Ben Rosen

1. Snapshot and testable edge thesis

This is a testable process thesis, not a greatness verdict. Rosen himself described Lotus and Compaq as lucky opportunities and denied that they came from one integrated portfolio plan. [investor-stated · strong] CHM, printed p. 16

2. Core philosophy

Rosen's clearest contemporaneous hierarchy is people first, an unmet market need second, and a product capable of serving that need third. [investor-stated · strong] Practical Computing, printed p. 81 His compact formulation—“We invest primarily in people”—does not authorize founder worship: the same answer retains need and product as separate gates.

His pre-outcome writing adds a demand mechanism. VisiCalc could make a personal computer useful to business and thereby pull the hardware platform into a larger market. [contemporaneous record · strong] Rosen Electronics Letter reprint, scan pp. 1–2 His 1982 interview similarly joined applications, simplicity, lower price, and reach rather than assuming that improving chips create demand by themselves. [investor-stated · strong] Boston Computer Update, printed pp. 23–24

The synthesized operating system is therefore:

1. Build an information surface before choosing. [researcher inference · strong] Publish, convene, use products, and treat proposals as market intelligence; do not confuse privileged access with selection skill. CHM, printed pp. 7–10 and 25–26
2. Underwrite people, need, and product in that order, but require all three. [investor-stated · strong] The people test concerns learning and company-building capacity, not conventional polish. Practical Computing, printed p. 81
3. Name the adoption bottleneck and the narrow wedge. [researcher inference · strong] Compaq borrowed IBM compatibility while differentiating through portability; Rosen's later Tesla analogy makes the beachhead logic explicit. CHM, printed pp. 13–16 · Driving the Tesla Roadster, “Tesla's opportunity”
4. Earn governance authority without taking product authorship. [researcher inference · strong] Rosen favored early entry and active boards after Osborne, but said principal product features should come from entrepreneurs. Practical Computing, printed p. 81
5. Treat every forecast and intervention as fallible. [researcher inference · strong] Borland, Kindle, Rosen Motors, Eclipse, and DEC demand written alternatives, kill criteria, and post-outcome scoring. Four Million Transistors, printed p. 18 · CHM, printed pp. 26–30

3. Sourcing

Sourcing checklist. [researcher inference · strong] For each opportunity record: source and prior relationship; the pre-existing market map; first direct user evidence; what access reveals that a cold analyst could not know; conflicts; who independently verifies it; and the full accepted/rejected denominator. A referral or conference encounter never receives picking credit by itself.

4. Picking

Rosen's screen is a hierarchy, not a slogan.

Diligence protocol. [researcher inference · strong] Freeze the decision-time state; interview users and channel partners; test the product against alternatives; map standards, complements, suppliers, regulation, and incumbent response; assess the team against the next failure mode; model total capital through the next value-inflecting milestone; record dissent, authority, conflicts, and falsifiers; and decide what evidence licenses a small option versus a full commitment. decision labs

5. Deal and portfolio mechanics

Portfolio decision rule. [researcher inference · moderate] Concentrate only where domain knowledge, governance capacity, and reserve capacity coexist; treat follow-on, intervention, and exit as new decisions. The historical record supports the components but not a Rosen-authored universal allocation formula. portfolio forensics

6. Vehicle and incentive fit

The vehicle is part of the method, not background.

Vehicle-fit test. [researcher inference · strong] Before investing, write the legal owner, decision makers, duration, reserve budget, liquidity clock, compensation, conflict duty, geographic authority, stakeholder claims, and succession owner. Reject the investment if the proposed industrialization or governance burden exceeds what that vehicle can finance or staff.

7. Team attribution and work with founders

Ten-role attribution protocol. [researcher inference · strong] For every deal name the source, sponsor, diligence owner, approver, negotiator, board representative, recruiter, follow-on decision maker, exit influencer, and later narrator. Use unknown rather than letting the best-known partner absorb the role. deal-level rights map

Rosen's supported founder-service doctrine is bounded: help with credibility, financing access, recruiting, governance, and difficult decisions; leave primary product features to entrepreneurs. [investor-stated · strong] Practical Computing, printed p. 81 Compaq's CEO removals and DEC acquisition show that active governance can create entrenchment, succession, and strategic-error risk, so interventions need precommitted measures and independent directors. [researcher inference · strong] Wired, 1999 · CHM, printed pp. 28–31

8. Views on the craft

• Risk and failure. [investor-stated · strong] Rosen said sufficiently aggressive venture investing should produce failures. The usable meaning is not “losses prove courage”; it is to distinguish thesis risk from process error and evaluate both against a complete denominator. Practical Computing, printed p. 81
• Power law and fund size. [researcher inference · moderate] Rosen sought companies large enough to justify fixed investor work, and Lotus/Compaq dominate public paper-value evidence. No public statement supplies a target concentration or fund-size formula, and no cash-flow ledger proves the power law. CHM, printed p. 16 · Pareto stress test
• Access versus selection. [investor-stated · moderate] Rosen said knowledge and networks made luck more likely, while explicitly naming luck. A great investor should therefore enlarge the opportunity surface and then prove selection and value-add separately. CHM, printed pp. 15–16 and 25–26
• Governance. [investor-stated · strong] The board selects and supervises management; Rosen generally favored separating chair and CEO. His Capellas exception shows that he allowed an independent board to substitute for formal separation. CHM, printed pp. 30–31 · retirement email
• Exits. [researcher inference · insufficient public record] No general sell discipline survives. Leaving Lotus's board over an IBM/Compaq conflict is governance evidence, not an exit rule; IPO marks and acquisition prices cannot establish realized skill. CHM, printed pp. 15–18 · return controls
• Greatness. [researcher inference · moderate] A defensible great-investor standard would require repeatable pre-outcome judgment, correct role attribution, aligned vehicle, useful intervention, ordinary-case resilience, and cash returns after outlier removal. Rosen's public record proves selected judgment and historical influence more strongly than personal alpha.

9. Signature bets and decision process

The labs below preserve the decision freeze before the outcome. Complete the four questions—invest/pass, kill criteria, ownership/terms, and next-check milestones—before opening each reveal.

1. Compaq: reject the first product, preserve the team. [researcher inference · strong] Test the team separately from the disk-drive concept, then validate compatibility, channel, supplier, and IBM-response risk. decision lab
2. Lotus: unconventional founder, powerful product layer. [researcher inference · strong] Evaluate product imagination and user switching without forcing a generic executive template; price conflict and recruiting risk. decision lab
3. Ansa: search breadth before conviction. [researcher inference · moderate] Reconstruct the category map, candidate exclusions, customer need, release capital, and acquisition dependence before treating “about 100” as diligence proof. decision lab
4. Osborne: passive later-round failure. [researcher inference · strong] Price the lack of rights and demand sell-through, runway, inventory, and transition evidence before investing. failure lab
5. Rosen Motors: technology without a committed integrator. [researcher inference · strong] Gate scale on OEM commitment, cost-down, total capital, certification, and salvage rights. failure lab
6. Borland: visible false negative. [researcher inference · strong] Do not backfill the later winner; ask whether a cheap milestone option could test a technically credible but underdocumented founder. miss lab
7. Eclipse: reserve strategy before apparent scale. [researcher inference · strong] Model supplier, certification, backlog quality, unit cost, round-by-round dilution, and reserve capacity before accepting the headline market. failure lab

The recurring sequence is: map → freeze → test people/need/product → identify wedge and bottlenecks → assign roles and rights → model total capital and reserves → precommit milestones and kill criteria → reveal and score. [researcher inference · strong] This is an analyst reconstruction from the seven cases, not a surviving Rosen IC template.

10. Mental models and maxims

The field guide promotes every distinct supportable rule rather than a top-N list:

1. Turn research into a compounding information network. [researcher inference · strong] Research, conferences, relationships, and proposals can widen access and comparison; they do not by themselves prove picking skill. model
2. Back people first, an unmet need second, and the first product third. [investor-stated · strong] All three gates matter; Kapor, Borland, Eclipse, and Kindle keep the hierarchy falsifiable. model
3. Find the application that pulls the platform. [researcher inference · strong] Start with the purchasing workflow that makes new infrastructure useful, then test whether demand is broad enough to matter. model
4. Enter through a narrow wedge while preserving the ecosystem. [investor-stated · strong] Borrow a standard or complement to lower adoption cost while building capabilities the incumbent lacks. model
5. Underwrite the company's next product, not only today's hit. [investor-stated · strong] Product enthusiasm is insufficient unless the team and organization can create successors. model
6. Invest early enough to earn active governance. [researcher inference · strong] Secure information and intervention rights, leave product authorship to founders, and audit board-caused harm. model
7. Match leaders to the failure mode, then separate strategy from execution. [researcher inference · moderate] Select against the next operating constraint and specify the evidence that distinguishes a bad strategy from weak execution. model
8. Separate technical proof from commercial scale and total capital. [researcher inference · strong] Require the customer, integrator, manufacturing, unit-economic, and total-funding gates that Rosen Motors and Eclipse missed. model
9. Preserve options and update at evidence thresholds. [researcher inference · moderate] Use small tests where uncertainty is cheap to resolve, precommit what changes the decision, and treat follow-ons as new underwritings. model
10. Use denominators and forecast ledgers to reason under uncertainty. [researcher inference · moderate] Define exposure and alternatives, freeze predictions, and score misses rather than rewarding a merely directional story. model
11. Write conflict, authority, and attribution boundaries before acting. [researcher inference · strong] Identify the legal owner, duty, recusal, ten decision roles, and later narrator so access and fame cannot absorb credit. model

These are model-family labels for navigation. The field guide controls exact provenance, examples, exceptions, support grade, and modern translation; no reconstructed label should be read as Rosen's wording.

11. Evolution and contradictions

The durable belief is not a fixed sector call. [researcher inference · strong] It is that technical capability becomes valuable through people, applications, complements, distribution, and organizational execution. The unresolved contradiction is whether Rosen consistently converted that systems view into disciplined capital and governance decisions outside a few visible cases.

12. Failures and limits

The principal limits are incomplete opportunity and cash-flow denominators, survivor-selected narratives, retrospective memory, partner and founder misattribution, era dependence, unknown terms/reserves, and no measured comparison cohort for active governance. [researcher inference · strong] uncertainty ledger

13. Historical operating environment

• PC platform formation. [contemporaneous record · strong] IBM's August 1981 PC used off-the-shelf components and an open architecture; its growing software ecosystem made compatibility commercially useful while lowering hardware defensibility. IBM history
• Applications as scarce complements. [contemporaneous record · strong] Rosen identified VisiCalc before the famous outcomes and later framed VisiCalc and Lotus 1-2-3 as catalysts for successive PC generations. 1979 scan · Four Million Transistors, pp. 24–25
• A smaller venture market. [contemporaneous record · moderate] A 1981 SEC speech reported roughly $900 million raised by organized venture funds in 1980 versus $319 million in 1979 and common six-to-ten-year investment periods. These are period estimates, not a Sevin Rosen benchmark. SEC speech, June 12, 1981
• Semiconductor capital and cycle. [contemporaneous record · moderate] Dataquest recorded a weak 1981–82 market and foundry access that let some startups postpone fab capital; this contextualizes but does not prove the Cypress or Crystal thesis. Dataquest archive, printed pp. 114–16 and 127–32
• Liquidity and regulation. [researcher inference · moderate] IPO and acquisition markets mattered to finite venture pools, while analyst-security conflicts shaped Rosen's transition. No public benchmark, partnership agreement, or cohort supplies period-adjusted alpha. vehicle analysis

The correct historical control asks what a diligent investor could know at the decision date. [researcher inference · strong] Open IBM architecture, a VisiCalc demonstration, processor capability, supplier plans, and customer interviews qualify; later company dominance, revised datasets, and retrospective memories do not enter the frozen case unless explicitly labeled.

