed p. 33, 02-00:46:19–48:15 printed p. 38, 03-00:01:18–03:16
- His strongest portable rule is conditional staging: identify whether the blocker is technical, market, people, patent, legal, production, or science risk; pay for a decisive test; increase capital only after proof. [investor-stated · strong] printed pp. 60–61, 04-00:14:04–17:05
- His strongest counterexample is Apple. Perkins says K&P rejected it without meeting Steve Jobs because he saw no technical risk and misread marketing as unstageable. The miss shows that the doctrine can discard exceptional distribution or category-creation skill. [investor-stated · strong] printed p. 61, 04-00:17:05–18:06
- He places the idea ahead of the person, yet much of his board work concerns recruiting missing functions, managing founder succession, and mediating high-ego conflicts. The record supports idea-first entry plus people-intensive execution, not a people-irrelevant thesis. [researcher inference · strong] printed pp. 62–65, 04-00:18:06–30:25
- Perkins treats a VC's non-cash product as deep crisis involvement. He also states that lead-investor follow-on power creates effective control even at low ownership, which is useful decision-rights evidence and a warning about founder dependency. [investor-stated · strong] printed pp. 62–64, 04-00:20:35–25:27
Quotes
- No transcript quotation is reproduced because Bancroft reserves publication rights. This note uses page- and timestamp-stable paraphrase.
Biography and decision evidence
- A White Plains physics teacher pushed Perkins toward college; scholarship economics determined MIT over Harvard, and he shifted from physics to electronic engineering after benchmarking himself against stronger physicists. [investor-stated · moderate] printed pp. 1–2, 01-00:00:21–04:24
- Georges Doriot's Harvard course supplied an early model of technically informed venture formation, while Hewlett and Packard supplied the operator apprenticeship and institutional example. The causal weight is Perkins's retrospective attribution. [investor-stated · moderate] printed pp. 5–9
- University Laboratories was the bridge from employee to principal: Perkins says he conceived the laser, financed and supervised the company while at HP with Packard's permission, and sold it into Spectra-Physics. [investor-stated · moderate] printed pp. 18–21
- Perkins and Eugene Kleiner met through Sandy Robertson, combined Hillman and Rockefeller-linked commitments, and closed Fund I at $8 million after a short but difficult raise. [investor-stated · moderate] printed pp. 31–33, 02-00:40:49–48:15
- He says the partnership's plan defined its identity as fellow entrepreneurs who would be deeply involved, rather than passive stock pickers. The original artifact was not located during the 2009 interview. [investor-stated · provisional] printed pp. 33–34, 02-00:48:15–50:05
Deal reasoning and decision evidence
Tandem Computers
- Jimmy Treybig originated the fault-tolerant-computer concept; Perkins and Treybig developed it, while James Katzman and Michael Green tested hardware/software feasibility. Perkins wrote the plan with Treybig and persuaded Kleiner to let the partnership finance nearly the entire initial round after outside investors declined. [documented behavior · strong] printed pp. 38–40, 03-00:01:18–09:26
- The ex-ante objections were incumbent competition, technical complexity, a poor public market, and LP skepticism about minicomputers. After the technical proof, the first customer accepted delivery risk because payment exposure was limited; sales then accelerated. [retrospective witness · moderate] printed pp. 40–42, 03-00:09:26–16:11
- Perkins reports spending roughly 60% of his time on recruiting, financing, and strategy, but chair/sponsor credit must remain distinct from Treybig's invention and operating leadership, Loustaunou's role, Kleiner's approval, Johnson's small co-investment, and the technical validators. [researcher inference · strong] printed pp. 39–42
Genentech
- Swanson recognized the recombinant-DNA opportunity after a Donald Glaser lunch, pursued Boyer, and with Boyer produced the venture concept. Perkins initially failed to grasp Glaser's point, then helped redirect the plan from a $2–2.5 million integrated build to subcontracted proof of principle. [retrospective witness · strong] printed pp. 45–48, 03-00:24:05–35:47
- Somatostatin compressed the scientific question into a smaller gene; Perkins recalled $100,000, $150,000, and about $250,000 total at different points, making the precise first-tranche figure unresolved. The reliable insight is staged proof, not a single clean number. [investor-stated · moderate] printed pp. 47–49, 03-00:34:48–38:19
- The three-person board was Boyer, Swanson, and Perkins. Perkins describes his role as chair, capital raiser, and strategy participant; Swanson ran the company and learned the science; Boyer and the scientific network supplied core scientific credibility. [retrospective witness · strong] printed pp. 49–51
- Perkins pushed the 1980 IPO to establish category leadership before Cetus. The strategy sought publicity, recruiting advantage, valuation anchoring, and future financing capacity, but post-IPO option pricing created a new hiring disadvantage. [investor-stated · strong] printed pp. 51–55, 03-00:43:40–55:09
- Junior common stock and R&D partnerships were financing responses to talent and income-statement constraints. Perkins credits himself heavily but names lawyers, the SEC, and CFO Fred Middleton; he also acknowledges conflicts in personal participation. [investor-stated · moderate] printed pp. 53–57, 03-00:51:43–04-00:04:12
Hybritech, Apple, Segway, and Acuson
- Hybritech was Brook Byers's deal: Byers followed Ivor Royston's monoclonal-antibody insight, persuaded Perkins to fund it, and chaired it; Ted Greene became the operator and Pitch Johnson co-invested. Perkins's “we started it” narration must not erase Byers's lead. [retrospective witness · strong] printed pp. 58–59, 04-00:05:52–08:49
- Perkins uses Segway as a later-firm violation of his principle because market risk could not be cheaply removed; he expressly says he was no longer involved. It is a doctrine counterexample, not his deal. [investor-stated · strong] printed p. 61, 04-00:15:30–17:05
- Acuson shows how technical adjacency substituted for a full-plan read: Perkins ignored most of Sam Maslak's long plan, used optics knowledge to interrogate acoustical diffraction, judged the mechanism differentiated, then backed an untested manager and helped recruit the team. [investor-stated · strong] printed pp. 69–70, 04-00:41:50–44:14
Philosophy signals
- sourcing / picking: seek a large opportunity with a specific barrier; an attractive market without a barrier invites herd competition. [investor-stated · strong] printed pp. 60–61
- staging / reserves: spend the minimum needed to test the decisive risk, then increase capital after proof; abandon cases where no cheap decisive test exists. [investor-stated · strong] printed p. 61
- people / ideas: begin with the opportunity, accept incomplete or inexperienced teams, and add management after technical or market proof. The evidence does not support ignoring character or execution. [researcher inference · strong] printed pp. 62–64
- ownership / control: Perkins preferred lead and effective control, but paired it rhetorically with long-duration support and founder visibility. [investor-stated · strong] printed pp. 62–63
- boards / crises: create enough trust that management reports trouble immediately; intervene early, mediate competing views in the same room, and keep the investor's ego behind the entrepreneur. [investor-stated · strong] printed pp. 63–65
- LP alignment: Perkins describes bans on opportunity diversion, preferential personal purchases, unaudited reporting, and cherry-picked co-investment. The industry-standard-setting claim lacks an adoption denominator. [investor-stated · provisional] printed pp. 34–35
- portfolio construction: the public evidence shows a small first fund dominated by two outliers; it does not show a repeatable hit rate or net LP return. [researcher inference · strong] printed pp. 33 and 38
Attribution and decision-rights map
- Eugene Kleiner: co-founder, co-GP, Fund I capital-commitment partner, technical/operating peer, and necessary approval on Tandem and Genentech; his quieter style is contrasted with Perkins's pushing. [researcher inference · strong] printed pp. 31–40 and 59–60
- Jimmy Treybig / Jack Loustaunou: Tandem concept/operating founders; Treybig supplied the core idea, Loustaunou the finance role, and both left K&P to operate it. [researcher inference · strong] printed pp. 38–42
- Robert Swanson / Herbert Boyer / Stanley Cohen: Swanson recognized and drove commercialization, Boyer supplied scientific partnership, Cohen and Boyer created the underlying recombinant-DNA work, and Swanson ran Genentech. [researcher inference · strong] printed pp. 45–51
- City of Hope / UCSF / Caltech: contractors made the staged experiment executable; their rights and later disputes are part of the deal, not background decoration. [researcher inference · strong] printed pp. 47–49
- Brook Byers / Ivor Royston / Ted Greene / Pitch Johnson: distinct sourcing, scientific, operating, and co-investment roles in Hybritech. [researcher inference · strong] printed pp. 58–59
- Fred Middleton: Genentech CFO and co-claimant to the R&D-partnership structure; Perkins's self-credit is explicitly disputed in friendly form. [researcher inference · moderate] printed pp. 56–57
Contradictions, counterevidence, and access limits
- The oral history is interviewee-reviewed but remains partisan retrospective evidence; Perkins frequently says he cannot remember dates, round amounts, company names, or exact investment figures. [researcher inference · strong] printed pp. 32–65
- The 2009 claim that the original business plan was missing is later updated by the 2011 CHM interview, where Perkins says Hillman produced a copy. The artifact itself still must be inspected before treating recollection as text. [researcher inference · strong] printed p. 33
- “Ideas over people” conflicts with his own statement that Hybritech's main risk was people and with years spent mediating Swanson/Raab. The more coherent rule is idea-first entry with intensive people work. [researcher inference · strong] printed pp. 61–65
- His anti-cherry-picking prescription sits uneasily beside personal participation in Genentech R&D partnerships; he concedes a conflict but says the structure was disclosed and approved. The transcript does not expose terms sufficient to resolve alignment. [researcher inference · strong] printed pp. 56–57
- Strong lead-investor control can improve follow-on credibility or suppress founder agency. The transcript gives Perkins's view, not a founder/employee denominator across ordinary companies. [researcher inference · moderate] printed pp. 62–63
- The Apple miss falsifies any claim that all large opportunities can be reduced to a cheap technical test; category creation and exceptional marketing can be the investable edge. [researcher inference · strong] printed p. 61
- Reported numbers—Fund I size/deployment, ownership, $3 billion raised for Genentech, and later carry—are participant recollections, not audited person-level economics. [researcher inference · strong] printed pp. 33 and 49–57
Forecasts and prediction record
- Perkins predicts that an apparently low-risk opportunity conceals risk and that no-barrier markets attract competition. This is a general decision rule, not a scored forecast. [investor-stated · moderate] printed pp. 60–61
- He retrospectively argues that Genentech's IPO would establish leadership and attract talent/capital; the transcript records the favorable outcome but no decision-date probability or alternative forecast. [retrospective witness · moderate] printed pp. 51–53
- He describes Apple as a false negative caused by an incorrect view of marketing risk. That is the clearest falsified judgment in the interview. [investor-stated · strong] printed p. 61
- His claim that Genentech would be unique in combining science and money is retrospective admiration, not a falsifiable forward call. [researcher inference · strong] printed pp. 57–58
Open questions
- What does the recovered 1972 business plan say verbatim, and which partnership agreements implemented it?
- What is the reconciled Fund I company denominator, gross/net cash-flow schedule, ownership ledger, and attribution by partner?
- Which Tandem decision-date documents distinguish Perkins's, Treybig's, Loustaunou's, Kleiner's, Johnson's, Katzman's, and Green's contributions?
- Which Genentech contracts and cap tables resolve the $100,000/$150,000/$250,000 staging and ownership recollections?
- How often did the cheap decisive-risk test work, fail, or exclude later winners across ordinary and written-off deals?
- Which founder, executive, employee, and LP accounts support or challenge the strong-control and crisis-help narrative?
[t] CBS 60 Minutes — “The Captain of Capitalism”
- URL: https://www.cbsnews.com/news/tom-perkins-the-captain-of-capitalism/ | Published: page 2007-11-01; broadcast 2007-11-04; page updated 2008-05-30 | Format: 13:26 broadcast video plus CBS-edited narrative transcript/report | Reliability: A for CBS custody and directly attributed interview speech; B for edited sequencing and reporter summaries; C for unverified deal economics | Access: full HTML report and hosted video duration through browser retrieval; a later CLI spot-check returned HTTP 406; no caption file or shot-level transcript processed | Accessed: 2026-08-23 | Locator scheme: paragraph openings in the HTML report; video duration is metadata, not a timestamp claim
Takeaways
- The broadcast uses the Maltese Falcon to establish an engineer-builder identity: Perkins says he designed the control software, describes the carbon-fiber masts, and admits that scale also served ego. This is biographical and temperament evidence, not evidence of venture returns. [investor-stated · moderate] paragraphs beginning “Perkins designed the software” and “Why did it have to be”
- Perkins describes Genentech's early objective as producing human insulin through recombinant DNA and says high-risk, high-technology venture returns encouraged broader Silicon Valley participation. CBS reports a $250,000 KPCB investment becoming $200 million, but the page does not provide tranches, ownership, distributions, or audited fund cash flows. [investor-stated · provisional] paragraphs beginning “He and his partners launched” through “And it took somebody”
- He gives a rough hit-rate claim of two home runs in ten projects. The undefined denominator—project, company, check, fund, or personal lead—makes this useful only as self-described power-law intuition. [investor-stated · provisional] paragraph beginning “And it took somebody like Perkins”
- The HP segment is adversarial rather than hagiographic. Stahl asks about gender and control, Dunn alleges a reputation campaign, and Perkins admits anger, a behind-the-scenes effort to remove Dunn, and possible error in how he resigned. [contemporaneous record · moderate] paragraphs beginning “Is it possible,” “After he resigned,” and “Asked if he misses”
- Perkins rejects the claim that he engineered Carly Fiorina's firing and disputes some reported hostile language toward Dunn. These are party positions in an edited interview, not findings about the HP board sequence. [investor-stated · moderate] paragraphs beginning “He denies he engineered” and “But Perkins says”
- His account of childhood scarcity connects money, display, and competition in his self-conception. Because this is retrospective self-interpretation prompted by a television profile, it should not be treated as a proven cause of investment behavior. [investor-stated · provisional] paragraphs beginning “Analyze that for us” and “An only child”
- CBS says the yacht was put up for sale after the construction challenge lost its novelty. The pattern—enjoying build-stage technical difficulty more than steady-state ownership—is a plausible temperament hypothesis, not a verified portfolio rule. [researcher inference · provisional] closing paragraph beginning “But there's no protection against boredom”
Quotes
- “The idea was to trick nature into letting us make something that didn't exist in nature, in particular, human insulin.” [investor-stated · strong] Genentech passage
- “Maybe I made a mistake in resigning as I did.” [investor-stated · strong] HP-regret passage
Biography and decision evidence
- The page identifies Perkins as partner emeritus thirty-five years after K&P's founding. Its opening list of AOL, Amazon, Netscape, and Google is firm history and must not be assigned to Perkins personally. [researcher inference · strong] opening paragraphs
- The Genentech account supports Perkins's chair/investor involvement and technical framing, but CBS's phrase that he and partners “launched” the company compresses Robert Swanson, Herbert Boyer, scientists, operators, and financing participants. [researcher inference · strong] Genentech paragraphs
- The HP interview supplies decision-near remorse and temperament evidence: Perkins acknowledges intense emotion and later says he may have erred in the manner of resignation. It does not establish the full May-to-September investigative chronology. [investor-stated · moderate] paragraphs beginning “Just slammed your briefcase” and “Asked if he misses”
- Perkins was then a News Corp director; the profile records board status only and cannot establish diligence quality, independence, or later board conduct. [contemporaneous record · strong] paragraph beginning “He has moved on”
Philosophy signals
- risk / portfolio: expect a small number of exceptional outcomes, but demand a defined denominator before translating a television hit-rate remark into portfolio construction. [researcher inference · strong] two-in-ten passage
- picking: technical fluency can help frame scientific feasibility, but CBS does not show a decision-date experiment, alternatives, or reject log. [researcher inference · moderate] Genentech passage
- boards / people: anger and costly dissent make conviction observable while increasing the need for process, minutes, recusal, and independent review. [researcher inference · strong] HP exchange
- temperament: the attraction to difficult construction and superlative scale may supply energy for frontier projects and also status-driven overreach. [researcher inference · provisional] yacht sections
Contradictions, counterevidence, and access limits
- This is a tightly edited television profile organized around spectacle and conflict. The HTML blends narration, interview fragments, and adversarial responses; it is not a complete chronological transcript. [researcher inference · strong] CBS page and 13:26 video metadata
- The Genentech reported: $250,000-to-$200 million claim is not reconciled to seed-versus-total investment, securities, ownership, dates, distributions, or fund-level economics. It cannot be treated as a verified 800x cash-on-cash return. [researcher inference · strong] paragraph reporting the 800-fold figure
- The “two out of ten” statement has no vehicle, vintage, loss rate, or coverage definition. It cannot establish a stable personal hit rate. [researcher inference · strong] hit-rate paragraph
- CBS's statement that Perkins “triggered” exposure of HP spying is a compressed outcome narrative. The interview itself shows later regret and dispute; issuer and regulatory records must control exact knowledge and disclosure chronology. [researcher inference · strong] opening HP paragraph and resignation exchange
- Self-awareness about ego does not neutralize governance or allocation consequences. Conversely, visible luxury is not evidence that a particular investment decision was unsound. [researcher inference · strong] yacht and self-protection passages
Open questions
- Locate a captioned or production transcript with timestamps and identify material interview segments omitted from the HTML edit.
- Reconcile Genentech check amounts, dates, ownership, dilution, distributions, and vehicle economics from contemporaneous records.
- Compare the HP interview chronology with board minutes, HP filings, and the SEC order rather than backfilling later privacy claims into resignation day.
- Define the population behind “two out of ten” and test it against ordinary, failed, and passed investments.
[t] Charlie Rose — Thomas Perkins
- URL: https://charlierose.com/videos/11352 | Aired: 2007-12-05 | Format: television interview; official speaker-labeled, millisecond-timestamped HTML transcript at
/video/transcripts/11352| Reliability: A for custody/transcript speaker labels, B for retrospective claims | Access: full transcript and hosted player route | Accessed: 2026-08-23 | Locator scheme:MM:SSfrom official transcript
Key claims and reasoning
- At 09:52–10:40, Perkins estimates $20–40 billion of available venture capital, says a universal 100x aspiration cannot be met collectively, and expects excess capital to inflate prices while some managers still perform. The forecast lacks a fund/vintage/net-return denominator but is a direct capital-supply thesis. [investor-stated · moderate] official transcript
- At 11:26–11:44, he credits John Doerr's earlier Internet optimism, a useful partner-attribution control against assigning later KPCB insight to Perkins. [investor-stated · strong] official transcript
- At 11:57–14:16, he treats university/corporate research intensity and a global problem as early green-market signals, then predicts biotechnology remains near a “ground floor” beyond copying natural genes. The thesis is expansive but supplies no company screen, capital plan, regulatory constraint, horizon, or scored outcome set. [investor-stated · provisional] official transcript
- At 16:53–17:14, he again describes roughly 525 financed companies as KPCB's referral sales force. The changing firm-level count is deal-flow evidence, not Perkins's personal portfolio or a complete outcome denominator. [investor-stated · moderate] official transcript
Biography and governance evidence
- At 19:51–21:25, Perkins says the memoir addresses HP and presents Packard/Hewlett as mentors whose internal new-market process inspired venture practice: idea, risk removal, capital, team, then repeat. This is a retrospective causal narrative, not an HP process audit. [investor-stated · moderate] official transcript
- At 21:42–25:13, he says the leak investigation was hidden from him despite chairing governance; he expected Dunn to approach the leaker privately and seek a promise; he opposed the full-board “trial,” secret ballot, and requested resignation, then left. This direct account matches the narrow private-resolution chronology in the SEC order more closely than later whistleblower compression. [investor-stated · strong] official transcript SEC order ¶¶4–9
- At 25:13–26:08, Perkins says he missed the board but would repeat his action. Costly consistency does not establish that the private process was correct. [investor-stated · moderate] official transcript
Ethics and adverse controls
- At 14:23–16:13, Perkins declares the credit crisis independent from venture capital, says there is no venture-capital scandal, and credits himself/Kleiner with a fair model competitors copied. These are self-attributed field-level claims; search-result counts are not legal denominators, and the missing 1972 plan/adoption/compliance records prevent causal confirmation. [investor-stated · insufficient public record] official transcript 1972-plan note
- His near-term claim that credit-market turmoil was independent from VC is conceptually narrow: venture funds may lack the same leverage, but exits, LP liquidity, valuation, fundraising, and company demand can still transmit shocks. The transcript offers no test. [researcher inference · moderate] official transcript
- Rose's opening assigns Amazon and Google to the firm; it is host framing, not Perkins speech and not person-level deal attribution. [researcher inference · strong] official transcript 00:00–00:22
Forecast ledger
- 09:52–10:40: aggregate $20–40 billion pool cannot achieve a 100x collective goal; excess capital inflates prices. [investor-stated · provisional] No defined vintage, asset boundary, net/gross metric, or horizon. Official transcript
- 11:57–12:30: green technology is brewing into a large potential market. [investor-stated · provisional] No market definition, timing, or investment-return metric. Official transcript
- 12:39–14:16: biotechnology remains underhyped/near a ground floor with an effectively unlimited synthetic future. [investor-stated · provisional] Too broad for binary scoring without decomposed milestones. Official transcript
- 17:25–18:13: China offers a positive long-term opportunity; KPCB is moving cautiously. [investor-stated · provisional] No portfolio, horizon, or metric. Official transcript
Mental-model tags
capital-supply crowding;price discipline;research-network signal;partner attribution;portfolio-network sourcing;operator apprenticeship;private-resolution governance;field-self-attribution;forecast ledger. [researcher inference · moderate] Official transcript
Contradictions, counterevidence, and access limits
- Transcript custody and speaker labels are strong; participant claims remain retrospective. [researcher inference · strong] Official speaker-labeled transcript
- “No venture scandal,” 525 companies, $20–40 billion, 100x goal, and industry-standard-setting are undefined/self-reported. [researcher inference · strong] Official transcript, 09:52–17:14
- The HP account is direct but interested; the SEC order controls the regulatory/disclosure result. [researcher inference · strong] Official transcript, 21:42–26:08 SEC order ¶¶4–9
- The appearance spends substantial time on the yacht and memoir; this note processes the complete investment/governance portion rather than treating personal display as an investment rule. [researcher inference · strong] Official transcript
Open questions
- How did the 2007 capital-supply forecasts score by defined venture vintage and net return?
- Which green/China/biotech opportunities were actually selected, by which partner and vehicle, and with what cash outcome?
- What independent adoption or governing-agreement evidence tests Perkins's industry-standard-setting claim?
[t] Computer History Museum oral history of Thomas J. Perkins
- URL: https://archive.computerhistory.org/resources/access/text/2012/03/102714336-05-01-acc.pdf | Recorded: 2011-07-22 | Published: 2012 | Format: 23-page oral-history transcript, CHM reference X6219.2012 | Reliability: A for Perkins's attributed statements; B for retrospective chronology and self-attribution; C for unsupported market-size, firm-count, and return forecasts | Access: full transcript | Accessed: 2026-08-23 | Locator scheme: CHM PDF page number, which matches the printed “Page N of 23” footer
Takeaways
- Perkins traces venture judgment to three layers: technical curiosity and early commerce, Doriot's business/leadership formation, and operating experience at University Laboratories and HP. This is a self-explanation of edge, not causal proof. [investor-stated · moderate] CHM pp. 2–6
- University Laboratories is the decisive operator bridge in this interview. Perkins says he conceived, funded, supervised, and exited the laser business, then realized the same activity could be repeated as venture capital. [investor-stated · strong] CHM pp. 5–6
- He presents K&P as hands-on, idea-first, and willing to build management around an inexperienced technical founder. The investor's non-cash product is crisis help, recruiting, and long-duration involvement. [investor-stated · strong] CHM pp. 6–7
- The clearest decision algorithm appears on pp. 18–19: locate the decisive technical, people, legal, or market risk; use the smallest tranche to test it; step valuation after success; walk away after failure. [investor-stated · strong] CHM pp. 18–19
- Acuson is the worked ordinary-success case: ignore the bloated plan, interrogate the physical mechanism, fund an inexperienced founder, recruit the management team, and later oppose a product roadmap whose long development gap threatened the existing franchise. [investor-stated · strong] CHM pp. 20–21
- The same model carries a strong-control premise. Perkins tells founders that follow-on signaling gives a first investor effective control even with small ownership. Long support can be valuable; the asymmetry also creates coercion risk. [investor-stated · strong] CHM pp. 19–20
- Perkins acknowledges many failures and lost capital but provides no denominator. The interview's famous-case selection cannot establish person-level hit rate or repeatable alpha. [researcher inference · strong] CHM pp. 18–20
Quotes
- No verbatim excerpt is necessary. The note uses page-stable paraphrase from the full public transcript.