14. Modern VC translation (as of 2026-08-23)

15. Comparison with completed peers

Arthur Rock is the closest people-first, concentrated-governance pioneer; Pitch Johnson is the closest early technical-network and patient-board comparator; Bob Kagle tests whether service, partnership design, and a visible outlier in a later internet regime produce the same claims. All three profiles are complete. The comparison is process- and vehicle-matched, not a performance ranking. [researcher inference · strong] Underlying comparison anchors are Rock, HBR, “Belief in the business,” projections, and closing passages, Johnson, CHM oral history, PDF pp. 17–29, and Kagle, Kettering transcript, eBay narrative and 10/90 exchange.

16. The strongest case against greatness

Skeptical thesis. [researcher inference · strong] Rosen may have been a historically important analyst, connector, and director whose reputation reflects extraordinary access, the PC/software regime, exceptional founders, partner and syndicate work, and two early visible outliers more than repeatable personal investment alpha.

1. Luck and outliers. [investor-stated · strong] Rosen called Lotus and Compaq lucky; those two dominate disclosed early paper values, and removing them destroys the public economic ranking. CHM, p. 16 · top-outcome removal
2. Privileged access. [researcher inference · strong] Morgan Stanley, the newsletter, conferences, executive relationships, and a small early venture network prefiltered the opportunity set. Access may explain why he saw deals without proving superior selection. CHM, pp. 7–10 and 25–26
3. Team misattribution. [researcher inference · strong] Founders built Compaq and Lotus; L. J. led semiconductor work; Bayless led Landmark/Ciena; co-investors and executives supplied capital and governance. people map
4. Missing denominator and economics. [researcher inference · strong] Five of 23 named paths remain unresolved, 30–31 names in the March-1990 45-entity census remain unallocated, and 0/23 has full cash returns. denominator
5. Survivorship and narration. [researcher inference · strong] Famous successes, instructive failures, biographies, and Rosen's retrospective accounts dominate; repetition is correlated evidence, not independent validation. source-map coverage boundary
6. Era effect. [researcher inference · strong] Open PC architecture, scarce organized venture capital, favorable software leverage, public-market liquidity, and a young category were conditions Rosen did not create. historical context
7. Governance contradiction. [researcher inference · strong] Board control did not prevent Compaq's DEC mistake; CEO removals created criticism; separate-chair doctrine admitted an exception. Active governance can add or destroy value. governance criticism
8. Non-repeatable or weak economics. [researcher inference · strong] No audited fund or personal return record shows that the process beat an era-, access-, partner-, and outlier-adjusted comparator. return controls

The favorable case wins if original ordinary-case memos, source/referral logs, votes, intervention records, and full cash flows show Rosen-specific pre-outcome judgment and value-add across more than Lotus and Compaq after controlling for team, vehicle, vintage, access, and syndicate. The skeptical case wins if roles are mostly access/narration, ordinary outcomes are weak, cash value is concentrated in the two famous companies, or matched partners seeing the same opportunities performed similarly. [researcher inference · strong]

17. Analyst study guide, glossary, and go deeper

Recommended reading path

1. Read the chronological biography and write the edge thesis, mechanism, predicted unseen evidence, and falsifier without using company outcomes.
2. Read the 1979 writing, then the 1982 and 1986 interviews; separate contemporaneous rules from later memory.
3. Use the people and rights map to assign sourcing, diligence, approval, negotiation, board, recruiting, follow-on, exit, and narration before crediting Rosen.
4. Complete all seven no-hindsight labs in order: Compaq, Lotus, Ansa, Osborne, Rosen Motors, Borland, and Eclipse. Write your answer before each reveal.
5. Apply every mental model to one ordinary or adverse case, then re-read the skeptical thesis.
6. Finish with the review, especially its central-claim audit, denominator checks, and unresolved evidence.

Investor-specific glossary

Reusable checklists

Sourcing. [researcher inference · strong]

• Define the market map, names/aliases, date, source, and complete funnel.
• Separate access, referral, sponsorship, and selection credit.
• Record conflict, incentive, and independent-verification boundaries.
• Preserve passes and ordinary opportunities, not only winners.

Diligence and picking. [researcher inference · strong]

• Freeze team, need, product, alternatives, price, and unresolved facts.
• Test the team after adverse evidence and specify exception criteria.
• Observe the customer job, repeat use, switching cost, and purchase owner.
• Name standard, complement, wedge, incumbent response, and dependency.
• For industrial systems, require certification, integrator, yield, cost-down, and total capital.

Investment committee. [researcher inference · strong]

• Name source, sponsor, diligence owner, approver, negotiator, and dissent.
• State invest/pass, kill criteria, ownership/security/terms, and next-check milestones.
• Model dilution and fund results with and without the dominant outcome.
• Match check, reserves, duration, board capacity, and liquidity to the legal vehicle.
• Record a small-option case separately from a full-underwriting case.

Governance and follow-on. [researcher inference · strong]

• Write board versus founder product rights, CEO review, recusals, and crisis authority.
• Assign recruiting, operating metric, financing, and succession owners.
• Re-underwrite each follow-on against milestone change and reserve opportunity cost.
• Preserve contemporaneous acquisition/integration assumptions and stop criteria.
• Audit whether the intervention helped, was neutral, or harmed the company.

Go deeper: ranked source path

1. Practical Computing, printed pp. 80–81 — best contemporaneous direct process statement. [researcher inference · strong]
2. 1979 VisiCalc reaction — strongest surviving pre-outcome category insight. [researcher inference · strong]
3. CHM oral history — deepest direct decision record, with memory and self-narration controls. [researcher inference · strong]
4. Boston Computer Update, printed pp. 23–24 — best early market forecast and sourcing-transition evidence. [researcher inference · strong]
5. L. J. Sevin oral history, printed pp. 20–25 and Kapor oral history, printed pp. 11–13 — essential partner/founder attribution controls. [researcher inference · strong]
6. SEC issuer filings — strongest ownership, board, and transaction controls; incomplete for full returns. [researcher inference · strong]
7. LA Times Rosen Motors closure — best contemporaneous commercial-failure control. [researcher inference · strong]

The package's unresolved record remains material: partnership agreements, LP reports, complete 45-company and lifetime lists, screening/pass logs, cash flows, original Compaq/Lotus/Ansa memos, board minutes, and disposition records could change the edge, economics, and governance conclusions. [researcher inference · strong] retrieval ledger

Pareto 80/20 — what drove the record

The public record supports concentration, but not a literal “80% of value from 20% of deals” calculation. There is no audited cash-flow ledger, complete company denominator, or LP-level DPI/TVPI/IRR. The useful Pareto result is therefore ordinal: two visible Fund I outcomes dominate the disclosed paper-value evidence, while a different set of failures and misses dominates what can be learned. [researcher inference · strong] TIME's contemporaneous paper-value report · searched-denominator reconciliation

Economic ranking — only what public evidence can support

Top-outcome removal test. Remove Lotus and Compaq and the surviving public evidence no longer supports an economic ranking, a positive aggregate return, or even a signed net cash result. It still supports multiple firm successes, but not their cost basis, dilution, realizations, write-offs, or opportunity cost. [researcher inference · strong] candidate ledger · return controls

Historical-significance ranking — category impact, not return

Learning-value ranking — failures and misses included

Evidence boundary and searched denominator

The source map began with 21 named opportunity/company paths. A DEALS-specific pass over acquired first-party and contemporaneous sources added Citrix (Rosen's post says Sevin Rosen financed it) and Capstone Turbine (the prospectus identifies both Ben/Harold investment and separate Sevin Rosen ownership), producing a deduplicated 23-path public candidate set. One company/opportunity is one path: follow-on rounds remain inside that path; passes, personal investments, family projects and firm investments stay in the candidate set but retain separate vehicle labels; Capstone's personal/family and firm interests remain one dual-vehicle company path; and the eight non-comparable controls are excluded rather than added. [researcher inference · strong] Eclipse post · Capstone prospectus, Product Development milestone table, printed p. 30; director biographies, printed p. 49; Principal Shareholders table and notes (5), (6), (11), printed pp. 60–62

The 18/23 (78%) coverage fraction applies only to the named public candidate set. It is not a firm-portfolio coverage ratio. Inc. reported 45 financed entities by March 1990, while Caltech says the firm supplied initial financing to more than 100 companies over its life; neither source provides a deduplicated company list, and the 23 paths also include passes and personal projects. The true firm-investment coverage fraction is therefore not computable. [researcher inference · moderate] Inc., 1990 · Caltech biography

The March-1990 firm denominator must be reconciled independently of the 23-path opportunity set. Fourteen names are confirmed or plausibly contemporaneous members: Compaq, Lotus, Cypress, General Parametrics, Acuson, Convex, Silicon Graphics, ProNet, Landmark Graphics, Electronic Arts, Synapse, Osborne, Ansa and Crystal Semiconductor. Citrix's financing cutoff is unresolved. Nine named paths or controls are necessarily outside the financed-entity census: Apple and Borland were passes; Eclipse, Ask Jeeves and Energy Vault were later personal investments; Ciena was a later firm investment; Capstone combined later Ben/Harold investment with separate firm ownership; Rosen Motors was a family operating company; and LSI Logic is an anti-attribution control. [researcher inference · strong] Inc., 1990 · Dataquest archive · Capstone prospectus, Product Development milestone table, printed p. 30; director biographies, printed p. 49; Principal Shareholders table and notes (5), (6), (11), printed pp. 60–62

Selection bias is severe: the named set is drawn from biographies, interviews, a success-oriented firm history, litigation, and Rosen's own later writing. Famous winners and narratively useful failures are more likely to survive than ordinary losses, small exits and rejected opportunities. No hit rate or greatness claim should be inferred from 18/23. [researcher inference · strong]