Biography and decision evidence
- Childhood tinkering included a model steam engine, a Tesla coil, television-kit assembly/sales, and a physics teacher who encouraged MIT. The commercial activity was small but establishes a recurring build-and-sell identity. [investor-stated · moderate] CHM pp. 2–3
- MIT was financially enabled by scholarship and academically difficult; Harvard/Doriot introduced broader leadership and industrial-development questions. Perkins later describes Doriot as an important mentor rather than a source of a specific K&P rule. [investor-stated · moderate] CHM pp. 3–5
- HP supplied technical, marketing, division-leadership, budgeting, and internal-entrepreneurship experience. Perkins repeatedly credits Hewlett and Packard, while also emphasizing his own computer-business contribution. [investor-stated · moderate] CHM pp. 8–17
- Packard's return from Washington and lack of an equally attractive next HP role precipitated the move into venture capital; HP later invested in the fund. [investor-stated · moderate] CHM p. 17
- Perkins says the 1972 K&P plan was finally recovered from Henry Hillman for the firm's 40th anniversary. This updates his 2009 Bancroft statement that the plan could not be found, but the public transcript does not reproduce the artifact. [investor-stated · moderate] CHM pp. 17–18
- The recovered plan reportedly committed K&P to hands-on work, ideas, team-building, and earliest-stage formation before it raised capital. Artifact inspection is still required before treating this paraphrase as the plan's text. [investor-stated · provisional] CHM p. 18
Deal reasoning and decision evidence
University Laboratories
- Perkins says the low-cost, adjustment-free laser idea could be developed nights/weekends with Packard's permission; he financed and supervised it, then merged the successful company into Spectra-Physics rather than leave HP. [investor-stated · moderate] CHM pp. 5–6
- The lesson he draws is operator empathy and confidence to intervene. The record does not show competing designs, invested cash, ownership, or counterfactual return. [researcher inference · strong] CHM pp. 5–6
Generic staged-risk method
- Perkins proposes replacing a founder's $5 million initial request with about $500,000, no automatic follow-on commitment, and a higher later valuation if proof succeeds. He argues this can preserve founder ownership while limiting loss. [investor-stated · strong] CHM pp. 18–19
- He expressly says the formula is difficult, sometimes unavailable, and sometimes wrong. That boundary matters: not every uncertainty is cheaply separable or testable. [investor-stated · strong] CHM pp. 18–19
- His long-hold rationale is learning and crisis capacity after IPO, not simply liquidity timing. He also admits 525 claimed K&P companies made indefinite board involvement impossible; “stay long term” is selective, not universal. [investor-stated · moderate] CHM pp. 19–20
Acuson
- Sam Maslak's plan projected large numbers and covered every conventional topic but under-explained the mechanism. Perkins used optical-wave knowledge to probe acoustical diffraction until he understood the differentiating technical claim. [investor-stated · strong] CHM pp. 20–21
- Maslak had never managed anyone. Perkins funded anyway, warned that leadership might change, and says he recruited the key management team; Maslak later proved an exceptional manager. The case falsifies a credentials-first screen but does not prove “idea over people” generally. [retrospective witness · strong] CHM pp. 20–21
- Later Perkins opposed concentrating resources on a five-to-seven-year product leap without nearer product releases. He resigned when Maslak refused incremental integration and retrospectively says the gap forced a sale. The account needs Acuson board/company records to test causality. [investor-stated · moderate] CHM p. 21
Philosophy signals
- formation: prefer operating and entrepreneurial apprenticeship before becoming a venture investor; do not treat venture as an entry-level profession. [investor-stated · strong] CHM p. 23
- picking: start with a differentiated technical or commercial idea, but retain a founder-replacement option if managerial capacity does not scale. [investor-stated · moderate] CHM pp. 5–6 and 20–21
- risk: identify a single decisive uncertainty only when the risk is actually separable; the model fails when proof requires the full build or when uncertainty is reflexive market adoption. [researcher inference · strong] CHM pp. 18–19
- ownership / pricing: stage capital at rising prices so early proof benefits founder ownership, while preserving investor choice on later rounds. [investor-stated · strong] CHM pp. 18–19
- boards / crises: deep involvement is justified by inevitable crises and requires management trust sufficient for immediate bad-news reporting. [investor-stated · strong] CHM pp. 6–7 and 19–20
- control: first-investor follow-on signaling creates practical authority regardless of formal ownership. This is decision-rights realism and a founder-power warning. [researcher inference · strong] CHM p. 19
- product cadence: avoid starving the current franchise for a distant breakthrough; decompose the new architecture into nearer customer-visible increments where possible. [researcher inference · strong] CHM p. 21
Attribution and decision-rights map
- Mr. Wilson / MIT mentors / Georges Doriot: educational access, technical formation, and business framing. These are Perkins's stated influences, not investors in K&P. [researcher inference · strong] CHM pp. 2–5
- Dave Packard / Bill Hewlett / HP teams: permission for University Laboratories, operating apprenticeship, computer/calculator initiatives, and division authority. Perkins's narrative sometimes centers himself despite distributed HP execution. [researcher inference · moderate] CHM pp. 5 and 8–17
- Eugene Kleiner: co-designer of the original model with independent entrepreneurial/Fairchild experience. The plan and firm results are partnership evidence, not Perkins-only evidence. [researcher inference · strong] CHM pp. 17–18
- Henry Hillman: early LP and custodian of the recovered plan copy; the public artifact route remains unresolved. [researcher inference · moderate] CHM p. 17
- Sam Maslak: Acuson founder/technical inventor/CEO; Perkins claims technical diligence, financing, recruiting, board service, and later dissent, not invention or company operation. [researcher inference · strong] CHM pp. 20–21
- later KPCB partners: the reported 525-company figure, Google board service, and later deals belong to the institution and partners; this source cannot assign them to retired Perkins. [researcher inference · strong] CHM pp. 19–20
Contradictions, counterevidence, and access limits
- The transcript is full and direct, but nearly every success mechanism is retrospective and voiced by Perkins; company records and other participants are required for independent causal weight. [researcher inference · strong] full CHM transcript
- The 2011 recovered-plan story supersedes the 2009 “missing plan” recollection, yet no public copy was attached. Treat plan content as recalled until inspected. [researcher inference · strong] CHM pp. 17–18
- He describes confidence as an asset and notes others might call it arrogance. Many failures and lost capital are acknowledged but unnamed; selection bias remains unresolved. [researcher inference · strong] CHM p. 18
- “Idea first” depends on intensive management recruitment and possible founder replacement. Acuson's unexpected managerial success is one favorable case, not a base rate. [researcher inference · strong] CHM pp. 6 and 20–21
- Effective control through follow-on signaling can reduce financing risk or create coercive dependency. The transcript does not include founder or minority-holder perspectives. [researcher inference · moderate] CHM p. 19
- The Acuson roadmap postmortem reveals long-horizon intolerance that could prevent a damaging product gap or underfund a true architectural leap. Actual alternatives and kill criteria are absent. [researcher inference · moderate] CHM p. 21
- The News Corp crisis is mentioned as current learning without details; it cannot support a favorable governance claim from this transcript alone. [researcher inference · strong] CHM pp. 6 and 20
Forecasts and prediction record
- Perkins estimated that US venture capital might approach $1 trillion and argued it could not all compound at 10x–50x; he predicted disappointment resembling the dot-com bust. The market-size premise is explicitly uncertain and unsourced, so only the directional overcapitalization thesis is usable. [investor-stated · provisional] CHM pp. 21–22
- He predicts recurring company crises and uses that expectation to justify long-duration board engagement. It is not scored against a company denominator. [investor-stated · moderate] CHM pp. 6 and 19
- The Acuson call was that an isolated five-to-seven-year development gap would damage the company; the eventual sale is offered as confirmation, but timing, alternatives, and counterfactual performance are not reconstructed. [retrospective witness · moderate] CHM p. 21
- His career advice—successful large-company apprenticeship, then entrepreneurship, then venture—is a normative sequence, not a forecast or demonstrated necessary condition. [investor-stated · moderate] CHM p. 23
Open questions
- Where can the recovered Hillman copy of the 1972 K&P business plan be lawfully inspected?
- What were the exact University Laboratories capital, ownership, operating, and exit economics?
- Which of the acknowledged failures used the staged-risk formula, and why did it fail?
- What is the defensible denominator behind the 525-company claim, and which vehicles/partners are included?
- What do Acuson board records show about the long-gap product decision, alternatives, warnings, and sale causation?
- How did the 2011 overcapitalization forecast score against contemporaneous venture commitments and later returns?
[t] Commonwealth Club — Tom Perkins 2007 appearance route
- URL: https://digitalcollections.hoover.org/objects/4419/thomas-j-perkins-speech | Date: Hoover record 2007-11-14; ABC7 says program recorded 2007-10-28 | Format: 59:27 sound-recording catalog record plus ABC7 event/video metadata | Reliability: A for each institution's displayed metadata; insufficient for speech content | Access: catalog full, media body not retrievable; live pages returned access-control failure in CLI | Accessed: 2026-08-23 | Locator scheme: catalog fields only
Retrieval result
- Hoover identifies “Thomas J. Perkins speech,” program
cc_20071114_sv_perkins, 59:27, Commonwealth Club records, with educational/research-use rights. ABC7 separately labels a Perkins Commonwealth Club program recorded October 28, 2007. The date discrepancy may represent two events, an air/recording distinction, or catalog error; it is unresolved. [contemporaneous record · moderate] Hoover catalog ABC7 metadata - The mapped Commonwealth route and Hoover object returned access-control errors to CLI retrieval; ABC7 describes a full video but exposes no playable transcript/body to the research route. No direct audio URL surfaced, so OpenAI transcription was not possible. [researcher inference · insufficient public record] Hoover catalog ABC7 metadata
Usable evidence
- Metadata supports at least one memoir-era public appearance about Valley Boy, HP, and Perkins's career. It does not support any claimed speech content. [researcher inference · strong] ABC7
Contradictions, counterevidence, and access limits
- Do not merge October 28 and November 14 without a program schedule or audio comparison. [researcher inference · strong] Hoover catalog ABC7 metadata
- The later February 2014 Commonwealth Club weighted-vote remarks are a separate appearance and are processed in the writing/adverse record. [researcher inference · strong] 2014 writing/adverse note
- No media wording is inferred from ABC7's promotional summary. [researcher inference · strong] ABC7 metadata
Open questions
- Can Hoover's reading room or a repaired public player expose the 59:27 recording?
- Was October 28 the recording date and November 14 the broadcast/catalog date, or were there two programs?
- If lawful media appears, transcribe through the OpenAI API and add timestamped claims.
[t] Creative Capital — *Valley Boy*, part 1
- URL: https://creativecapital.wordpress.com/2008/01/17/valley-boy-part-1-of-an-interview-with-vc-pioneer-thomas-j-perkins/ | Published: 2008-01-17; interview conducted November 2007 | Format: full edited written Q&A by Spencer Ante, with a derivative Venture Hacks excerpt | Reliability: B for attributed remarks/editing | Access: full part 1; promised part 2 not located in the mapped route | Accessed: 2026-08-23 | Locator scheme: question headings and opening words
Key claims and reasoning
- Under “worst investment,” Perkins says there were hundreds and repeats the fail-cheap rule: put the risk first and use initial capital to reduce it; one Google can cover many early losses. This is a firm-level power-law claim that assigns Google to “one” return without specifying the partner, vehicle, cost, ownership, cash realization, or loss denominator. [investor-stated · moderate] Q&A
- On bubbles and capital supply, he says KPCB made money in bubbles, excess VC is persistent, and public-market moves should not drive startup timing because technology growth and eventual liquidity matter more. The argument separates entry thesis from near-term exit conditions but can underweight fund duration and capital-market dependence. [investor-stated · moderate] Q&A
- On value/liquidity, Perkins alternates between earnings, market share, and growth rate, then says ground-floor entry makes it “always” acceptable. The last claim is too absolute: early price cannot rescue weak unit economics, dilution, follow-on burden, governance, or no exit. [investor-stated · provisional] Q&A
- On career path, he again advises becoming an entrepreneur rather than a VC and calls money an undifferentiated commodity VCs sell. This makes sourcing, judgment, operating help, governance, reputation, terms, and reserves the differentiation thesis—but does not measure their effect. [investor-stated · moderate] Q&A
Deal, team, and attribution evidence
- Perkins says Rupert Murdoch consulted him before MySpace and an unnamed rejected acquisition; he was enthusiastic because network platforms could scale and the people impressed him. This is a News Corp public-board decision, not a KPCB investment. [investor-stated · moderate] Q&A, MySpace questions
- Succession is attributed to Ray Lane, John Doerr, and Brook Byers as managing GPs; Perkins says future durability depends on bets rather than only pipeline. This protects later-firm decision attribution. [investor-stated · strong] Q&A, succession question
Forecast ledger
- Green technology has enough momentum/technology to matter; trendiness is not a reason to ignore it. [investor-stated · provisional] No market, company, horizon, or return metric. Creative Capital Q&A
- KPCB would raise its thirteenth fund in 2008 and survive another thirty years depending on bets. [investor-stated · provisional] Vehicle identity and outcome need later firm records. Creative Capital Q&A
- China expansion is experimental and US-China relations are viewed positively. [investor-stated · provisional] No decision/outcome ledger. Creative Capital Q&A
Mental-model tags
fail-cheap;power-law portfolio;cycle-independent entry;value-before-liquidity;money-as-commodity;operator apprenticeship;firm-partner attribution;trend-without-chasing. [researcher inference · moderate] Creative Capital Q&A
Contradictions, counterevidence, and access limits
- This note uses the full interviewer-controlled Creative Capital page; no separate Forbes body was processed. [researcher inference · strong] Creative Capital Q&A
- “Hundreds” of worst investments, “one Google,” and “always” acceptable ground-floor entry are rhetoric without denominators. [researcher inference · strong] Creative Capital Q&A, worst-investment and value questions
- Venture Hacks preserves selected aphorisms and is a derivative excerpt, not a second interview family. [researcher inference · strong] Venture Hacks excerpt
- The public part 1 does not contain the promised HP-focused part 2; no claims from it are inferred. [researcher inference · strong] Creative Capital part 1
Open questions
- Where is part 2 and does it preserve decision-near HP detail?