Historical market, regulatory, and base-rate context

• The platform inflection was visible but unsettled. IBM introduced the PC in August 1981 using off-the-shelf components and an open architecture; within a year, its ecosystem had hundreds of software packages. Compatibility could therefore be an adoption strategy, but it also invited clones and compressed durable hardware differentiation. [contemporaneous record · strong] IBM history
• Venture capital was expanding and still exit-dependent. A June 1981 SEC speech described organized venture investments as commonly lasting six to ten years, noted that public equity and secondary markets supplied liquidity, and reported about $900 million raised by organized venture funds in 1980 versus $319 million in 1979. These are period estimates, not Sevin Rosen fund data. [contemporaneous record · moderate] SEC speech, June 12, 1981
• Semiconductor startups faced a cyclical and capital-cost problem. Dataquest's historical directory records a weak 1981–82 market and a wave of 1977–85 startups; foundry access allowed some entrants to postpone fab capital. This contextualizes, but does not prove, the Cypress and Crystal theses. [contemporaneous record · moderate] Dataquest archive
• The relevant base rate is missing. There is no surviving Rosen screening log, rejected-deal list, fund write-off schedule, ownership ledger or vintage benchmark in the public corpus. Rosen's recollection that the first two investments failed and that he reviewed roughly one in fifty proposals cannot substitute for a cohort. [investor-stated · moderate] CHM, pp. 12–13 · TIME, January 23, 1984

Vehicle economics and incentives

The incentive-sensitive conclusion is narrow: Fund I's small size and bilateral veto favored concentrated partnership judgment; later growth and succession distributed selection authority; personal projects allowed long-duration technical conviction without transparent institutional portfolio controls. No public evidence supports a claim about carry-driven behavior or fund-level performance. [researcher inference · moderate] CHM, pp. 11–24 · Inc., 1990 · Ask Jeeves prospectus, printed pp. 60–65 · Eclipse post

People, capital, and influence map

Deal-level decision-rights map

Material-case ten-role matrix

“Source” means who originated the opportunity, not the documentary source. unknown means the opened public record does not establish the role; firm association never fills it.

Comprehensive named candidate ledger

“Win,” “loss,” “pass” and “unresolved” describe the evidenced company path, not an audited fund result. Every row uses the same fields; explicit unknowns are evidence limits, not omissions.

Excluded and non-comparable controls

No named post-1995 successor-firm deal in the acquired corpus is assigned to Ben. That negative boundary prevents the successor firm's later portfolio from being folded into his record. [researcher inference · strong]

Return and economics controls

No IRR is calculated. Paper value is not cash realization; company financing is not investor cost; acquisition price is not proceeds; IPO size is not gain; company market value is not fund NAV; and firm economics are not Rosen's personal economics. [researcher inference · strong]

Ex-ante case studies and no-hindsight decision labs

Each lab freezes the evidence at the stated decision point. The prompts are analyst-created exercises, not Rosen quotations. Outcomes appear only after the exercise.

Case 1 — Compaq: reject the first product, preserve the team

Decision freeze — Houston, 1981

Three experienced Texas Instruments managers first propose a disk-drive company. IBM has just launched a PC using open architecture, but the future dominance of its standard is unproved. The team returns with a portable computer intended to be completely IBM-compatible. Rosen believes partial compatibility would be commercially weak; the product must win through portability while borrowing the new ecosystem. [investor-stated · strong] CHM, pp. 13–16 · IBM history

Analyst-created exercise

[researcher inference · strong] Underwrite the team separately from its first proposal. Test BIOS/legal implementation, application compatibility, bill of materials, dealer willingness and an IBM-response scenario before treating “compatible” as a moat.

• Would you invest? Yes, conditionally, if independent compatibility tests and dealer interviews validate the wedge.
• What would kill the deal? Material application failures, dependence on IBM-controlled components without alternatives, or no channel willingness to stock a new brand.
• What ownership/terms would you require? A board seat, milestone-tranched capital, information rights and enough pro rata capacity to support the launch; exact percentage cannot be set without valuation and round data.
• What next-check evidence would change the decision? A working prototype against a defined application suite, supplier quotes, dealer letters and a bottom-up launch cash plan.

Actual decision and outcome

The investors rejected the disk-drive concept but funded the portable-compatible concept. Ben became chair; Kleiner Perkins and others joined the syndicate. Compaq reported $111 million of first-year sales and completed a 1983 IPO, but the public record does not supply complete Sevin Rosen cash returns. [documented behavior · strong] Kleiner Perkins · SEC proxy

Case 2 — Lotus: unconventional founder, powerful product layer

Decision freeze — 1982

Mitch Kapor has a pre-existing relationship with Rosen and unusual product imagination, but does not fit a conventional executive template. The PC installed base is growing; a business-productivity layer could create leverage above hardware. No public decision memo supplies price, ownership, burn or competitor analysis. [retrospective witness · strong] Kapor oral history, pp. 11–13

Analyst-created exercise

[researcher inference · strong] Evaluate the product insight and founder learning rate rather than force a generic CEO archetype. Separate product-market evidence from the investor's ability to confer legitimacy.

• Would you invest? Yes, if users switch from incumbent tools and the product has a credible route to distribution.
• What would kill the deal? Weak repeat usage, dependence on one hardware vendor without portability, or a founder unwilling to recruit complementary operating talent.
• What ownership/terms would you require? Board representation, pro rata rights and explicit conflict management because Rosen also chairs Compaq.
• What next-check evidence would change the decision? Cohort usage, willingness-to-pay, performance benchmarks, dealer/software-channel interviews and a recruiting plan.

Actual decision and outcome

Sevin Rosen backed Lotus; Rosen served as founding director and later left the board when IBM-related discussions conflicted with Compaq. Kapor credited him principally with credibility and access. TIME's IPO paper-value figure is not a realized-return record. [documented behavior · strong] CHM, pp. 15–18 · TIME

Case 3 — Ansa: search breadth before conviction

Decision freeze — mid-1980s database market

Rosen says he evaluated roughly 100 database products before selecting Ansa and becoming non-executive chair. The surviving Q&A gives a decision hierarchy—people, need, product—but not unit economics, valuation, security, ownership or comparative scorecard. [investor-stated · strong] Practical Computing, pp. 80–81

Analyst-created exercise

[researcher inference · moderate] Rebuild the category map rather than accept “100” as proof of diligence. Compare target users, query model, hardware requirements, distribution, performance and switching cost across a representative sample.

• Would you invest? Only if Ansa is top-decile on a predefined user problem and has a credible path to revenue before the next round.
• What would kill the deal? No differentiated user need, benchmark inferiority, or a funding plan dependent on an acquisition.
• What ownership/terms would you require? Board seat, budget approval rights, pro rata participation and milestones tied to product release and paid adoption.
• What next-check evidence would change the decision? Blind product tests, reference calls, sales pipeline conversion, burn-to-release and a competitor matrix preserved before investment.

Actual decision and outcome

Rosen invested and chaired. Borland agreed to acquire the loss-making company in 1987 for a reported amount near $38 million. Without stake and proceeds, this is a strategic exit, not a verified win multiple. [contemporaneous record · moderate] Tech Monitor

Case 4 — Osborne: the passive later-round failure

Decision freeze — second financing round

Osborne is already operating when Rosen enters a later round. He has neither board representation nor effective control. The public record does not preserve valuation, liquidity runway, channel inventory, product roadmap or the syndicate's governance. [investor-stated · strong] Practical Computing, p. 81

Analyst-created exercise

[researcher inference · strong] Treat absence of rights as an underwriting fact, not a post-failure excuse. Price the inability to intervene and require operating evidence that makes passive ownership tolerable.

• Would you invest? No on the surviving facts; reconsider only with sufficient downside price protection and reliable operating data.
• What would kill the deal? Short runway, inventory/channel deterioration, a product-transition gap or refusal to grant information and governance rights.
• What ownership/terms would you require? At minimum information rights, protective provisions and a board observer; a board seat if the check is material.
• What next-check evidence would change the decision? Weekly sell-through, returns, cash conversion, product-transition schedule and downside liquidation analysis.

Actual decision and outcome

Rosen reported losing the full $400,000. Inc. uses first-person loss language while TIME attributes $400,000 to the fund, so the capital owner remains unresolved. He later made active earlier-stage board participation part of his lesson. The causal counterfactual remains unproved: board control might not have rescued Osborne. [investor-stated · strong] Inc., 1986 · TIME, January 23, 1984 · Inc., 1990

Case 5 — Rosen Motors: technology without a committed integrator

Decision freeze — 1993

Ben and Harold Rosen pursue a turbine-flywheel hybrid drivetrain. The architecture promises performance and efficiency but must be integrated by an automaker, reach automotive cost and reliability, and survive a long capital cycle. No signed OEM program or full capital-to-production budget is in the public record. [investor-stated · moderate] Car of the Future post

Analyst-created exercise

[researcher inference · strong] Underwrite the customer and integration path before the prototype. Make an OEM commitment, cost-down curve and total funding plan gates to scale.

• Would you invest? Only a staged technical-validation amount before OEM evidence; not the full commercialization program.
• What would kill the deal? No paid OEM pilot by a fixed date, component cost above the adoption envelope, or funding needs beyond an agreed ceiling.
• What ownership/terms would you require? Milestone tranches, independent technical review, board control over budget expansion and clear IP ownership/salvage rights.
• What next-check evidence would change the decision? OEM engineering commitment, duty-cycle data, reliability testing, production quotes and a financing plan through certification and launch.

Actual decision and outcome

After nearly five years, about $24 million and 70 employees, Rosen Motors closed without an automaker commitment. Rosen characterized the work as technically but not commercially successful. Capstone, which had been adjacent since the 1993 Ben/Harold investment, bought $590,000 of Rosen Motors equipment and improvements in 1997 and assumed leases; that transaction and later Pentadyne licensing do not establish total recovery. [documented behavior · strong] LA Times · Capstone prospectus, Certain Transactions, printed p. 59 · first-person post

Case 6 — Borland: the visible false negative

Decision freeze — before Borland's founding

Philippe Kahn seeks venture backing. Rosen later says every approached VC, including him, passed; Kahn then founded Borland with $5,000. The surviving report supplies neither pitch nor rejection rationale, so market, product, founder and price cannot be reconstructed without hindsight. [investor-stated · strong] Caltech report, p. 18

Analyst-created exercise

[researcher inference · strong] Do not reverse-engineer a “yes” from Borland's later success. Build a small-option path for a technically credible but unconventional founder when core uncertainty can be tested cheaply.

• Would you invest? Insufficient evidence for a full investment; consider a small milestone option only if a product demonstration is compelling.
• What would kill the deal? No usable product, no identifiable customer, founder integrity concerns or an untestable capital plan.
• What ownership/terms would you require? A simple seed security with pro rata rights; governance proportional to the small check, not control by default.
• What next-check evidence would change the decision? Prototype, early paid users, distribution experiment and the original rejection notes.

Actual decision and outcome

Rosen passed; Kahn founded Borland, and Rosen later became a director. The later role is evidence of updating or relationship, not proof of original foresight or later investment. [documented behavior · strong] Caltech report, p. 18

Case 7 — Eclipse: reserve strategy before apparent scale

Decision freeze — early personal round

Eclipse proposes a low-cost very-light jet and a linked air-taxi market. Rosen invests a modest personal amount. Aircraft certification, engine supply, manufacturing scale, customer financing and repeated follow-on rounds all sit between prototype and sustainable economics. [investor-stated · strong] Eclipse post

Analyst-created exercise

[researcher inference · strong] Model total capital and reserve behavior before accepting the headline market. A small first check without a follow-on rule is not exposure to the ultimate outcome.

• Would you invest? Only with a financing-to-certification model and a precommitted reserve rule; otherwise pass.
• What would kill the deal? Engine/supplier delay beyond runway, gross margin requiring implausible volume, refundable rather than firm backlog, or no financing path to certification.
• What ownership/terms would you require? Pro rata rights, information rights, milestone tranches and governance around supplier and liquidity risk.
• What next-check evidence would change the decision? Certification schedule, supplier guarantees, audited backlog quality, unit-cost curve and round-by-round dilution model.

Actual decision and outcome

Rosen did not follow on, expected severe dilution and later recorded that Eclipse ran out of money and entered bankruptcy. The case falsifies the equation of ambitious engineering, backlog or capital raised with durable value. [investor-stated · strong] Eclipse post · bankruptcy follow-up

Portfolio-level forensics

The most defensible portfolio mechanism is specialized partnership coverage plus active governance, not a single Ben Rosen formula. Ben's PC/software and capital-markets access, L.J.'s semiconductor operating depth, younger partners' deal leadership and outside syndicates produced the visible set. [researcher inference · strong] Ben Rosen CHM, pp. 11–24 · L.J. Sevin CHM, pp. 20–25 · Dataquest · Kleiner Perkins on Compaq The economic claim remains weaker than the historical one because attribution is better documented than cash flows.

Failure, miss, and adverse-evidence postmortems

Governance criticism, litigation, and contradictory evidence

• Rosen's philosophy favored separating chair and CEO, yet Compaq later combined the roles under Michael Capellas. Rosen justified the exception by pointing to an independent board. The episode shows a principle with contextual exceptions, not a mechanically applied rule. [investor-stated · strong] retirement email
• Rosen participated in the removals of Rod Canion in 1991 and Eckhard Pfeiffer in 1999. Contemporary critics questioned whether his influence complicated CEO recruitment; Pfeiffer later argued the abrupt removal hurt the company. These are relevant adverse accounts, not adjudicated causality. [retrospective witness · moderate] Wired, 1999 · Computerworld
• Rosen was a named defendant in Berger v. Compaq, concerning alleged misleading 1997–98 disclosures and channel inventory practices. The Fifth Circuit reversed class certification for legal error and remanded; it did not decide the allegations' merits. [contemporaneous record · strong] Fifth Circuit opinion
• HP later disclosed a Compaq-related securities settlement substantially covered by insurance. That company settlement is not a finding of Rosen's personal wrongdoing. [contemporaneous record · strong] HP 2002 10-K
• No Rosen-specific SEC enforcement action or substantiated LP suit was found in the reviewed public routes. This is a bounded retrieval result, not legal clearance. [researcher inference · moderate] adverse-source map

Uncertainty and retrieval ledger

Three terminal search passes—company-name plus Rosen/Sevin, company-name plus board/financing, and company-name plus filing/outcome/adverse terms—produced no new named Ben-attributable candidate beyond Citrix and Capstone and no additional audited economics. The search is saturated for this public web run, not historically exhaustive. [researcher inference · moderate]

Continue with the source map, profile, writings, talks, and the case notes in notes/.

Corpus definition and reconciliation

The public written corpus is bounded rather than assumed complete. The first-party RSS feed contains 43 items: 41 substantive Ben-authored posts plus two administrative notices, from November 2007 through December 2009. Thirteen concern investing, technology, company operations, institutions, Rosen's biography or decision reasoning closely enough for deep processing; the other 28 substantive posts remain routed rather than silently omitted. first-party RSS archive

Four other usable writing families are separated: the 1979 VisiCalc letter scan; the 2011 “Memories of Steve” post; the 2020 essay; and the January 1983 Rosen-branded newsletter. The last is organizational evidence only because no Ben Rosen byline was verified. Four archival newsletter leads remain partial or unavailable. [researcher inference · strong] annotated source map

Source-by-source analysis

1979 — Rosen Electronics Letter reaction to VisiCalc

The two-page scan is the most important surviving pre-outcome writing. Rosen sees that a spreadsheet can make the personal computer useful to business users and may pull hardware demand behind it. [contemporaneous record · strong] This is stronger than a later “saw the PC revolution” recollection because it freezes the software-as-demand-driver thesis before Lotus or Compaq. letter scan · source note

Its limitation is equally important: recognizing VisiCalc did not itself specify the later IBM-compatible hardware wedge, the Lotus team, price, ownership or exit. It demonstrates category observation, not a complete venture decision.

1983 — Rosen-branded newsletter after the editorial transition

The January 1983 issue shows a continuing research product organized around the IBM PC and networking. It is not treated as Ben-authored: Rosen had sold the publication to Esther Dyson, the issue is unbylined, and brand continuity is not authorship. [contemporaneous record · strong] source note

2008 — I Coulda Been a Contenda

Rosen recalls building a searchable restaurant application on an Apple II and distributing printouts inside Morgan Stanley. [investor-stated · moderate] It shows prototype-first curiosity and limited observed utility, but the later Zagat comparison is humorous hindsight: there was no demonstrated data-acquisition, distribution, monetization or founder-commitment advantage. post · source note

2008 — Eclipse Takes Off

Rosen describes Eclipse Aviation's linked aircraft-manufacturing and air-taxi bets, more than $1 billion of capital, an engine delay, scale-dependent gross margins and roughly 1,800 employees. [investor-stated · moderate] The strongest portfolio lesson is disclosed dilution after Rosen declined follow-ons: early access is not retained economic exposure. Eclipse's later bankruptcy is the mandatory outcome control. post · source note

2008 — Potpourri

Rosen treats the Segway as impressive engineering for a weak customer problem, proposes field evidence to decide between electric-vehicle architectures, and records founder-board conflict at Eclipse. [investor-stated · moderate] His initial Kindle skepticism is a scored counterexample: a good “problem, not technology” heuristic can still misread a changing product. post · source note

2008 — Reflections

Rosen challenges equilibrium assumptions with models of interacting expectations and challenges football convention using conversion and yardage base rates. [investor-stated · moderate] Both moves ask whether inherited rules survive measurement, while raw averages still require causal controls. post · source note

2008 — On... a Lot of Topics

Rosen audits a presidential-election forecast, judges policymakers partly by pre-crisis foresight, distinguishes Tesla builders from commentators, and revisits GM's rejection of Rosen Motors. [investor-stated · moderate] The Eclipse bankruptcy discussed in the same piece checks the temptation to equate risk-taking with value creation; the GM story validates a category direction, not Rosen's specific architecture. post · source note

2007 — Car of the Future

Rosen revisits Rosen Motors and the larger transition to electric propulsion. The useful investment signal is not technological triumphalism but the mismatch between a demonstrable subsystem and an automotive-scale route to market. [investor-stated · moderate] His retrospective can underweight selection effects and the financing/channel constraints documented when the company closed. post · source note · LA Times failure control

2008 — The Merger That Worked

Rosen argues that Compaq and HP possessed complementary products, distribution and scale, and that execution—not combination logic—was the principal risk. [investor-stated · moderate] The post includes a useful disclosure that he did not participate in the transaction and owned no HP stock, but its stock-price comparison is endpoint-sensitive and does not isolate merger causality. post · source note

2008 — Science, Technology and America's Future

The post applies the same ecosystem lens at national scale: universities, research laboratories, engineering talent, capital and policy jointly shape innovation capacity. [investor-stated · moderate] The lesson is a constraint map, not a forecast of individual-company returns. International degree-count comparisons and institutional decline claims need dated source control. post · source note

2008 — Met Opera — A New Act

Rosen describes a turnaround through product improvement, marketing and new revenue/audience channels. [investor-stated · moderate] It is relevant because it reveals an operating decomposition outside technology: improve the offering, increase awareness and expand distribution/monetization. An arts institution is not a startup control, and Rosen's management role creates an interested-narrator limit. post · source note

2008 — Driving the Tesla Roadster

The post begins with firsthand product use and links delight to broader electric-vehicle potential. [investor-stated · moderate] It is a valuable match to Rosen's habit of learning through use, but also exposes sample bias: an affluent early adopter's experience does not establish mass-market cost, service, manufacturing or charging readiness. post · source note

2009 — Fear of Flying

Rosen contrasts vivid accident salience with transportation base rates. [investor-stated · moderate] The portable lesson is to compare risks against a denominator, not against emotional availability. The post is not venture analysis and its chosen statistics must remain tied to the publication date. post · source note

2009 — It's Time for a Change

The post criticizes leverage, complexity, incentives and opacity in crisis policy. [investor-stated · moderate] For investing, it suggests a system-risk checklist: trace leverage, payoff asymmetry, transparency and unintended consequences before accepting an elegant rescue mechanism. It does not document Rosen applying this checklist to his venture portfolio. post · source note

2009 — One Mistake of the First 100 Days

Rosen evaluates public appointments through prior evidence of understanding systemic risk, capital deterioration, financial innovation, leverage and instability. [investor-stated · moderate] The underlying investment principle is a role-specific evidence test: select leaders by demonstrated comprehension of the failure mode they must manage. Political judgment and venture hiring remain distinct contexts. post · source note

2011 — Memories of Steve

Rosen recounts showing the Apple II inside Morgan Stanley, recognizing VisiCalc's importance, declining an Apple security because of conflict, and later corresponding with Jobs. [investor-stated · moderate] The narrative connects access, use and ethics, but memorial writing is highly selective and should be read beside the 1979 scan and Morgan Stanley's institutional history. post · source note

2020 — A Miserable Old World, A Better New World

The late essay makes a broad systems forecast about crises accelerating institutional and technological change. [investor-stated · provisional] Its value is as a dated prediction ledger, not as evidence that Rosen's earlier venture selections followed the same framework. Outcomes should be scored against explicit propositions, time horizons and alternatives rather than narrative plausibility. essay · source note

Detailed source worksheets

Cross-source synthesis

1. Product use becomes valuable only when it reveals a purchasing job. Rosen's 1979 VisiCalc reaction identifies application software as demand-pull for PCs; his Tesla post again starts from firsthand use, but a luxury early-adopter experience does not establish mass-market cost or serviceability. [researcher inference · strong] VisiCalc scan · Tesla post
2. Adoption is a complement system, not a stand-alone product score. Software pulled PC hardware; Rosen judged HP/Compaq through product, geography, channel and scale; his Met Opera account joins offering, marketing and distribution. This is a bottleneck checklist, not proof any one complement caused an outcome. [researcher inference · moderate] merger post · Met post
3. Technical proof and commercial scalability are separate gates. Rosen's EV writing recognizes that automotive incumbents excel at high-volume, low-cost manufacturing; Rosen Motors' closure supplies the adverse case—reported $24 million spent, 70 employees and no automaker commitment. [researcher inference · strong] Car of the Future · LA Times closure
4. Risk should be decomposed into denominator, leverage, incentives and opacity. “Fear of Flying” attacks salience without a base rate; the crisis posts ask whether leaders recognized leverage, deteriorating capital and unstable incentives before failure. These are public-policy arguments, not evidence Rosen used the checklist in an IC. [researcher inference · moderate] Fear of Flying · One Mistake
5. Role and conflict boundaries are part of the decision. Rosen says he declined private Apple stock while covering Apple and explicitly says he neither participated in HP/Compaq nor owned HP stock when defending the merger. The first is a conduct example; the second limits his authority to retrospective commentary. [documented behavior · moderate] Memories of Steve · merger disclosure

The tensions remain visible: firsthand use generated insight but Rosen later admitted his preferences could have led him to miss eBay; technical optimism coexists with Rosen Motors' commercial failure; base-rate skepticism coexists with broad 2020 optimism; and the favorable merger narrative is ex post and endpoint-sensitive. [researcher inference · strong] CHM, pp. 25–27 · 2020 essay

Chronology, evolution, and contradictions

Decision playbook derived from the corpus

This is a researcher reconstruction, not a claimed Rosen checklist. Osborne, Rosen Motors, the DEC acquisition, Eclipse dilution, Kindle skepticism and admitted sample bias are mandatory adverse tests.