- Which hundreds of losses and which partner/vehicle Google economics support the power-law claim?
- Did KPCB XIII, green, and China forecasts meet defined net-return and attribution tests?
[t] Kleiner Perkins, Venture Capital, and the Chairmanship of Genentech, 1976–1995
- URL: https://digicoll.lib.berkeley.edu/record/54368/files/klienerperkins00perkrich.pdf | Recorded: 2001-10-24, 2001-10-31, and 2001-11-14 | Published: 2002 | Format: three-session oral history with participant introductions, appendices, and interviewee-reviewed transcript | Reliability: A for Perkins's attributed statements and reproduced primary appendices; B for retrospective chronology/credit; C for unsupported comparative returns and industry-wide claims | Access: full transcript, printed pp. 1–54 plus appendices; quotation restricted by Bancroft publication-rights notice | Accessed: 2026-08-23 | Locator scheme: printed transcript page; PDF page is printed page + 38
Takeaways
- This is the deepest public record of Perkins's Genentech work, but it is deliberately “Genentech-centric,” funded as part of a biotechnology-history program, and surrounded by laudatory participant introductions. Its value is detail; its risk is success-selected memory. [researcher inference · strong] interview history, roman pp. xvi–xvii
- Perkins reconstructs Genentech as a staged scientific experiment rather than a conventional laboratory build: subcontract the minimum proof, preserve founder ownership, and commit larger amounts only after expression worked. [investor-stated · strong] printed pp. 4–7
- He explicitly gives Swanson primary credit for assembling the company, building the scientific culture, and seeing the technology more clearly than anyone; this constrains any founder narrative centered on Perkins. [investor-stated · strong] printed pp. 7 and 53
- Perkins defines his operating lane as finance, capitalization, board leadership, licensing/partnership architecture, management mediation, and ensuring the company never ran out of money—not choosing research programs. [investor-stated · strong] printed pp. 21–24 and 49
- The transcript records multiple genuine disagreements: IPO timing, founder/CEO succession, Roche, patent/FDA posture, and public-company forecasts. The package is therefore more useful as a governance record than as an unbroken success story. [researcher inference · strong] printed pp. 12–17, 36–48
- Perkins admits that clinical R&D partnerships created conflicts and that he invested personally. Disclosure and SEC review are evidence, but the transcript lacks terms and independent assessment needed to determine whether personal/fund/company incentives were fairly aligned. [investor-stated · strong] printed p. 10
- He repeatedly emphasizes focus: license or spin out adjacent fields, preserve the core scientific mission, and build a fully integrated pharmaceutical company only when capital and organizational capacity can support it. [investor-stated · strong] printed pp. 8–9, 14–15, and 23–24
Quotes
- No transcript quotation is reproduced because Bancroft reserves publication rights. This note uses page-stable paraphrase.
Biography and decision evidence
- Perkins says K&P began in 1972, Tandem in 1974, and Genentech in 1976. He served as Genentech chair from formation until the 1990 Roche transaction, then chaired the executive committee and remained a director until 1995. [investor-stated · moderate] printed p. 1 and biographical information, roman pp. xvii–xviii
- The source describes the early US venture industry as tiny and Fund I as $8 million, then makes a participant claim of roughly 40% annual compounded K&P returns over almost 30 years. No audited cash-flow schedule is attached, so the performance claim remains reported. [investor-stated · insufficient public record] printed p. 2
- University Laboratories, HP's computer business, and K&P formation are presented as the operator/finance preparation that made Perkins useful to Swanson. This is causal self-assessment, not independent attribution. [investor-stated · moderate] printed pp. 15–16
- Perkins says he spent roughly one afternoon or day per week at Genentech, discussed board agendas with Swanson, and learned enough science to follow reviews without imposing research choices. [investor-stated · moderate] printed pp. 21–22 and 46–47
- His 1990s disengagement followed his wife's lymphoma diagnosis and a deliberate exit from most boards; he retained Genentech partly for access to cancer expertise before resigning after her death. [investor-stated · strong] printed p. 48
Deal reasoning and decision evidence
From Cetus to Genentech
- K&P invested about $500,000 in Cetus at what Perkins later called too high a price, then assigned Swanson to salvage progress. Donald Glaser surfaced gene splicing; Swanson—not Perkins or Cetus management—recognized the opportunity and pursued Boyer. [retrospective witness · moderate] printed pp. 3–4
- Swanson and Boyer proposed a conventional roughly $1 million build. Perkins's diligence questions were equipment, sequence, measurement, time, and success criteria; Boyer's prepared answers established experimental coherence without eliminating scientific risk. [investor-stated · strong] printed pp. 4–5
- The revised plan distributed the experiment across UCSF, City of Hope, and Caltech. Swanson negotiated most contracts, Kiley handled legal detail, and Perkins checked capability/approved the arrangement. Later patent litigation shows that technically elegant staging can transfer rather than eliminate rights risk. [retrospective witness · strong] printed pp. 5–6
- Perkins recalls $100,000 for somatostatin, another $50,000, and small monthly commitments toward insulin. Later recollections elsewhere vary, so the reliable conclusion is staged scale, not an exact first-check figure. [investor-stated · moderate] printed pp. 6–7
Licensing, integration, and focus
- Perkins preferred broad patent licensing over attempting to exclude the field, reasoning that royalties could finance the company and avoid endless litigation before products generated cash. Lilly manufacturing was a capital-sharing choice, not abandonment of integration. [investor-stated · strong] printed pp. 8–9
- He says Swanson deserves full credit for regional licensing, Japanese relationships, and much of the early scientific/commercial vision. Perkins led parts of the Lilly negotiation but not Kabi or early Roche licensing. [investor-stated · strong] printed pp. 22–23
- The 1978 FIPCO plan envisioned manufacturing and marketing, while licensing and partnerships funded the bridge. Perkins defines his job as preventing a cash-out; Swanson's job was making the company merit each higher-priced round. [investor-stated · strong] printed pp. 23–24
- Agriculture and veterinary products were considered but licensed/spun away to preserve organizational focus. Perkins says Genentech avoided the bovine-growth-hormone controversy by licensing it to Monsanto; the claimed economics and counterfactual are retrospective. [investor-stated · moderate] printed pp. 14–15
Financial engineering and IPO
- Clinical R&D partnerships moved trial funding into separate vehicles; junior common stock attempted to restore startup-like employee economics after IPO. Both structures were disclosed to the SEC, later copied/abused elsewhere, and eventually restricted. [investor-stated · moderate] printed pp. 10–12
- Personal investment in R&D partnerships and the company repurchase right created an acknowledged conflict. The transcript asserts profitability and SEC acceptance but provides no terms, fairness analysis, or distribution ledger. [researcher inference · strong] printed p. 10
- Perkins pushed the IPO to be first ahead of Cetus; Swanson resisted workload/public-market burdens. The eventual decision was strategic branding rather than K&P liquidity, because Tandem had already returned capital and Fund II was fundable. [investor-stated · strong] printed pp. 12–13
- The first-day price spike exceeded the intended underpricing. Perkins later treats the publicity as beneficial while conceding the company had to earn back the transient price over years. [retrospective witness · strong] printed pp. 29–31
Board construction and governance
- Perkins recruited Dave Packard and industrial directors rather than filling the board with venture capitalists. He risks over-crediting board composition, but the source identifies the deliberate goal: add operating scale, industrial credibility, and public-market confidence. [investor-stated · moderate] printed pp. 20–21
- Swanson/research leaders controlled research priorities; Perkins/Middleton and management concentrated on finance, production, and marketing. This is clearer functional attribution than later generic “hands-on” claims. [investor-stated · strong] printed pp. 21–22
- Perkins preferred decisions to be unanimous and used the executive committee for emergencies, issue-framing, and housekeeping. That can improve preparation but risks suppressing recorded dissent or centralizing authority. [investor-stated · moderate] printed pp. 45–47
- He describes repeatedly mediating Swanson/Raab by restating each position in front of both parties. The arrangement preserved two valuable executives for years but failed soon after Perkins left, suggesting dependence on a person rather than a durable governance system. [researcher inference · strong] printed pp. 16–17 and 42–44
Roche transaction
- Concern over product-pipeline weakness, share-price pressure, and hostile takeover led Genentech to a structured Roche deal. Fred Frank was the key banker; Raab promoted him, Swanson agreed, and Perkins became increasingly involved as board recommendation approached. [retrospective witness · strong] printed pp. 17–18 and 36–38
- They selected Roche without an auction to protect scientific independence and prevent a hostile acquisition. Fairness work used earnings/revenue multiples and comparable transactions, but Perkins admits his own preferred venture exit was ordinarily an IPO and market price. [investor-stated · moderate] printed pp. 37–38
- Governance terms left Roche with a minority of board seats despite economic power. Perkins characterizes Roche's conduct as good during his tenure; this is a participant assessment, not an independent post-deal outcome study. [retrospective witness · moderate] printed pp. 39–40
Philosophy signals
- technical diligence: translate unknown science into equipment, sequence, measurement, time, and falsifiable success criteria rather than pretending domain mastery. [documented behavior · strong] printed pp. 4–5
- risk / staging: use existing institutions to run the smallest proof, then step valuation and capital after evidence. Rights allocation must be treated as a coequal risk. [researcher inference · strong] printed pp. 5–7
- focus: license, partner, or spin out adjacencies when manufacturing, regulation, distribution, or scientific scope would dilute the core. [investor-stated · strong] printed pp. 8–9 and 14–15
- talent: world-class research required university-like intellectual freedom, publication after patents, sufficient resources, and meaningful equity. Perkins assigns this system primarily to Swanson/Boyer. [investor-stated · strong] printed p. 7 and p. 53
- capital: the chair's obligation was financing continuity across private rounds, licenses, partnerships, public markets, and strategic capital—not minimizing absolute capital. [investor-stated · strong] printed pp. 23–24 and 31–34
- boards: recruit complementary authority, prepare issues before meetings, preserve formal full-board legitimacy, and use executive power sparingly. [investor-stated · moderate] printed pp. 20–22 and 45–47
- exits: an IPO can be financing, talent branding, and category definition—not merely liquidity; strategic sale can preserve independence if governance is explicitly structured. [researcher inference · strong] printed pp. 12–13 and 37–40
- forecast discipline: management forecasts can create legal and valuation damage. Perkins criticizes Gower's tPA call and admits his own casual Compaq profitability forecast was too aggressive. [investor-stated · strong] printed pp. 44–45
Attribution and decision-rights map
- Robert Swanson: founding commercial entrepreneur; recognized gene-splicing potential; recruited Boyer; negotiated core contracts/territories; created culture; selected science and people; operated the company. [researcher inference · strong] printed pp. 3–9, 21–24, 52–53
- Herbert Boyer / Stanley Cohen: scientific foundation and patents; Boyer co-founded, joined the early board, and guided research culture. [researcher inference · strong] printed pp. 4–8
- Eugene Kleiner / K&P: first-fund vehicle and initial capital; Kleiner knew/hired Swanson and approved the investment but is less visible in Perkins's Genentech-centered narration. [researcher inference · strong] printed pp. 1–4
- UCSF / City of Hope / Caltech / Kiley: executed or contracted the initial experiment and shaped rights; later litigation makes them part of the economics and risk ledger. [researcher inference · strong] printed pp. 5–7
- Fred Middleton: CFO and financial-engineering collaborator; attribution for R&D partnerships is disputed between him and Perkins. [researcher inference · moderate] printed pp. 9–10
- Brook Byers / Ivor Royston / Ted Greene: lead source/science/operator chain for Hybritech; Perkins supplied approval/capital/board support but was not the originating deal lead. [researcher inference · strong] printed p. 14
- Kirk Raab: operating scale, marketing/manufacturing, and Roche execution; the Swanson/Raab tension required board mediation. [researcher inference · strong] printed pp. 16–18 and 42–44
- Fred Frank / Genentech board / Roche: transaction design, fairness process, approval, and post-deal governance; Perkins was chair/negotiator, not sole architect. [researcher inference · strong] printed pp. 36–41
Contradictions, counterevidence, and access limits
- The transcript is interviewee-reviewed direct voice, not a final verified history; it repeatedly flags memory limitations and the need for records on specific rounds. [researcher inference · strong] printed p. 13
- Perkins's claim that K&P was the world's most successful financial institution and returned reported: roughly 40% annually is unsupported by audited fund cash flows here. [researcher inference · strong] printed p. 2
- His “no strategic error” claim is hindsight-heavy and conflicts with admitted weak contract drafting, later litigation, failed research, excessive forecasts, and management conflict. [researcher inference · strong] printed pp. 8–9, 44–45
- Personal investment in clinical R&D partnerships is in tension with his later anti-conflict rhetoric; SEC review does not itself prove fair economic allocation. [researcher inference · moderate] printed p. 10
- Unanimity and executive-committee control can improve deliberation or hide dissent; the transcript lacks board-minutes evidence to distinguish them. [researcher inference · moderate] printed pp. 45–47
- Perkins assigns himself financial/governance importance but explicitly denies selecting Genentech's later research programs. Do not attribute cancer-therapy choices or later scientific success to him. [researcher inference · strong] printed p. 49
- The archive forbids unauthorized publication quotation; this note paraphrases. Appendices can corroborate formal roles but do not independently prove the interview's causal claims. [researcher inference · strong] archive rights statement
Forecasts and prediction record
- In 2001 Perkins predicted biotechnology would remain profoundly important for a century and advised young people to work/invest in it for the long term. This is a broad, unscored forecast. [investor-stated · provisional] printed pp. 49–50
- He predicted a contemporary implantable-heart device would disappear because FDA iteration requirements made the program infeasible. The transcript supplies no later score; treat it as an open forecast, not a validated lesson. [investor-stated · provisional] printed pp. 50–51
- He expected biotech market risk to rise after obvious targets were exhausted and technology became more powerful. This is a falsifiable sector-development thesis but lacks a dated measurement plan. [investor-stated · moderate] printed pp. 52–53
- He identifies Gower's billion-dollar tPA sales forecast and his own Compaq profitability comment as overly aggressive. These are documented calibration failures, not evidence that all forecasts should be avoided. [investor-stated · strong] printed pp. 44–45
- The IPO leadership thesis was made before listing but is recounted after success; the source has no preserved decision-date memo or probability. [researcher inference · strong] printed pp. 12–13
Open questions
- Can the 1976–80 contracts, cap tables, board minutes, and bank records reconcile the exact first-check sequence, ownership, and partner contribution?