Modern translation as of 2026-08-23

• AI applications. [researcher inference · provisional; analyst-created scenario] For a hypothetical coding assistant, transfer the workflow/adoption-unlock test. Assume model capability and price can change quickly while data rights, inference cost and bundling dominate. Signals: paid repeat use, retention after model changes, workflow displacement and gross margin after inference. Misuse: calling a benchmark or pilot product-market fit. The example is deliberately hypothetical, not a claim about a named 2026 company. historical analogue
• AI infrastructure. [researcher inference · provisional] Transfer compatibility and complement mapping. Current example: CoreWeave's Q2 2026 filing describes customer concentration, NVIDIA-dependent GPUs, substantial capex and future financing needs. Signals: multi-customer utilization, contracted power, financing headroom and portability. Misuse: treating incumbent compatibility as protection from incumbent response. CoreWeave Q2 2026 10-Q · Compaq analogue
• Climate and industrial technology. [researcher inference · strong] Require industrialization, certification, channel/offtake and total-capital plans. Current example: Rivian's Q2 2026 filing reports Volkswagen milestone funding while retaining JV, supplier, factory and additional-capital risks. Signals: yield, unit cost, homologation and binding milestone funding. Misuse: treating an OEM logo as guaranteed distribution. Rivian Q2 2026 10-Q · Rosen Motors
• Fintech. [researcher inference · moderate] Map leverage, incentives, custody and record ownership through every intermediary. The FDIC's proposed custodial-account recordkeeping rule is a current public example of ledger/beneficial-owner risk across intermediaries. Signals: reconciliation, beneficial-owner records, direct bank contracts and failure playbooks. Misuse: treating a partner-bank logo as end-to-end safety. Federal Register, RIN 3064-AG07 · historical crisis post
• Boards and portfolio conflicts. [researcher inference · moderate; analyst-created scenario] For a hypothetical 2026 investor holding competing AI application seats, pre-agree information rights, recusal triggers, CEO-review cadence and crisis authority. Signals: written recusals, independent committees and rehearsed succession. Misuse: counting seats as value-add without interventions and counterfactuals. The example is hypothetical; Rosen's Apple conduct is the historical control. Apple conflict

Five highest-value writings

1. 1979 VisiCalc reaction — best contemporaneous category observation. [researcher inference · strong]
2. The Merger That Worked — best complement-versus-execution framework and explicit disclosure. [researcher inference · strong]
3. Car of the Future — strongest prototype, capital and channel tension when paired with failure. [researcher inference · strong]
4. Driving the Tesla Roadster — firsthand product judgment constrained by manufacturing economics. [researcher inference · moderate]
5. One Mistake of the First 100 Days — sharpest evidence-based leadership and systemic-risk test. [researcher inference · moderate]

Access, attribution, and prediction gaps

• The 1973–82 newsletter archive is fragmentary: the March 31, 1980 “Quality” item survives only as OTA quotation/metadata; the July 14/15, 1980 Semiconductor Forum only as a JEC derivative citation; the August 24, 1981 IBM PC issue only as catalog metadata; and the June/November 1981 PC items only as museum paraphrase without scans.
• A Rosen-branded issue after the Dyson transition is not Ben-authored without a byline or editorial record.
• Blog statistics and broad forecasts are time-bound and have no externally scored denominator.
• No private investment memo, IC record or LP letter links the public writing directly to allocation decisions.
• The 43-item feed denominator is complete only for the preserved 2007–09 first-party export; it contains 41 substantive posts and two admin notices, not Rosen's lifetime writing.
• The corpus supports a testable process more strongly than repeatable personal economic alpha.

Routed first-party blog ledger

The routing rule excludes a post from deep investment analysis when it contains no direct investment, fund or company-board decision and is unnecessary to support a claimed investing rule. The 28 routed substantive titles are: Healthcare, Academia, Perps; Three Visual Treats; Inaugurating a President; Inaugurating a New Year; Archeos and Elyn Zimmerman; Election Prediction; Letter from London; Tabula Rasa No. 1—A Conversation with Edmund Morris; North Korea and the N.Y. Philharmonic: The “Why?” Question Gets Answered; Revisiting “The Gates”; Dinner with Barack; A Good-News Katrina Story; Spin Me to the Moon; Letter from North Korea Parts 1, 2, 3 and 4; Pre-Pyongyang: Japan and China; On Reading Books: A Better Way?; On to Pyongyang Part 2; Movies Are ??? Than Ever; Democratizing Art; Car-Pooling—An Idea Whose Time Has Come—and a Solution; Obama—A Personal Take; On to Pyongyang Part 1; Hurricanes—Whither Thou Blowest?; Horses vs. Humans—Improvement of Which Breed?; and Male vs. Women Swimmers. Peripheral technology hypotheticals may be cited only as such. RSS inventory

The two administrative notices are Until Later… and New Blog Site. They count in the 43-item feed denominator but not the 41 substantive-post denominator.

Appearance denominator and access reconciliation

Twelve public direct-communication families were mapped: 11 provide usable full or partial attributable text in different forms, while one documentary is metadata-only and supports no content claim. No report, feature, excerpt, email or edited Q&A is upgraded to a transcript. [researcher inference · strong] source map

No private conference, board, LP, IC or classroom transcript is implied by this public denominator. The corpus contains no locally transcribed audio; all analysis uses publisher text, scans, reports or the explicit unavailable record.

Source-by-source analysis

1982 — Boston Computer Update

Rosen argues that falling cost and easier use can move personal computers from enthusiasts toward both homes and businesses. [investor-stated · strong] The interview is valuable because it predates Compaq's outcome and speaks from an analyst/investor transition point. scan, printed pp. 23–24

The reasoning links hardware capability, application availability, price and ease of use rather than treating unit growth as autonomous. The assumptions are that useful applications will arrive, corporate buyers will legitimate the category and customers will tolerate incompatibilities. Compaq later supports the category direction but does not validate every penetration estimate or timeframe. [researcher inference · moderate] Boston feature, pp. 23–24

Operating implication: freeze the adoption chain before a deal—job, application, standard, distribution, price and user friction. Access is a full scanned feature with partial quoted voice, not a verbatim interview; OCR should be checked against printed pages. [researcher inference · strong] source note

1986 — Practical Computing

Rosen's clearest contemporaneous underwriting hierarchy is people first, an unfilled need second and product quality third. He prefers large-company potential and close board involvement. [investor-stated · strong] interview, pp. 80–81

Osborne supplies the adverse example: a passive investment without sufficient control or representation. Ansa supplies a search example—Rosen says he reviewed roughly 100 database products before investing. The assumption is that people, need and product can be assessed early enough to justify ownership and governance; the missing comparison cohort prevents causal claims. [investor-stated · moderate] Practical Computing, pp. 80–81

The interview also forecasts new software opportunities as 286/386-class capability expands. Byte's 1987 update shows the view evolving rather than remaining a single frozen call. [investor-stated · moderate] Practical, pp. 80–81 · Byte, p. 6

1987 — Byte conversations

Two short conversations record Rosen changing his view of software opportunity after seeing what the 80386 enabled. [investor-stated · moderate] Byte, printed p. 6

The process lesson is updating on capability thresholds, not merely defending an earlier forecast. The evidence is editor-mediated and brief; it identifies no investment, price, team, market size or allocation. Later software success cannot be scored as Rosen alpha without a frozen security decision. [researcher inference · strong] source note

1988 — Four Million Transistors lecture series

Rosen describes the PC industry's “hyper-growth” as the interaction of enabling technology, entrepreneurs and capital, while acknowledging that venture investors including him rejected Philippe Kahn/Borland. [investor-stated · moderate] Caltech four-lecture report

The Borland admission falsifies a simple “technical-network access finds every winner” story. Rosen's low then-current penetration observation supported continued category growth, but a category forecast and company decision are separate. The four-lecture report is not a transcript, so only explicitly attributed wording is Rosen voice. [researcher inference · strong] source note

1990 — Inc., Risky Business

Mamis's reporter profile—not a visible Rosen answer—reports Fund I at $25 million, the first two investments as failures, $2.5 million in Compaq, roughly $2 million in Lotus, Fund II at $60 million and Fund III at $65 million. It supplies the best public denominator: 45 financed entities by early 1990. [contemporaneous record · moderate] Inc. profile

The two accessible Rosen answers say venture remained vigorous and that he and Sevin had reduced new-company selection while retaining Fund I/II board work and Fund III adviser/investor roles. Mamis attributes the active-board process change to the early failures; the missing Q&A body does not preserve Rosen confirming it. Quantities are unaudited, and differences from L. J. Sevin's recollection stay unresolved. [investor-stated · strong] source note

1990 — Raising Money foreword excerpt

The reproduced foreword argues that business decisions, unlike many engineering problems, often lack a uniquely correct answer. [investor-stated · moderate] Caltech review, printed p. 39

Its operating implication is scenario work under uncertainty: record alternatives, assumptions and reversible tests rather than presenting a single optimized answer. The full foreword is unavailable, and the item is written prose rather than an appearance. [researcher inference · moderate] source note

1992 — Compaq annual meeting

After Rod Canion's removal, Rosen publicly argued that Compaq's new strategy would restore competitiveness. [investor-stated · moderate] Washington Post

The report suggests a crisis-governance test: a chair replacing a founder-CEO should name the failure, new strategy, indicators and review date. Any later recovery cannot assign sole credit to Rosen; executives, employees, products, prices and markets matter. Full remarks and Q&A are unavailable. [researcher inference · moderate] source note