- What were the clinical R&D partnership terms, conflict approvals, personal allocations, cash returns, and losses?
- Which contemporaneous documents separate Perkins's and Middleton's contributions to financial engineering?
- Did unanimous board practice suppress meaningful recorded dissent, or did pre-meeting work improve decisions?
- How did ordinary, failed, or non-Genentech biotech investments perform relative to the two outliers?
- Which Swanson, Boyer, Kiley, Middleton, Raab, employee, and Roche accounts most strongly contradict Perkins's version?
- How did the 2001 biotech, FDA, and market-risk forecasts score afterward using contemporaneous definitions?
[t] CHM “Pioneers of Venture Capital” panel — event 2002, not 2010
- URL: https://www.youtube.com/watch?v=ult7VFgxiKo | Event: 2002-09-30 | Official video upload: 2017-11-21 | Format: 1:54:12 official Computer History Museum panel video with machine-generated
en-origcaptions | Reliability: A for CHM custody and event metadata; B for speaker-bounded paraphrase from machine captions; C for retrospective performance claims | Access: full public video; existing machine-caption track retrieved without audio transcription; three moderator-bounded Perkins turns processed | Accessed: 2026-08-23 | Locator scheme: video timestamp, moderator direct address, and explicit panel handoff
Takeaways
- The requested filename's “2010” date is not supported by the official records. CHM dates the event to September 30, 2002; video metadata dates the upload to November 21, 2017. [contemporaneous record · strong] CHM event record official video
- Asked how he entered venture capital, Perkins calls it a byproduct of entrepreneurship. He describes HP machine-shop apprenticeship, management of its computer business, funding University Laboratories from household savings, obtaining Packard's permission, raising outside capital, and merging the laser company into Spectra-Physics before forming K&P. [investor-stated · moderate] video, 00:35:52–00:39:11
- Perkins remembers roughly $15,000 of family savings for the laser experiments and an $8 million first K&P fund. These amounts are retrospective and the savings figure differs from the $10,000 reported in the 1993 Red Herring introduction; neither should be silently harmonized. [investor-stated · provisional] video, 00:37:31–00:38:19 and 00:39:00–00:39:11 Red Herring introduction
- For Tandem, Perkins starts with Jimmy Treybig's fault-tolerant-computer idea and unsuccessful effort to interest HP. He describes Treybig joining K&P, continuing the concept with partnership help, K&P financing the initial effort, Pitch Johnson joining later as the validating outside investor, and the team adding operating and finance leadership. [investor-stated · moderate] video, 00:50:54–00:53:18
- The same Tandem turn includes Snow-Job as a failed Treybig-associated investment. Its inclusion next to Tandem makes survival and team learning visible, but the panel gives no check, ownership, loss, or shutdown chronology. [investor-stated · provisional] video, 00:51:23–00:51:49
- Asked about Apple, Perkins says he never met Steve Jobs: after seeing several computer kits he disliked, he categorized the scheduled opportunity as another similar computer and did not take the meeting. The miss occurred before full diligence, making category fatigue and delegated screening the informative failure modes. [investor-stated · moderate] video, 01:08:03–01:08:43
Quotes
- None retained: the available caption track is machine-generated and materially garbles names and wording. It supports speaker-bounded paraphrase, not verbatim quotation.
Biography and decision evidence
- The panel places Perkins alongside William Draper III, Pitch Johnson, Burt McMurtry, Arthur Rock, and Don Valentine. Only turns beginning with a moderator address to Tom and ending at the next explicit handoff are assigned to Perkins. [researcher inference · strong] event participants full video
- Perkins's operating sequence is specific: hands-on shop work, HP computer management, a side venture built with employer permission, outside fundraising, chair-level oversight, acquisition, and then full-time venture work. It is direct self-narration, not proof that each stage caused investing performance. [researcher inference · moderate] video, 00:35:52–00:39:11
- Tandem credit is distributed in Perkins's own telling: Treybig conceived and worked the fault-tolerant idea; unnamed colleagues and Perkins assisted; K&P supplied the first partnership financing; Johnson validated from outside; later hires filled operations and finance. [researcher inference · strong] video, 00:50:54–00:53:18
- Perkins remembers Tandem's 1974 formation, rapid public offering, long operating life, and acquisition by Compaq. His imprecise “I believe” dating is retrospective memory and requires company records before use as exact chronology. [investor-stated · provisional] video, 00:52:30–00:53:18
Philosophy signals
- operator apprenticeship: firsthand production, management, fundraising, and board exposure form the basis for a hands-on investor identity. [researcher inference · moderate] biography turn
- sourcing / incubation: recruit an operator already carrying a technical concept, develop it inside the partnership, and seek an external validator before scaling. [researcher inference · moderate] Tandem turn
- people / boards: company creation requires distinct technical, operating, finance, and governance contributors; financier involvement is not founder substitution. [researcher inference · strong] Tandem turn
- anti-portfolio: repeated weak examples can make a later exceptional opportunity look like another member of the failed category, causing rejection before evidence collection. [researcher inference · strong] Apple turn
Contradictions, counterevidence, and access limits
- No 2010 CHM event was authenticated for this title. The repository filename is preserved as assigned, while the note records the official 2002 event and 2017 upload dates. [contemporaneous record · strong] CHM event video
- Machine captions repeatedly corrupt names, companies, and clauses. Garbled passages were omitted; no automated-caption sentence is promoted to a quote. [researcher inference · strong] official video
- Perkins's claims that the $8 million vehicle was the world's largest, K&P supplied all initial Tandem venture capital, and Johnson was the only validating outside investor remain participant memory until partnership, financing, and cap-table records control them. [researcher inference · strong] video, 00:39:00–00:39:11 and 00:50:54–00:53:18
- The Apple story documents a missed meeting, not an evaluated investment rejection. The later TechCrunch report compresses the miss into not evaluating Wozniak; neither supplies a decision memo or offered terms. [researcher inference · strong] CHM Apple turn TechCrunch report
- The panel is nearly three decades after Tandem formation and uses success-selected anecdotes. Repetition by Perkins in other appearances is consistency, not independent corroboration. [researcher inference · strong] event record
Open questions
- Locate a speaker-labeled human transcript or creator captions and recheck every Perkins turn and proper name.
- Reconcile University Laboratories capitalization and the $10,000/$15,000 savings discrepancy from contemporaneous records.
- Reconstruct Tandem's idea ownership, initial checks, securities, cap table, external investors, hires, milestones, and IPO timing.
- Recover the Apple meeting referral, delegation, opportunity materials, proposed terms, and decision-date communications.
[t] Red Herring interview — December 1993
- URL: https://cryptoniteventures.substack.com/p/venture-pioneer-tom-perkins-rip-unveils | Published: original interview December 1993; mirror 2023-08-27 | Format: interviewer-posted mirror of an edited Red Herring Q&A | Reliability: B for attributed speech and original-interviewer custody; C for unverified performance totals and retrospective causal claims | Access: full mirrored Q&A; no audio or contemporaneous scan located | Accessed: 2026-08-23 | Locator scheme: interviewer-question headings and opening words of Perkins answers
Takeaways
- Perkins describes becoming a venture capitalist through operating work: an unsuccessful HP-backed company, HP's computer business, and University Laboratories preceded the 1972 partnership with Eugene Kleiner. The account is direct memory twenty years later, not a contemporaneous employment file. [investor-stated · moderate] questions beginning “What inspired you” and “Were there any role models”
- He reports an $8 million first fund, unused commitments, and only two “super” outcomes—Tandem and Genentech. This is unusually helpful self-reported denominator evidence, but “hundreds of millions” returned is not tied to audited cash flows, dates, fees, or net LP return. [investor-stated · provisional] questions beginning “How big was your first fund?” and “What were your biggest hits?”
- His explicit underwriting formula is to identify the deal's largest risk and make the initial check fund a decisive test. He assigns Genentech's main early uncertainty to technical feasibility and Tandem's to multiprocessing software, while describing market opportunity as visible in both cases. [investor-stated · moderate] question beginning “Why were those deals so successful?”
- The Snow-Job motorcycle/snowmobile is his named early failure. He attributes failure to launch timing during the oil embargo, but also admits reluctance to discuss failures; the anecdote does not reveal product demand, unit economics, financing, governance, or stop criteria. [investor-stated · provisional] question beginning “What were some of your more memorable failures?”
- Perkins identifies firm design—not only deal picking—as KPCB's advantage: partner selection, LP treatment, conflict rules, immediate profit distributions, separation of old- and new-fund follow-ons, founder economics, portfolio cooperation, and succession. These are prescriptions and self-reported practices, not inspected agreements. [investor-stated · moderate] questions beginning “What is KPCB's competitive advantage?” and “You've been credited”
- He says the firm sought partners with technical backgrounds, hands-on orientation, ambition, competitiveness, and impatience. The passage names Caufield, Byers, Doerr, Lally, and Khosla, so later KPCB work remains team output rather than a Perkins portfolio. [investor-stated · moderate] question beginning “What are the elements to success”
- Perkins treats people problems—recruiting, retention, motivation, and founder succession—as the hardest operating work, even though his selection rhetoric elsewhere emphasizes ideas and technical risk. Acuson's Sam Maslak is the favorable founder-development example. [investor-stated · moderate] questions beginning “Do you ever take a risk” and “What's the hardest part”
- The reported sourcing engine is founder referral: Perkins says almost every deal arrived through a previously backed entrepreneur and none through unsolicited mail. This is a strong network hypothesis but lacks an intake denominator or exception log. [investor-stated · provisional] question beginning “Where does KPCB find its deals?”
Quotes
- “We try to isolate whatever the biggest risk is in a given deal.” [investor-stated · strong] risk-formula answer
- “Once we sign the check, we’re on the team.” [investor-stated · strong] competitive-advantage answer
Biography and decision evidence
- Perkins credits Sandy Robertson with introducing him to Kleiner after each separately expressed interest in venture capital. This routes partnership formation credit beyond a two-founder self-creation story. [investor-stated · moderate] question beginning “How did you get together with Eugene Kleiner?”
- For Tandem and Genentech, Perkins says the partnership incubated both and that Jimmy Treybig and Bob Swanson were partners in the fund. The account supports partnership involvement but does not erase Treybig's operating authorship, Swanson and Herbert Boyer's founder/scientist roles, or later partner work. [researcher inference · strong] questions beginning “What were your biggest hits?” and “Why were those deals so successful?”
- The 1993 snapshot says Perkins had reduced seventeen board seats and sixteen chairmanships to four—Tandem, Genentech, Acuson, and Internet Systems—and created a gradual partner scale-back mechanism. It supplies a succession rationale, not a legal ownership schedule. [investor-stated · moderate] question beginning “You started reducing your participation”
- Perkins explicitly tells prospective VCs that operator experience in a growing enterprise is necessary for his hands-on model and says he would choose entrepreneurship over venture capital if restarting at thirty. [investor-stated · moderate] closing two questions
Philosophy signals
- risk / picking: fund the cheapest decisive experiment against the dominant uncertainty before scaling capital. [researcher inference · strong] risk-formula answer
- portfolio: pursue non-consensus, new-market companies whose upside can compensate for a high failure rate; avoid undifferentiated copycats. [investor-stated · moderate] question beginning “KPCB is known for swinging for the fence”
- boards / people: capital creates a duty to help finance, recruit, resolve conflict, and identify risks, but intensive involvement can become investor overreach. [researcher inference · moderate] questions beginning “How would you summarize” and “What's the hardest part”
- sourcing: prior founders act as a reputation-mediated referral network; the mechanism depends on earned trust and may also narrow access to outsiders. [researcher inference · moderate] sourcing answer
- governance / conflicts: precommitment on personal investments, follow-ons, distributions, competitor funding, and founder economics is presented as a brand asset. [investor-stated · moderate] LP- and entrepreneur-interest answers
Contradictions, counterevidence, and access limits
- The source is a 2023 HTML mirror posted by original interviewer Anthony B. Perkins, not a scanned December 1993 Red Herring issue; layout, edits, and omissions cannot be compared with the original. [contemporaneous record · moderate] mirror masthead
- Fund-return, $30 billion distribution, 100,000-job, 180-deal, and industry-creation figures are participant claims without definitions, vehicle boundaries, dates, or audited controls. They must not become person-level return or coverage denominators. [researcher inference · strong] fund, motivation, and entrepreneur-interest answers
- “Only two” super outcomes and admitted failure reluctance cut against a smooth repeatability narrative. Snow-Job also shows that isolating technical risk does not remove macro timing or demand risk. [researcher inference · strong] hits, failures, and risk answers
- The statement that KPCB's investments gave birth to biotechnology and ASICs over-compresses founder, scientist, operator, university, co-investor, and broader market contributions. [researcher inference · strong] motivation answer
- Strong founder alignment sits beside partners being more ambitious than entrepreneurs and being “too involved.” The interview supplies the tension but no independent founder survey or board record to resolve it. [researcher inference · moderate] partner-selection and competitive-advantage answers
Open questions
- Locate and compare the original December 1993 issue scan, including any sidebars, tables, edits, or omitted questions.
- Reconcile the first-fund commitments, invested capital, company count, cash distributions, fees, carry, timing, and net LP return from partnership records.
- Reconstruct Snow-Job from decision-date product, market, financing, board, and shutdown evidence rather than the oil-embargo anecdote alone.
- Test the claimed referral-only sourcing and conflict rules against an intake log, rejected deals, exceptions, partnership agreements, and LP reporting.
[t] Risk & Reward — alleged Tom Perkins episode route
- URL: https://www.riskandreward.org/episode/tom-perkins | Date: unresolved | Format: putative interview URL | Reliability: insufficient | Access: unavailable/unauthenticated | Accessed: 2026-08-23 | Locator scheme: none
Retrieval result
- The URL returned HTTP 200 with only a 114-byte body and no title, guest metadata, media, transcript, date, or source attribution. Exact-name/site/title searches did not surface a target-specific body. A nominal success code is not evidence of an episode. [researcher inference · strong] degraded route
- No audio surfaced, so there was nothing legitimate to send to the OpenAI transcription API. [researcher inference · insufficient public record] degraded route
Usable evidence
- None. The route is retained to document an exhausted lead and prevent later researchers from silently treating it as processed speech. [researcher inference · strong] degraded route
Contradictions, counterevidence, and access limits
- Date, host, program identity, guest identity, duration, and content are all unresolved. [researcher inference · insufficient public record] degraded route
- Do not merge this route with Something Ventured: Risk, Reward and the Original Venture Capitalists merely because “risk and reward” overlaps in title. [researcher inference · strong] degraded route Smithsonian Something Ventured item
Open questions
- Was this a malformed derivative URL for a different program or a removed episode?
- Do DNS history, a verified archive snapshot, or creator catalog expose authentic metadata and media?
[t] *Something Ventured* — raw Tom Perkins interview route
- URL: https://americanhistory.si.edu/collections/archival-item/sova-nmah-ac-1467-ref12 | Interviewed: 2008-10-08 | Format: raw documentary-interview transcript archival item, collection NMAH.AC.1467_ref12 | Reliability: A for Smithsonian custody/metadata | Access: in-person research; no public item body | Accessed: 2026-08-23 | Locator scheme: collection/item identifiers only
Retrieval result
- Smithsonian's item record identifies “Transcripts of interviews with Tom Perkins of Klein Perkins Caufield & Byers,” dated 2008-10-08, in the Something Ventured Film Transcripts series. The collection contains 23 interview transcripts and marks this item for in-person research. [contemporaneous record · strong] Smithsonian item record
- Stanford and Smithsonian collection routes expose custody/finding-aid metadata but no public raw transcript file. The edited film and press accounts are not substituted for Perkins's complete speaker record. [contemporaneous record · strong] Stanford route Smithsonian collection
Usable evidence
- Interview existence, exact date, target identity, custody, collection identifier, and in-person access boundary are usable; no substantive quote, deal claim, or philosophy is drawn from the inaccessible body. [contemporaneous record · strong] Smithsonian item record
Contradictions, counterevidence, and access limits
- The archive spells the firm “Klein Perkins” in the item title; this is catalog text, not evidence of a different identity. [contemporaneous record · strong] Smithsonian item record
- Edited documentary sequencing can compress attribution and chronology; raw Perkins text remains unavailable without authorized archival research. [researcher inference · strong] Smithsonian item record
- No archive or person was contacted because the workflow does not authorize outreach. [researcher inference · strong] Smithsonian item route
Open questions
- Can authorized in-person research produce a lawful note set under Smithsonian rights conditions?
- Which edited-film claims differ from the raw transcript, and what material was omitted?
[t] TechCrunch Disrupt SF — Perkins and Don Valentine
- URL: https://techcrunch.com/2013/09/11/vc-titans-tom-perkins-and-don-valentine-articulate-what-makes-a-good-vc/ | Published: event and report 2013-09-11 | Format: co-bylined, edited event report associated with a hosted Disrupt appearance | Reliability: B for event custody and attributed fragments; C for omitted context and reporter compression | Access: full report and official agenda available; hosted video/caption/transcript body did not surface through the available article route, so no invented timestamps | Accessed: 2026-08-23 | Locator scheme: article paragraph openings and subheadings; official agenda slot 2:40–3:05 p.m.
Takeaways
- TechCrunch's official agenda places Perkins and Sequoia founder Don Valentine in a twenty-five-minute “Something Ventured” session. The report says the discussion centered on venture experience, but it is a short selection rather than a complete transcript. [contemporaneous record · strong] official agenda report opening
- Perkins says he begins with the idea, using a deliberately provocative character inference, and then checks whether back-page numbers are large before reading the business description. The report does not define which numbers, forecast horizon, market measure, or diligence stage he meant. [investor-stated · moderate] report paragraph beginning “I feel you invest”
- The sequencing implies upside screen before narrative: first ask whether a successful outcome can be large enough, then examine the proposed business. It does not establish that Perkins ignored team, technical proof, terms, or downside in a full process. [researcher inference · moderate] same opening quotation paragraph
- Perkins acknowledges K&P failed even to evaluate Steve Wozniak and calls that a major mistake. The fragment is anti-portfolio evidence but supplies no decision date, intake route, delegation, memo, valuation, or opportunity cost. [investor-stated · moderate] closing paragraph beginning “Perkins essentially admitted”
- Valentine—not Perkins—supplies the panel's quoted visionary, risk-taker, business-school, and HR arguments. Adjacency in the report does not make them Perkins rules. [researcher inference · strong] subheadings “I'm against all business schools” and “Human Resources”
Quotes
- “I feel you invest in the idea because bad people don’t have good ideas.” [investor-stated · strong] opening quotation paragraph
- “We didn’t even look at Steve Wozniak. Big mistake.” [investor-stated · strong] closing paragraph
Biography and decision evidence
- The session joins two retired founding-era investors in connection with Something Ventured. It records their views in 2013, not the contemporaneous information set for 1970s investments. [contemporaneous record · strong] report context
- The Apple fragment establishes a K&P miss and Perkins's later acknowledgment. It does not show that Perkins personally saw Wozniak, met Steve Jobs, or made a documented investment-committee vote. [researcher inference · strong] closing paragraph
- TechCrunch's introductory references to firm portfolios are context, not personal deal attribution. Sequoia and KPCB outcomes also remain separate from each other. [researcher inference · strong] opening paragraphs
Philosophy signals
- picking / market: test whether the opportunity can support an exceptional outcome before investing time in the full plan. [researcher inference · moderate] Perkins opening fragment
- people: idea-first rhetoric is not a license to infer character from idea quality; evidence on integrity, collaboration, and execution still requires independent diligence. [researcher inference · strong] Perkins opening fragment
- anti-portfolio: category fatigue or failure to inspect an unconventional builder can prevent the underwriting process from starting at all. [researcher inference · moderate] Apple fragment
Contradictions, counterevidence, and access limits
- The report contains only a few selected fragments and no timestamps. Without the hosted video's accessible captions or transcript, no claim is made about full context, frequency, audience reaction, or omitted qualifications. [researcher inference · strong] event report
- The idea/character sentence is aphoristic and logically invalid as a universal rule: harmful or incapable people can propose attractive ideas, and strong builders can begin with poorly expressed ones. [researcher inference · strong] opening fragment
- The back-page-first method may prioritize scale efficiently or invite spreadsheet anchoring. The article supplies no ordinary cases, forecast scoring, or rejection denominator. [researcher inference · moderate] opening fragment
- The Apple fragment compresses a longer missed-meeting/delegation story preserved in the 2002 CHM panel. The two can coexist, but the short report cannot establish who rejected whom or why. [researcher inference · strong] TechCrunch closing paragraph CHM panel, 01:08:03–01:08:43
- TechCrunch's host copy does not provide investment memos, terms, ownership, returns, or counterfactual decision evidence for any named company. [researcher inference · strong] report
Open questions
- Recover the complete hosted video or a legitimate caption/transcript track and timestamp every Perkins turn.
- Identify which “back-page” measures Perkins used, how he normalized early-stage uncertainty, and whether forecasts were scored later.
- Reconstruct the Apple intake, scheduled meeting, delegation, pass rationale, and opportunity terms from decision-date records.
- Test idea-first selection against ordinary and failed deals and distinguish Perkins from Valentine's adjacent remarks.
[t] Venture Voice — Tom Perkins episode route
- URL: https://venturevoice.com/tom-perkins-kleiner-perkins-caufield-byers | Date: circa 2005–2008; exact episode date unresolved | Format: historic podcast appearance; current slug resolves to generic series landing page | Reliability: A for current host roster metadata, insufficient for episode content | Access: unavailable body | Accessed: 2026-08-23 | Locator scheme: none; no media or transcript body surfaced
Retrieval result
- The exact slug returned HTTP 200 but a generic 41,774-byte Venture Voice landing page, not a Perkins episode. The current roster authenticates Tom Perkins as a guest but supplies no episode date, audio URL, show notes, transcript, or remarks. [contemporaneous record · insufficient public record] current series roster
- Exact-name/title/host and legacy-slug searches surfaced the current roster but no legitimate target MP3, enclosure, archive snapshot body, or transcript. Because no audio file surfaced, no OpenAI transcription could be performed. [researcher inference · insufficient public record] current series roster
Usable evidence
- Appearance metadata only: Venture Voice lists Tom Perkins among its historic guests. The landing page's promotional “Funded Google, AOL, Genentech” is not person-level attribution evidence and is excluded. [researcher inference · strong] current roster
Contradictions, counterevidence, and access limits
- The SCOUT entry's “2006” date and audio format were not verified from a target-specific public body. [researcher inference · insufficient public record] current series roster
- A current HTTP 200 is not episode access when the slug silently serves a generic page. [researcher inference · strong] legacy episode slug
- Nothing said in other interviews is imputed to this lost appearance. [researcher inference · strong] current series roster
Open questions
- Does the original RSS feed, Internet Archive capture, Apple podcast catalog, or host archive preserve the episode enclosure and date?
- If a lawful MP3 surfaces, transcribe it through the OpenAI API and compare it with the 2007 memoir-era interview family.