1999 — Compaq annual meeting

Rosen acknowledges that Compaq's strategy had not been executed well and identifies bureaucracy, speed, simplicity, services, Internet distribution, channel choice, profitability and growth as priorities. [investor-stated · strong] Wired

The admission distinguishes strategy from operating capacity, but it is also a defense by the chair who oversaw the period. Hostile shareholder reaction is counterevidence; no full transcript or board packet is available. [researcher inference · strong] Wired report · source note

2000 — Retirement email

Rosen endorses Michael Capellas as both chair and CEO while emphasizing an independent board. [investor-stated · strong] email reproduction

That decision conflicts with Rosen's general preference for separating the roles. The exception may reflect changed circumstances or a belief that board independence substitutes for separation; the email does not compare alternatives. It is full direct prose preserved by a third party, not a speech or primary company archive. [researcher inference · moderate] retirement email · source note

2017 — Computer History Museum oral history

The transcript supplies the fullest reasoning chain: engineering and analyst formation; newsletter/conflict history; fund formation; Synapse and Osborne; Compaq and Lotus; firm attribution; Rosen Motors; Compaq governance; and explicit memory uncertainty. [investor-stated · strong] CHM transcript

Three testable ideas recur: full IBM compatibility plus portability created Compaq's wedge; rare creative judgment justified backing unconventional Kapor; and knowledge, networks and deal flow improve the conditions in which luck operates. Each is retrospective and requires contemporaneous controls. [investor-stated · moderate] CHM, pp. 13–19, 25–26

Rosen admits his product-use heuristic is sample-biased and that he likely would have passed on eBay; calls DEC a mistake; cannot reconstruct the board's ex-ante reasoning; and separates firm successes from his roles. These admissions make the record useful while its dates/economics require controls. [investor-stated · strong] CHM, pp. 23–31

2019 — Bloomberg Markets/Yahoo Q&A

The recovered published highlights add four distinct ideas. Rosen says strong founders can survive rejection of a narrow first idea, as the Compaq team did; product enthusiasm must be paired with an organization's ability to generate successors; “we” backed Electronic Arts around Trip Hawkins, preserving firm rather than sole-Ben attribution; and Energy Vault was his only investment in roughly 20 years. [investor-stated · moderate] published highlights · source note

The Energy Vault statement is a late personal thesis, not a Sevin Rosen decision or verified personal-investment denominator. Rosen also links career learning to company experience and its network. Bloomberg calls the article “highlights”; the full interview/audio is unavailable. [investor-stated · moderate] source note

2016 — Silicon Cowboys

IMDb lists Rosen in the cast of the 2016 Compaq documentary, but metadata alone does not verify an on-screen appearance, speaking role, or content; no stable lawful full stream or complete transcript was verified in this research environment. [contemporaneous record · provisional] film metadata

The metadata route is listed to make the appearance denominator auditable. No scene, quotation, timestamp, decision claim, forecast, or confirmed screen appearance depends on it. [researcher inference · strong] source note

Chronology and evolution of the spoken record

Recurring frameworks and strongest counterarguments

Deal and team attribution ledger

Forecast scorecard

Five highest-value appearances

1. CHM oral history — deepest direct decision record; must be externally controlled.
2. Practical Computing, 1986 — strongest contemporaneous process interview.
3. Inc., 1990 — best public fund and denominator interview.
4. Boston Computer Update, 1982 — best early pre-outcome category thesis.
5. Wired, 1999 — best contemporaneous execution admission and adverse governance context.

Gaps and access limits

• Full prepared remarks and Q&A from the 1992 and 1999 Compaq meetings were not found.
• The complete Raising Money foreword and canonical 2019 Bloomberg body remain unavailable.
• Silicon Cowboys has no verified stable full transcript in the research environment.
• Private partnership, IC, LP, board and conference material is outside the public denominator.
• Reported excerpts are not silently upgraded to transcripts, and a reproduced email/foreword is not called a talk.
• No spoken forecast has a complete externally scored universe; directional calls are not treated as audited alpha.

Appearance note index

• Boston Computer Update, 1982
• Practical Computing, 1986
• Byte conversations, 1987
• Four Million Transistors, 1988
• Inc. Risky Business, 1990
• Raising Money foreword excerpt, 1990
• Compaq new strategies, 1992
• Compaq promises better days, 1999
• Compaq retirement email, 2000
• Silicon Cowboys, 2016 — metadata only
• CHM oral history, 2017
• Bloomberg Markets/Yahoo highlights, 2019

1. Turn research into a compounding information network

• Rule — Publish useful market maps, convene practitioners, test products, and log incoming opportunities before trying to pick winners; use the resulting network to enlarge and improve the comparison set, while measuring selection quality separately from access.
• Evidence status — [documented behavior · strong] for Rosen's newsletter, conferences, and product demonstrations; [investor-stated · moderate] for the reported Ansa search and his later claim that knowledge, relationships, and deal flow improve the conditions in which luck operates. No matched opportunity cohort proves that the network generated superior returns.
• Provenance and origin — By 1979 Rosen was publishing the Rosen Electronics Letter and evaluating unreleased products; his two-page VisiCalc reaction recorded the business-use implication before the famous PC investments. VisiCalc letter and scans In 1982 a contemporary feature described the newsletter and conferences as an information-and-relationship system, not merely personal commentary. Boston Computer Update, printed pp. 23–24
• Mechanism and reasoning — Repeated research creates a taxonomy; convening supplies expert correction and proprietary-seeming but legitimately shared observations; incoming proposals then reveal where founders are clustering. The causal chain is research → informed relationships → broader deal flow → better comparisons → a larger surface on which judgment and luck can operate. The model fails if popularity merely increases low-quality inbound or if access is mistaken for selection skill. [researcher inference · strong] CHM oral history, printed pp. 7–8, 25–26
• Historical and vehicle context — Rosen developed the system as a sell-side analyst and independent publisher when semiconductor and PC information was fragmented, then divested the research business because venture ownership conflicted with publication independence. That pre-fund institution and early-1980s geography were necessary advantages; the precise subscriber list, conference attendee set, proposal funnel, and economic conversion are not public. CHM, printed pp. 8–12 Dyson collection finding aid
• When to use — Use when a market is new, vocabulary and value chains are unsettled, experts are dispersed, and a fund can contribute a credible research product. Do not use it as a reason to manufacture thought leadership in a mature market, to solicit confidential information improperly, or to assume that large inbound volume is evidence of an edge.
• How to apply it — (1) Define the market map and falsifiable questions; (2) publish or convene around evidence useful to operators; (3) record every opportunity and rejection under a stable taxonomy; (4) compare each candidate with the live set; (5) separate source, sponsor, diligence, and decision credit. IC questions: What did the network reveal that public data did not? Is it lawful and repeatable? What is the comparable set? Kill criteria: no differentiated learning after two research cycles, conflicts that cannot be managed, or a funnel whose conversion and miss rates cannot be logged. Evidence threshold: at least three independent expert corrections and a dated candidate comparison before conviction; next milestone is a scored funnel review.
• Examples —
  • Documented pre-investment application — VisiCalc: [documented behavior · strong] Rosen used an unreleased spreadsheet demonstration to identify a business workflow that could pull PC demand, then demonstrated it inside Morgan Stanley. The observation was real; it did not by itself select Lotus or Compaq. 1979 letter 2011 first-person recollection, “Easy Come, Easy Go” and VisiCalc sections
  • Documented investment application — Ansa: [investor-stated · moderate] Rosen said he reviewed roughly 100 database products before backing Ansa and becoming non-executive chair. The list and scorecard do not survive, so “100” is a reported funnel, not an audited denominator. Practical Computing, printed pp. 80–81
  • Researcher-applied illustration — Compaq: [researcher inference · moderate] the newsletter/conference network and Rosen-Sevin partnership plausibly improved access to the founders, but no source shows that publishing caused the deal. The evidenced decision was shared diligence after the team arrived. CHM, printed pp. 11–16
  • Counterexample — Borland: [investor-stated · strong] Rosen's category expertise and network did not prevent him from passing on Philippe Kahn; later board service cannot be rewritten as originating foresight. Caltech lecture report, printed p. 18
• Evidence — The contemporaneous 1979 letter establishes the research act; the 1982 feature independently records the research/conference system; the 1986 Q&A records comparative search; and the 2017 oral history supplies Rosen's later mechanism and its luck qualification. 1979 scan 1982 feature, printed pp. 23–24 1986 Q&A, printed pp. 80–81 2017 CHM, printed pp. 7–8, 25–26
• Limits, failure modes, and exception ledger — Network prestige can create adverse selection, consensus herding, and unearned attribution; the public record contains no complete inbound, pass, or return cohort. Ansa's reported search did not preserve its alternatives, and Borland is a visible false negative. Rosen's own formulation includes luck, so this is a process model, not proof of personal alpha. searched-denominator limits
• Evolution and contradictions — The mechanism begins as paid research in the 1970s, becomes a venture sourcing advantage after 1981, and is described in 2019 as career learning through company experience and relationships. Rosen also withdrew from new selection by Fund III, showing that a network can outlive direct decision authority. 2019 Q&A, network and career exchanges Inc., 1990, second visible Q&A answer
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] What transfers is the research flywheel and scored opportunity set; what fails is assuming a public content audience equals privileged access. Search, data rooms, virtual communities, and AI lower mapping cost but also increase copied theses and conflict risk. Applicability signals are expert correction, unique lawful data, high-quality founder referrals, and measurable funnel learning; likely misuse is vanity-media volume. A climate-infrastructure fund could publish a grid-constraint map, convene utilities and developers, and require three independently verified bottlenecks before investment; that example is hypothetical, not Rosen behavior. The contemporary need for capital and ecosystem mapping is illustrated, not validated, by Rosen's late systems essay. 2020 essay, PDF p. 1

2. Back people first, an unmet need second, and the first product third

• Rule — Rank the team's ability and integrity first, prove a consequential unmet need second, and treat the initial product as a testable implementation third; reject a weak proposal without discarding a strong team, but require evidence that the revised thesis can build a large company.
• Evidence status — [investor-stated · strong] for the 1986 hierarchy and the 2019 Compaq iteration rule; [retrospective witness · strong] for the independently bounded Lotus sequence and [retrospective witness · moderate] for Compaq; support for predictive superiority remains moderate because the rejected and ordinary-deal denominator is missing.
• Provenance and origin — The earliest explicit public statement recovered is Rosen's 1986 order: people, an unsatisfied need, then product, paired with a preference for large-company potential. Practical Computing, printed p. 81, selection-criteria answer His later Compaq account makes the rule dynamic: the disk-drive proposal failed, but the founders were invited back. 2019 Q&A, Compaq discussion
• Mechanism and reasoning — Products change as customers and platforms reveal constraints; capable, honest, adaptive founders can generate a second answer. Need prevents technology-first enthusiasm, while product supplies a concrete test. The causal bet is that team quality persists across iterations more reliably than a first specification, but “team quality” must be decomposed into relevant creation, recruiting, learning, and operating evidence rather than charisma. [researcher inference · strong] CHM, printed pp. 13–19
• Historical and vehicle context — A small early-1980s fund could work closely with a few founding teams before product-market fit. Rosen and Sevin each held a veto, so optimism about an unconventional founder was checked by partner dissent. That concentrated governance context is necessary to interpret the rule; it is not equivalent to a modern multi-stage fund selecting from thousands of software companies. CHM, printed pp. 11–18
• When to use — Use at pre-seed through Series A when the problem is important, product evidence is early, and iteration can be cheap enough to preserve runway. Do not use when the team lacks domain capacity for an irreversible safety, manufacturing, or regulatory task; when the “need” has no buyer; or when iteration consumes industrial-scale capital.
• How to apply it — (1) Score each founder on learning velocity, integrity, recruiting, and relevant creation; (2) interview users to quantify pain and buyer authority; (3) test the product against a specific workflow; (4) if product fails but team/need pass, define one bounded return experiment. IC questions: What would this team have to learn? Who urgently buys? Which result would invalidate founder or need? Kill criteria: integrity failure, no reachable buyer, repeated uninstrumented pivots, or inability to recruit missing expertise. Evidence threshold: two reference-checked examples of learning, five qualified user interviews, and one measurable product test; next milestone is paid repeat use or an industrial equivalent.
• Examples —
  • Documented application — Compaq: [documented behavior · strong] Ben and L. J. rejected the founders' disk-drive proposal but encouraged them to return; the compatible portable PC became the investable thesis. Founders Canion, Harris, and Murto—not Rosen—built the company. CHM, printed pp. 13–16 Kleiner Perkins retrospective
  • Documented exception — Lotus: [retrospective witness · strong] Kapor did not fit Rosen's usual executive pattern, but Rosen judged his creative product capacity exceptional; Kapor later bounded Rosen's contribution to backing, credibility, and access. Kapor oral history, printed pp. 11–13 CHM, printed pp. 17–19
  • Researcher-applied adverse illustration — Osborne: [researcher inference · moderate] a later-round, passive investment supplied neither the early team-building relationship nor control Rosen later preferred; the record does not show that the hierarchy was applied before entry. Practical Computing, printed p. 81 Inc., 1986
  • Counterexample — Borland: [investor-stated · strong] a technically capable unconventional founder was still rejected. The pitch and rationale are unavailable, preventing a claim that the rule should have compelled investment. Caltech lecture report, printed p. 18
• Evidence — The rule is explicit in the 1986 Q&A and 2019 highlights; the CHM and Kapor oral histories independently reconstruct Compaq and Lotus; Osborne and Borland prevent winner-only inference. 1986 interview, printed pp. 80–81 2019 Q&A Kapor, printed pp. 11–13 Caltech, printed p. 18
• Limits, failure modes, and exception ledger — “People first” can rationalize charisma, pedigree, or founder domination after weak product evidence. A rigid founder template would have excluded Kapor; unlimited patience would subsidize Eclipse-like capital burn; a severe regulatory or technical constraint may make initial domain competence non-negotiable. No complete Rosen screening log tests false positives, false negatives, or whether team scores preceded outcomes.
• Evolution and contradictions — The 1986 hierarchy is qualified by the later Kapor exception and strengthened in 2019 by “preserve the team after rejecting the idea.” Rosen's later auto-industry writing asks for industry expertise in a turnaround, while the Met Opera case praises hiring an outsider with the missing production capability. The consistent core is role-specific capability, not a universal résumé template. auto-management discussion Met Opera post, Gelb selection section
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] The hierarchy transfers to cheap-to-test software and AI applications, but fails when model dependence, safety, data rights, or hardware capital makes iteration non-reversible. Signals: rapid evidence-based revision, senior recruits, paid repeat use, and explicit abandonment of failed assumptions. Misuse: funding endless pivots because the founder is impressive. For a hypothetical AI compliance startup, reject a generic chatbot but fund a 60-day audit-workflow test only if references show regulatory depth and buyers precommit; kill if the team cannot secure lawful data or two design partners. No named current company is asserted.

3. Find the application that pulls the platform

• Rule — Underwrite a technology transition through the concrete workflow that makes a user buy or adopt the enabling platform; then update the application map when a new capability threshold changes what software or services can do.
• Evidence status — [contemporaneous record · strong] for Rosen's 1979 VisiCalc observation and [investor-stated · strong] for his 1982 adoption chain; [investor-stated · moderate] for the 1986–87 processor update. Company-selection efficacy is only moderate because category recognition does not specify team, security, price, ownership, or timing.
• Provenance and origin — The earliest recovered articulation is the 1979 letter: Rosen treated VisiCalc's business workflow as a reason personal computers might matter beyond hobbyists. letter transcription and scans In 1982 he explicitly linked PC adoption to cost, ease of use, and useful applications. Boston Computer Update, printed pp. 23–24
• Mechanism and reasoning — Platform specifications are inputs, not demand. A “killer application” changes a valuable job enough to justify hardware, learning, switching, and complement costs; adoption then attracts more complements. A later capability threshold can reopen an application frontier, so the map must be revised rather than extrapolated from the old constraint. [researcher inference · strong] 1986 Practical Computing, printed pp. 80–81 1987 Byte, printed p. 6
• Historical and vehicle context — The IBM PC used off-the-shelf components and an open architecture, and its software ecosystem expanded quickly; compatibility therefore reduced adoption friction while inviting clones. The fund entered during an unsettled standard transition, when hands-on software access could precede reliable shipment datasets. IBM PC history This context is necessary; a mature platform with entrenched distribution may not yield the same venture opening.
• When to use — Use at platform transitions—compute, interface, network, energy, or regulatory—when the enabling technology is visible but the purchasing job is not. Do not use when adoption is compulsory regardless of application, when infrastructure economics dominate end-user value, or when a demo has no buyer, budget, or retention evidence.
• How to apply it — (1) Name the changed capability; (2) identify one before/after workflow; (3) measure time, cost, quality, or revenue improvement; (4) list required complements and switching costs; (5) test whether repeat use pulls platform adoption. IC questions: Who buys and why now? Is the application causal or merely correlated with platform growth? What old constraint disappeared? Kill criteria: no budget owner, novelty without repeated use, or dependence on an uncommitted complement. Evidence threshold: three workflow observations, one paid pilot, and retention after novelty; next milestone is repeatable deployment without founder handholding.
• Examples —
  • Documented pre-investment application — VisiCalc: [documented behavior · strong] Rosen saw financial modeling become a legible business-PC job and used a live demonstration inside Morgan Stanley. The evidence supports demand insight, not a Lotus investment rationale. 1979 letter 2011 recollection, VisiCalc section
  • Researcher-applied investment illustration — Compaq: [researcher inference · moderate] full IBM compatibility preserved the application base while portability supplied the differentiated use case. Rosen directly states the compatibility reasoning; the founders and syndicate supplied execution. CHM, printed pp. 13–16
  • Documented forecast update — 80386 software: [investor-stated · moderate] Rosen revised his software opportunity view after seeing proposals enabled by additional processor capability. No named investment or shipment denominator survives. Byte, printed p. 6
  • Counterexample — personal-preference blind spots: [investor-stated · strong] Rosen later said his own use test could have caused him to miss eBay, and his initial Kindle skepticism proved wrong. A workflow can matter to others before it matters to the investor. CHM, printed pp. 25–27 Potpourri, Kindle discussion
• Evidence — Four independent records cover formation, repetition, update, and self-critique: the 1979 letter, 1982 feature, 1987 report, and 2017 oral history. 1979 scan 1982, printed pp. 23–24 1987, printed p. 6 CHM, printed pp. 23–27
• Limits, failure modes, and exception ledger — A selected winning application can be hindsight camouflage; no complete set of Rosen's application forecasts survives. Product use overweights the investor's demographics, and a platform can win through distribution, regulation, bundling, or cost rather than one application. VisiCalc does not prove Lotus selection, and the 80386 forecast lacks a scored horizon or security action.
• Evolution and contradictions — The rule moves from one observed application in 1979 to a multi-factor adoption chain in 1982, then to capability-driven software updating in 1986–87. Late self-critique narrows it: personal use is a discovery method, not a universal demand oracle. Practical Computing, printed p. 81 CHM, printed pp. 25–27
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] The workflow test transfers to AI, robotics, and new interfaces; what fails is treating benchmark gains as user demand when capability and prices change rapidly. Signals are paid repeated use, displacement of an incumbent workflow, retention across model changes, and gross margin after inference; misuse is calling a pilot a killer application. A hypothetical coding assistant should pass only if teams keep it after a model swap and materially change deployment throughput. The example is hypothetical; CoreWeave's filing supplies a grounded infrastructure reminder that customer concentration, GPU dependence, capex, and financing can dominate application enthusiasm. CoreWeave Q2 2026 Form 10-Q, Part II, Item 1A, customer, supplier, and capital-expenditure risks