[t] WBUR *On Point* — “Silicon Valley's Tom Perkins”
- URL: https://www.wbur.org/onpoint/2007/11/28/silicon-valleys-tom-perkins | Aired: 2007-11-28 | Format: 45:59 official public-radio interview with Jane Clayson and callers | Reliability: A for custody/speaker turns, B for retrospective claims, B for OpenAI
whisper-1segment wording | Access: full official MP3; full machine transcript produced through the OpenAI API on 2026-08-23 | Locator scheme:HH:MM:SSfrom the official MP3; host/Perkins turns bounded by direct handoffs; advertisements at 00:00–01:08 excluded
Key claims and reasoning
- At 00:02:55–00:03:56, Perkins says the first fund raised $8 million, invested about $4 million, and produced two major hits, Tandem and Genentech. This is a fund-level participant claim without audited cash flows or a 17/18-company roster. [investor-stated · moderate] official episode
- At 00:04:17–00:05:40, he says KPCB prefers an idea before a complete team, helps remove the idea's principal risk with early money, then adds capital and management. He further says Tandem and Genentech were “started by ourselves,” a firm-level compression that requires founder and partner controls. [investor-stated · moderate] official episode
- At 00:06:58–00:07:57, risk signals early entry and potential uniqueness; KPCB's claimed specialty is recognizing, managing and reducing it, with failures ideally occurring before substantial capital. The rule is explicit, but the episode supplies no portfolio-wide application denominator. [investor-stated · moderate] official episode
- At 00:23:23–00:25:23 and 00:29:31–00:31:38, he operationalizes the rule: hold team size and burn down until the specific technical, manufacturing, people, patent, or customer risk is controlled; stage financing; then invest more at higher prices. [investor-stated · strong] official episode
- At 00:34:31–00:36:23, he says excess venture capital pushes newer managers to invest weaker deals, while KPCB can wait; his screen orders customers and competitive path, then risk, then founder identity. This ordering tensions with the simpler idea-first formulation and makes market evidence more important than its slogan suggests. [investor-stated · moderate] official episode
Biography and decision evidence
- Snow-Job, 00:12:35–00:15:43: Perkins describes an enjoyable motorcycle/snowmobile conversion that failed as gasoline restrictions disrupted distribution, then admits it lacked Genentech-scale importance even if it worked. This is both an external-shock story and a retrospective market-size correction. [investor-stated · moderate] official episode
- Long holding, 00:15:44–00:16:32: he says an IPO is a way station and investors may remain on boards. His Google example explicitly names John Doerr as the continuing director, preserving later-partner attribution. [investor-stated · moderate] official episode
- Satellite case, 00:31:51–00:33:08: Perkins reports that a roughly $20 million satellite-capacity position was heading toward a $15 million write-off before launch failure triggered full insurance. He calls it luck. Company, vehicle, audited amounts, insurance terms, and cash-flow outcome are unnamed/unverified. [investor-stated · provisional] official episode
- Operating path, 00:43:23–00:45:31: he advises aspiring VCs to build a company first and links his own path to laser work and University Laboratories. The sequence is documented elsewhere; the causal selection advantage remains unmeasured. [investor-stated · moderate] official episode CHM operator control, pp. 6–7
Governance and sourcing
- At 00:19:09–00:21:16, Perkins contrasts directors who understand/direct the business with compliance-focused box-checking, using Enron as a negative example. The dichotomy is rhetorically sharp but false if read as a choice: technical understanding, legal compliance, independence, and risk control can all be necessary. [investor-stated · moderate] official episode
- At 00:25:51–00:26:58, he says KPCB reads unsolicited plans but has never funded one; he calls the network around more than 525 firm-created companies its sales force. This is firm-level, shifting-denominator evidence of referral advantage, not proof that good outsiders receive equal access or that Perkins personally built every node. [investor-stated · moderate] official episode
Attribution and contradiction controls
- At 00:37:22–00:38:53, Perkins uses “we” when saying KPCB started Compaq and then describes prototype-first team building. KPCB's own history attributes the original opportunity to Sevin Rosen and internal sponsorship to John Doerr; Perkins's supported role is later board/M&A work. The process anecdote may be firm knowledge without being Perkins's deal. [researcher inference · strong] KPCB Compaq history official episode
- The 00:03:47–00:04:01 statement that “we” started biotechnology overstates the causal field. Genentech founders Bob Swanson and Herbert Boyer, university science, research laboratories, public research support, City of Hope, Eli Lilly, employees and competitors remain material. [researcher inference · strong] Genentech founders official episode
- At 00:43:36–00:44:08, he says K&P's fair treatment made competitors follow. The missing 1972 plan, absent adoption denominator, and lack of compliance records leave this industry-causation claim insufficient. [investor-stated · insufficient public record] writing note official episode
Forecast ledger
- 00:22:08–00:22:16: Perkins predicts executive exit packages will taper as boards respond. [investor-stated · provisional] No horizon, metric, comparison set, or later score is supplied; it is routed for external scoring rather than called correct. Official episode
- 00:34:31–00:35:15: he predicts that $20–40 billion of available US high-tech venture capital is too large to deliver investors' desired 10–100x multiplication. [investor-stated · provisional] The amount, fund vintage, desired-return denominator and outcome window are undefined. Official episode
Mental-model tags
idea-before-team;specific-risk decomposition;staged financing;fail-cheap;customer sequence;operator apprenticeship;portfolio-network sourcing;director-as-business-partner;attribution control;luck disclosure. [researcher inference · strong] Official episode
Contradictions, counterevidence, and access limits
- The OpenAI transcript is machine-generated and not independently word-verified; wording is paraphrased, while official-audio timestamps allow reproduction. [researcher inference · strong] Official episode
- Speaker turns are strong when the host hands directly to Perkins; caller/host framing is not assigned to him. [researcher inference · strong] Official episode
- The 525-company count, $20–40 billion pool, satellite amounts, and Fund I figures are investor-stated and definition-incomplete. [researcher inference · strong] Official episode, 02:55–03:56, 25:51–26:58, 31:51–35:15
- This appearance is unusually valuable because it includes Snow-Job, an unnamed luck-driven satellite case, a failed-source denominator, and a Compaq attribution leak rather than only signature wins. [researcher inference · moderate] Official episode, 12:35–15:43 and 25:51–38:53
Open questions
- Can the unnamed satellite company and insurance/cash-flow record be identified?
- How many mailed plans, referrals, screened deals, failures, and successful referrals constitute the sourcing denominator?
- Does a complete K&P/KPCB vehicle ledger show that staged financing consistently improved founder ownership and fund returns?
[t] Knowledge at Wharton — risk and venture practice
- URL: https://knowledge.wharton.upenn.edu/article/venture-capitalist-tom-perkins-if-there-is-no-risk-you-have-already-missed-the-boat/ | Published: 2012-12-05; event described only as a recent Wharton San Francisco visit | Format: staff-edited event report, not a transcript | Reliability: B for attributed speech and institutional custody; C for retrospective deal totals and generalized industry claims | Access: one direct page retriever returned HTTP 403 while a later CLI spot-check returned 200; the full search-indexed report body was processed; no recording, captions, or verbatim transcript located | Accessed: 2026-08-23 | Locator scheme: article section headings and paragraph openings
Takeaways
- Perkins attributes to David Packard a minimum-contribution test: a new company should seek roughly an order-of-magnitude improvement, which normally entails high risk. This is a differentiation heuristic, not a measured threshold or sufficient investment condition. [investor-stated · moderate] opening paragraphs before “Getting the Right Answers”
- He pairs large ambition with cheap early falsification. In the Genentech example, the report says two specialized research groups tested whether DNA could be opened at the target and closed again before the venture received fuller commitment. [investor-stated · moderate] section “Getting the Right Answers”
- Perkins reports a $250,000 initial investment instead of a hypothetical $3 million commitment. The staging logic is clear, but the event report does not reconcile check tranches, laboratory payments, ownership, dilution, or the final return. [investor-stated · provisional] Genentech paragraphs in “Getting the Right Answers”
- He argues that venture decisions require a strong individual decision-maker and criticizes committee and corporate-VC incentives: career penalties can make employees avoid failures that independent venture portfolios must tolerate. The claim is mechanism-based but unsupported by a comparative dataset in this appearance. [investor-stated · moderate] paragraphs beginning “Perkins worked at HP” and “And for that reason”
- Perkins uses HP's calculator and computer entries to illustrate founder-led decisions made against staff and expert skepticism. Retrospective success does not reveal the contemporaneous alternatives, downside limits, or frequency of similarly forceful failed decisions. [investor-stated · provisional] paragraphs beginning “Two of HP's most important”
- The report connects Packard's permission for Perkins to build University Laboratories while at HP with Perkins's operator identity, and says K&P's early conflict document reflected HP ethics. The original document is not reproduced here. [investor-stated · moderate] section “A Personal Congratulations”
- Perkins contrasts smaller, low-priced IPOs when a company's best years are ahead with maximizing initial valuation. This is a long-term capital-formation prescription; the report supplies no matched outcome study or contemporaneous board records. [investor-stated · moderate] paragraphs beginning “Perkins also noted” and “But current fashion”
- In Q&A, Perkins says he weighs the idea before a complete management team, because strong executives may not join before technical risk is reduced. The sequence is idea → early evidence → capital → executive recruitment, not a claim that people never matter. [investor-stated · moderate] Q&A paragraph beginning “During the Q&A session”
- He distinguishes networking from cronyism and says prior KPCB executives become the firm's referral network. The mechanism is plausible; the accompanying meritocracy claim and “more than 500 companies” count are untested firm-wide self-reports. [investor-stated · provisional] closing Q&A paragraphs beginning “Another student wondered”
Quotes
- “We wanted to structure the deal so the risk was up front and not at the end.” [investor-stated · strong] Genentech staging paragraph
Biography and decision evidence
- Perkins names Packard as his largest career influence and credits HP with technical, operating, leadership, and ethics lessons. This is mentor attribution, not independent evidence that each later rule originated with Packard. [investor-stated · moderate] opening and HP paragraphs
- University Laboratories supplies direct founder experience before K&P: permission, an operating side venture, acquisition by Spectra-Physics, and Packard's approval. Exact capitalization and sale economics are outside this report. [investor-stated · moderate] section “A Personal Congratulations”
- The Tandem passage records Perkins's 23-year chair role and eventual competitive pressure from Intel servers, but it is staff narrative rather than a decision-date statement. [contemporaneous record · moderate] paragraph beginning “Some of Perkins' greatest hits”
- The Genentech anecdote credits Perkins and collaborators with commissioning tests; it does not identify Robert Swanson, Herbert Boyer, the laboratories, contract terms, or which individual approved each payment. [researcher inference · strong] Genentech section
Philosophy signals
- risk / picking: demand discontinuous contribution, then isolate the decisive uncertainty and buy evidence cheaply before company-scale spend. [researcher inference · strong] opening and Genentech sections
- people: a complete team can follow proof; do not require proven executives before the venture has reduced the risk that keeps them away. [investor-stated · moderate] Q&A management-team passage
- governance: individual conviction should be paired with explicit accountability, because anti-committee rhetoric otherwise concentrates authority without improving truth. [researcher inference · moderate] HP and corporate-VC paragraphs
- exits: price an IPO to support durable company financing rather than maximize first-day extraction; performance evidence remains absent here. [investor-stated · moderate] IPO paragraphs
- sourcing: portfolio executives act as distributed scouts, creating high-signal access and possible network exclusion at the same time. [researcher inference · moderate] networking Q&A
Contradictions, counterevidence, and access limits
- This is an eight-minute staff report, not a transcript. Question wording, omitted exchanges, edits, and event date are unavailable, and direct page access was blocked to automated retrieval during this run. [researcher inference · strong] canonical article
- The Genentech staging story gives a useful ex-ante rule but only one famous survivor. It does not show how often cheap tests failed, produced false positives, or were overridden. [researcher inference · strong] “Getting the Right Answers”
- The $250,000 figure may describe a stage rather than total capital. Without contracts, cap tables, and cash flows it cannot be combined with other public check figures. [researcher inference · strong] Genentech staging paragraph
- Anti-committee and top-down prescriptions can overcome conformity or suppress dissent. HP's successful examples do not establish the base rate of forceful but wrong leadership decisions. [researcher inference · strong] HP/committee discussion
- “Silicon Valley is a meritocracy” and “cronyism is almost non-existent” are broad participant claims. The referral-network description itself identifies access dependence; no demographic, allocation, or outcome evidence tests equality. [researcher inference · strong] networking Q&A
Open questions
- Locate event video, host notes, or a transcript that exposes exact question wording, speaker sequence, and omitted qualifications.
- Identify the two Genentech test providers, contracts, result dates, amounts, kill criteria, and approval sequence.
- Test the tenfold-improvement and corporate-VC claims across ordinary and failed investments, not only HP and Genentech winners.
- Compare the IPO prescription against actual offering sizes, pricing, secondary sales, post-IPO performance, and dilution across Perkins-led boards.
[w] Hewlett-Packard resignation and disclosure record
- URL: https://www.sec.gov/Archives/edgar/data/47217/000110465906059471/a06-19246_18k.htm | Published: corrective filing 2006-09-06; events from 2006-05-18 onward | Format: HP-authored Form 8-K, controlled by SEC Release 34-55801 | Reliability: A for issuer/regulatory record, with party positions distinguished | Access: full HTML and five-page SEC order; command-line requests may return 403 | Accessed: 2026-08-23 | Locator scheme: 8-K Item 8.01 paragraph openings; SEC order numbered paragraphs and PDF pages
Takeaways
- HP says Perkins supplied no written explanation at resignation; this is not a Perkins letter. [contemporaneous record · strong] 8-K Item 8.01, paragraph beginning “On May 22”
- The immediate dispute favored private handling: Perkins objected to bringing the leak matter to the full board and to seeking Keyworth's resignation. [contemporaneous record · strong] 8-K paragraph beginning “George A. Keyworth” SEC order ¶¶8–9
- The public record first documents Perkins learning enough about pretexting after resignation to request an inquiry; later privacy framing cannot be backfilled as the complete May rationale. [contemporaneous record · strong] 8-K paragraph beginning “On June 19”
- The SEC found HP's original resignation disclosure deficient; the respondent and cease-and-desist addressee was HP, not Perkins. [contemporaneous record · strong] SEC order ¶¶1, 12–15 and §IV
Quotes
- No extended quotation is required; the primary records are fully linked and cited by numbered paragraph or Item 8.01 paragraph opening.
Biography and decision evidence
- During roughly 90 minutes of deliberation Perkins argued for confidential/off-line handling and proportionality, then resigned after the board's secret-ballot decision. [documented behavior · strong] SEC order ¶¶8–9
- The SEC concluded that the disagreement concerned operations, policies, or practices and should have been disclosed. HP settled the administrative proceeding without admitting or denying findings except jurisdiction. [contemporaneous record · strong] SEC order preliminary section and ¶¶12–15
Philosophy signals
- confidential resolution: Perkins preferred private correction before public/full-board sanction in this case. [documented behavior · strong] SEC order ¶¶8–9
- proportionality: he weighed the perceived offense against Keyworth's prior contribution. [documented behavior · moderate] SEC order ¶8
- governance decomposition: investigation, board authority, director discipline, recusal, and issuer disclosure require separate rules. [researcher inference · strong] 8-K Item 8.01 SEC order ¶¶3–15
Contradictions, counterevidence, and access limits
- Perkins's preferred private process may protect trust or conceal misconduct; the record does not establish it was correct. SEC order ¶¶8–9
- His initial confirmation that he had no disagreement on HP operations/policies conflicts with the SEC's later finding that the dispute qualified as such for disclosure. 8-K Item 8.01 SEC order ¶¶12–15
- HP's 8-K reports party positions; the SEC order provides the controlling regulatory finding but is a settled order, not a trial verdict against Perkins. SEC order preliminary section and §IV
- Automated SEC requests returned 403 during one CLI check; canonical routes remained retrievable through browser access.
Open questions
- What did Perkins know about each investigative technique at resignation time?
- What do full board minutes and counsel memoranda show about authority, recusal, proportionality, and alternatives?
- Did his venture boards use comparable private-resolution and disclosure rules?
[w] Perkins's oral description of a purported “Investor's Bill of Rights”
- URL: https://digicoll.lib.berkeley.edu/record/218580/files/perkins_tom.pdf | Stated origin: circa 1972; described in interviews 2009 and transcript published 2010 | Format: oral description of a section in K&P's missing original business plan | Reliability: A for custody/attributed speech; B for memory; insufficient for the missing text and industry-causation claim | Access: full transcript, original plan unavailable | Accessed: 2026-08-23 | Locator scheme: printed pp. 32–35 (authorship/location/abuses/rule) and p. 40 (GPs as LPs/automatic participation)
Takeaways
- Perkins says he wrote the original K&P plan but did not know where it was; the public source is therefore his oral recollection, not an inspected 1972 artifact. [investor-stated · provisional] printed pp. 32–34
- He names abuses the rule was meant to prevent: preferential founder-stock purchases before fund entry, diversion of attractive opportunities, missing audits, and selective/potentially preferential GP co-investment. [investor-stated · moderate] printed pp. 34–35
- He says K&P prohibited personal investments in anything the fund might invest in and that the GPs invested as LPs, automatically participating in partnership deals. [investor-stated · moderate] printed pp. 34–35 and 40
Quotes
- No quotation is reproduced. The note uses page-stable paraphrase to respect the archive's publication-rights notice and repository policy; it does not claim short fair-use excerpts are categorically impermissible.
Biography and decision evidence
- The described rule concerns GP-LP allocation, pricing, audit, and personal-investment conflicts—not generic board veto or founder-control rights. [researcher inference · strong] printed pp. 34–35
- Perkins's automatic-participation statement supports alignment through personal LP capital but does not expose fees, carry, taxes, dilution, side arrangements, or later-fund practice. [researcher inference · strong] printed p. 40
Philosophy signals
- allocation: define eligible vehicles before deal quality becomes visible. [researcher inference · strong] printed pp. 34–35
- price and information: ban the GP from using privileged access to obtain better personal terms. [researcher inference · strong] printed pp. 34–35
- audit: reporting is an alignment mechanism, not administrative decoration. [researcher inference · moderate] printed pp. 34–35
Contradictions, counterevidence, and access limits
- No original plan, signed agreement, revision history, portfolio-wide compliance file, or enforcement record surfaced. printed pp. 32–35
- Perkins's claim that K&P “cleaned up” and set the industry standard is self-attribution without an adoption denominator. [investor-stated · insufficient public record] printed pp. 34–35
- The perimeter “anything the fund might invest in” is not operationally defined in the public record. printed pp. 34–35
- Alignment through GP capital does not eliminate allocation, fee, carry, governance, follow-on, or disclosure conflicts. printed p. 40
Open questions
- Where is the original 1972 plan, and what did the relevant partnership agreement actually require?
- Which investments or personal opportunities tested the rule, and who audited exceptions?
- Did later KPCB vehicles, SPVs, scouts, opportunity funds, and partner investments retain or revise the restriction?
[w] 2014 wealth letter, reported retraction, and weighted-vote remarks
- URL: https://www.newyorker.com/business/currency/tom-perkins-and-schadenfreude-in-silicon-valley | Published: WSJ letter 2014-01-24; New Yorker 2014-01-27; TIME 2014-02-14 | Format: investor-authored letter authenticated through contemporaneous reproductions, followed by attributed TV/event remarks | Reliability: B for reported direct fragments and context | Access: partial; original WSJ body not openly accessible | Accessed: 2026-08-23 | Locator scheme: New Yorker paragraphs beginning “On Saturday,” “The letter was controversial,” and “Soon after”; TIME paragraphs beginning “Later in the interview,” “Perkins' latest comments,” and “Perkins would later retract,” with event-video locator 1:03:00 reported by TIME
Takeaways
- Authenticated fragments show Perkins comparing criticism of wealthy Americans to Nazi persecution of Jews; he later called the central historical term a terrible choice while retaining the group-hatred premise. [investor-stated · strong] New Yorker opening paragraphs TIME final paragraphs
- At a later event he suggested taxpaying as a voting qualification and taxes paid as a vote multiplier. TIME reports he called it provocative and a reporter said he later denied seriousness, so context confidence is moderate. [investor-stated · moderate] TIME, event paragraphs and 1:03:00 locator
- The episode is counterevidence to a global-good-judgment thesis, not proof that Perkins lacked technical or board skill. [researcher inference · strong] New Yorker opening and firm-response paragraphs
Quotes
- The note does not reproduce the full letter. Short fragments remain at the linked paragraph-stable reproductions; paraphrase is sufficient for analysis.
Biography and decision evidence
- KPCB stated Perkins had not been involved in years and publicly rejected the views, protecting the contemporary-firm/retired-founder boundary. [documented behavior · strong] New Yorker paragraph beginning “Soon after”
- The reported retraction addresses the word choice but not the underlying class-hostility argument; weighted-vote seriousness remains ambiguous. [researcher inference · strong] TIME final paragraphs
Philosophy signals
- domain calibration: success in one domain does not license analogy in another. [researcher inference · strong] New Yorker opening and defense paragraphs
- power: tax-weighted voting, even as provocation, exposes a wealth-to-political-authority premise worth testing in governance analysis. [researcher inference · moderate] TIME, event paragraphs and 1:03:00 locator
- public-claim process: examine mechanism, proportionality, harmed parties, incentives, and less inflammatory alternatives. [researcher inference · moderate] New Yorker TIME
Contradictions, counterevidence, and access limits
- The original WSJ page is paywalled; this note uses authenticated fragments, not a complete-text claim. New Yorker opening paragraphs
- “Retraction” and “terrible word” come through contemporaneous reporting; this note does not upgrade them into an unqualified apology. TIME final paragraphs
- The remarks are public-judgment evidence, not fund-performance or securities-law evidence. New Yorker TIME
- A charitable warning about class hostility does not make the historical mechanisms proportionate. New Yorker opening and context paragraphs
Open questions
- Does a licensed original expose qualifications omitted from public reproductions?
- Does full Commonwealth Club video resolve the seriousness and audience context of the voting proposal beyond TIME's 1:03:00 locator?
- Did Perkins connect these political priors to LP, founder, employee, or public-board decision rights?
[w] *Valley Boy: The Education of Tom Perkins*
- URL: https://books.google.com/books/about/Valley_Boy.html?id=AwkVP65ZTiIC | Published: 2007 | Format: participant memoir; Google Books limited preview, publisher/audio metadata, and interviews about the book | Reliability: B for Perkins's attributed recollection; C for causal self-assessment | Access: partial | Accessed: 2026-08-23 | Locator scheme: no stable public preview-page inventory exposed; only metadata/scope claims use this route, while independently locatable Bancroft/CHM pages control repeated life claims
Takeaways
- The title and publisher description frame technical education, HP operating work, company formation, venture partnership, governance, and wealth as a single “education.” This is authenticated framing, not a content-complete analysis. [investor-stated · provisional] Google Books record publisher record
- Perkins's independently available oral histories repeat an operator-to-investor sequence: Doriot at printed pp. 5–7, HP at pp. 10–27, University Laboratories at CHM pp. 6–7, and the K&P plan at printed pp. 31–35. These are controls on self-conception, not memoir excerpts. [investor-stated · moderate] Bancroft/NVCA CHM, pp. 6–7
- The memoir can generate hypotheses about judgment but cannot provide a screening denominator, pass log, cap tables, board counterfactuals, LP cash flows, or independent causal proof. [researcher inference · strong] source-map denominator
Quotes
- No memoir quotation is reproduced: the full body and stable public page inventory were unavailable. This is a project-policy and reproducibility choice, not a claim that fair-use quotation is categorically barred.
Biography and decision evidence
- The 2007 publication follows Perkins's 2006 HP resignation and comes decades after Tandem and Genentech; selection and reputation management are material retrospective risks. [researcher inference · strong] publisher date HP 8-K
- University Laboratories is the strongest public operator bridge because Perkins describes actually forming and selling a technical business before K&P. The record does not isolate which lesson caused which later decision. [investor-stated · moderate] CHM, pp. 6–7
Philosophy signals
- operator apprenticeship: use firsthand building experience to formulate product and organization questions, but require domain relevance. [researcher inference · moderate] CHM, pp. 6–7
- active involvement: the K&P self-description emphasizes work beyond stock selection; performance impact remains unverified. [investor-stated · provisional] Bancroft/NVCA, pp. 33–34
- narrative control: a decision rule derived from memoir needs contemporaneous documents, an ordinary case, and a counterexample. [researcher inference · strong] limited preview Bancroft/NVCA
Contradictions, counterevidence, and access limits
- The lawful preview is incomplete; reviews and snippets are not treated as Perkins prose.
- The famous Fund I outcomes dominate later memory while public sources conflict on a 17/18-company denominator and disclose no net LP cash-flow schedule. source-map denominator
- Repetition by Perkins across book, Bancroft, CHM, CBS, and other interviews is consistency, not independent corroboration. Bancroft/NVCA CHM
- Founder, partner, scientist, operator, co-investor, LP, and later-partner roles remain separately attributed. SCOUT people map
Open questions
- Can a library-licensed full copy provide page-stable analysis under repository quotation policy?
- Which memoir propositions match contemporaneous HP, Tandem, Genentech, partnership, and board records?
- Does the full body treat ordinary investments, passes, write-offs, reserves, dilution, and attribution as deeply as the famous wins?