4. Enter through a narrow wedge while preserving the ecosystem

• Rule — Join the standard or complement system customers already value, remove a specific adoption constraint, and choose an asymmetric beachhead where incumbents cannot or will not respond immediately; underwrite every required complement and its owner.
• Evidence status — [investor-stated · strong] for Compaq's full-compatibility-plus-portability logic; [investor-stated · moderate] for Tesla's asymmetric entry and HP/Compaq complement analysis; overall support is moderate because only Compaq is a directly connected venture application.
• Provenance and origin — Rosen's 1981–82 Compaq account is the earliest reconstructed use: he regarded partial IBM compatibility as commercially weak and portability as the opening. CHM, printed pp. 13–16 IBM's own history documents the open-architecture ecosystem that made compatibility consequential. IBM PC history
• Mechanism and reasoning — Compatibility lowers switching and complement costs; the wedge supplies differentiation while the entrant builds distribution, operations, and follow-on products. The complement map—product, standard, channel, supply, trust, financing, and regulation—shows which external owner can veto adoption. The mechanism fails if compatibility makes the product a commodity, the standard owner forecloses access, or an indispensable complement remains uncommitted. [researcher inference · strong] Compaq case
• Historical and vehicle context — The open IBM PC architecture and rapid software growth made full compatibility unusually valuable, while a small venture syndicate could finance a focused portable entrant. This was not a generic “copy the incumbent” rule. In automobiles, incumbents retained volume manufacturing, supply, service, and certification advantages, so Rosen later praised Tesla's high-end entry as time to build a broader system. Tesla Roadster post, “Tesla's opportunity” and competitor sections
• When to use — Use when a dominant standard has valuable complements, one underserved job is urgent, and the wedge can become a distribution or learning advantage. Do not use where certification, proprietary interfaces, supplier control, or platform retaliation can close the route; where the wedge is too small to support a company; or where “compatibility” erases differentiation.
• How to apply it — (1) Draw the customer workflow and complement owners; (2) identify the standard that reduces adoption friction; (3) specify one narrow constraint the entrant removes; (4) model incumbent response and platform retaliation; (5) require a path from wedge to a durable capability. IC questions: Which complement is non-negotiable? Who can revoke it? Why is the beachhead unattractive or inaccessible to incumbents? Kill criteria: no control over the differentiator, a hostile gatekeeper, no second market, or switching cost greater than user value. Evidence threshold: interoperable product proof, three reference customers, a named distribution route, and an incumbent-response scenario; next milestone is repeat purchase without bespoke integration.
• Examples —
  • Documented investment application — Compaq: [documented behavior · strong] full IBM compatibility preserved applications while portability supplied the wedge. Founders and co-investors share creation credit; the public record does not isolate returns from this mechanism. CHM, printed pp. 13–16 Kleiner Perkins retrospective
  • Researcher-applied non-investment illustration — Tesla: [researcher inference · moderate] Rosen argued that a premium sports car avoided immediate symmetric competition and could buy time to build manufacturing and distribution. His product purchase is not an equity investment and mass-market proof did not follow from his drive. Roadster post
  • Researcher-applied corporate illustration — HP/Compaq: [researcher inference · moderate] Rosen's retrospective case joined complementary products, geographies, channels, and scale while treating integration as the risk. He disclosed no role and no HP ownership, so this is analytical evidence, not deal credit. “The Merger That Worked,” fit and disclosure sections
  • Failure boundary — Rosen Motors: [documented behavior · strong] a technically promising drivetrain had no committed automaker integrator; absent complement ownership defeated the route to market after reported capital of about $24 million. Los Angeles Times, November 19, 1997
• Evidence — The CHM oral history connects the rule to Compaq; IBM independently establishes the standard context; Rosen's Tesla and merger essays extend the complement logic; Rosen Motors supplies a contemporaneous failure control. CHM, printed pp. 13–16 IBM Tesla LA Times
• Limits, failure modes, and exception ledger — Wedges can remain niches; compatible entrants can face price compression; complement owners can integrate, bundle, or revoke access. Compaq's success does not establish the base rate. HP/Compaq is an endpoint-sensitive retrospective, and Rosen Motors warns that a technical component without the industrial integrator is not an ecosystem strategy.
• Evolution and contradictions — The model begins as compatibility reasoning at Compaq, broadens into an ecosystem analysis of products, channels, and geographies, and becomes an explicit warning that a period-fit business model is not permanent. merger post, closing lessons Its tension is that preserving an incumbent ecosystem can delay necessary self-obsolescence, addressed by the next-product model below.
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate] What transfers is complement ownership and asymmetric entry; what fails is assuming API or model access is durable. AI platforms can change price, terms, routing, or bundle an application. Signals: multi-provider portability, proprietary workflow data, repeat distribution, and a wedge expanding into owned capability; misuse: calling thin integration a moat. Grounded current premise: CoreWeave's Q2 2026 filing describes customer concentration, NVIDIA-dependent GPUs, substantial capex, and future financing needs—complements that a modern infrastructure IC must model. CoreWeave Q2 2026 Form 10-Q, Part II, Item 1A, customer, supplier, and capital-expenditure risks

5. Underwrite the company's next product, not only today's hit

• Rule — Invest only when the organization can generate, select, and ship successor products before the current wedge matures; require evidence of a repeatable renewal system, not merely enthusiasm for one product.
• Evidence status — [investor-stated · strong] as a 2019 rule; [researcher inference · moderate] across Compaq, Lotus, and the Met Opera. The public record connects no complete three-company venture cohort to Rosen applying this test ex ante, so predictive support is provisional even though the rule itself is explicit.
• Provenance and origin — The earliest explicit articulation recovered is the 2019 published Q&A, where Rosen says product enthusiasm should be paired with evidence that the organization can create successors. 2019 Q&A, product-lifetime exchange A 1988 lecture report supplies an earlier adjacent belief: technology firms must accept self-obsolescence when discontinuities transfer leadership. Caltech lecture report, printed p. 26
• Mechanism and reasoning — A venture outcome compounds only if the company renews product, channel, and operating capabilities before competitors and standards move. The mechanism is discovery cadence → resource reallocation → deliberate cannibalization → successor adoption. Necessary assumptions are that the organization can learn before runway expires and that the next market remains economically adjacent; serial launches without retention are not renewal.
• Historical and vehicle context — PC hardware and packaged software product cycles shortened rapidly, while Fund I's long-lived ownership and Rosen's 18-year Compaq chair tenure exposed him to succession beyond the initial wedge. The model may be partly retrospective learning from Compaq's later bureaucracy and acquisition errors, not a frozen 1981 underwriting criterion. Compaq proxy, director tenure CHM, printed pp. 28–31
• When to use — Use after initial product-market evidence and before underwriting a long-duration ownership or growth round. Do not penalize a focused company before its first wedge works; do not demand unrelated diversification; and do not use launch count as a proxy for renewal quality.
• How to apply it — (1) Map current product maturity and replacement risk; (2) inspect how customer evidence enters the roadmap; (3) review the last two resource-allocation choices; (4) identify the successor thesis and the capability it reuses; (5) set cannibalization and adoption milestones. IC questions: Who can kill the incumbent product? What was learned from the last miss? Can a second team ship without founder heroics? Kill criteria: no renewal owner, roadmap driven only by incumbent defense, acquisition as a substitute for integration capacity, or no evidence of a second creation cycle. Evidence threshold: one shipped extension with retained users, a resourced successor experiment, and a board-reviewed sunset criterion; next milestone is successor revenue or usage independent of the first hit.
• Examples —
  • Researcher-applied investment illustration — Compaq: [researcher inference · moderate] compatibility and portability created the first wedge, but later bureaucracy, weak speed, and the DEC acquisition show that a successful first system did not guarantee renewal. Rosen later called DEC a mistake and could not reconstruct the board rationale. Wired, April 22, 1999 CHM, printed pp. 28–31
  • Researcher-applied investment illustration — Lotus: [researcher inference · moderate] 1-2-3 demonstrated the leverage of an application atop the PC standard, but the public record here does not establish that Rosen tested Lotus's successor process before investing. It is therefore an example of why the question matters, not a documented use of it. Kapor oral history, printed pp. 11–13
  • Researcher-applied institutional illustration — Metropolitan Opera: [researcher inference · moderate] Rosen's account links product improvement, marketing, and a new cinema distribution/revenue channel, showing renewal outside venture. His governance role and interested narration constrain causal credit. Met Opera post, “Measures of success” and HD sections
  • Failure boundary — DEC acquisition: [investor-stated · strong] buying a new capability can magnify integration burden instead of creating renewal; board approval is not evidence of organizational capacity. CHM, printed pp. 28–30
• Evidence — The 2019 Q&A establishes the rule; the 1988 lecture report establishes self-obsolescence; the Compaq proxy and 1999 meeting report document the long governance window and execution concern; the Met essay supplies a distinct operating illustration. 2019 Q&A Caltech, printed p. 26 Compaq proxy Met Opera post
• Limits, failure modes, and exception ledger — The rule can encourage premature diversification, waste, or narrative excuses for a fading core. A single durable product can be an excellent business. Model-specific attempted-route ledger: the research reviewed Rosen's 43-item blog feed, 12 direct-communication families, CHM/Kapor/L. J. oral histories, Compaq proxy and meeting reports, and the 23-path deal ledger; it found the explicit 2019 rule but no private IC memo, Lotus roadmap record, complete Compaq board packet, or three-company ex-ante application cohort. corpus reconciliation appearance reconciliation deal retrieval ledger
• Evolution and contradictions — The early public record celebrates platform applications and hyper-growth; the late rule reflects exposure to product-cycle decay. Rosen's 2008 HP/Compaq essay also says business models fit periods rather than remaining permanently superior. merger post, closing lessons The contradiction is productive: focus wins the wedge, but renewal must begin before focus becomes inertia.
• Modern VC translation (as of 2026-08-23) — [researcher inference · provisional; analyst-created hypothetical] The renewal test transfers, but AI-era release frequency makes launch count nearly meaningless and model vendors can absorb features quickly. Signals: retained cohorts across versions, a roadmap informed by proprietary workflow evidence, lower dependence on one model, and a second product with independent pull; misuse: demanding a “platform” before the wedge works. A hypothetical AI application passes only after its second workflow retains without bundling subsidies. This remains hypothetical because no grounded 2026 company case in the research record tests Rosen's rule.

6. Invest early enough to earn active governance

• Rule — When the investment is material and the company is early enough to shape, secure explicit information and board rights, work on management and financing rather than the founder's product, and record the conditions under which intervention, recusal, or exit becomes necessary.
• Evidence status — [investor-stated · strong] for Rosen's lesson from Osborne and his board-role boundary; [documented behavior · strong] for the Compaq chair role and [retrospective witness · moderate] for Lotus and Ansa. Outcome superiority is moderate because active-board successes lack a matched passive cohort and Compaq supplies adverse intervention evidence.
• Provenance and origin — Rosen says the passive, later-round Osborne loss taught him to invest earlier, obtain representation, and work actively. Practical Computing, printed p. 81 Inc. independently reported the first two failures and the subsequent emphasis on active boards, though its causal account is reporter-framed. Inc., 1990, web paragraphs 68–70
• Mechanism and reasoning — Early rights increase information, financing, recruiting, succession, and crisis options before a company is locked into a path. The board supervises management and capital allocation; it should not substitute investor taste for founder product judgment. The mechanism assumes the investor has relevant skill, time, and legitimate authority. More control can also create entrenchment, correlated errors, and unearned credit.
• Historical and vehicle context — Fund I was approximately $25 million, Ben and L. J. each held a veto, and a few major board roles could absorb meaningful partner time. Rosen's Compaq chair tenure ran from 1982 to 2000; Lotus and Ansa supplied other direct roles. That concentrated model differs from a high-velocity seed portfolio or a late-stage crossover vehicle. CHM, printed pp. 11–24 Compaq proxy
• When to use — Use when ownership and company risk justify time, the investor can fill an actual governance gap, and information or crisis rights are negotiated before trouble. Do not demand control for a small optional check, enter a later round merely to obtain a title, or intervene in product choices outside demonstrated competence.
• How to apply it — (1) Name the governance gap before term-sheet negotiation; (2) assign source, sponsor, diligence, approval, board, and follow-on roles; (3) agree information cadence and reserved matters; (4) define CEO evaluation, conflict, recusal, and succession triggers; (5) log interventions and counterfactuals. IC questions: What decision improves because we have a seat? Who is accountable? What would make us leave? Kill criteria: ceremonial seat, no time budget, authority without expertise, unresolved conflict, or founder/board expectations that cannot be written. Evidence threshold: rights in documents, monthly operating data, and a board skills map; next milestone is a dated governance review.
• Examples —
  • Documented negative application — Osborne: [investor-stated · strong] Rosen reported neither control nor representation and a complete $400,000 loss. Capital owner and the “control would save it” counterfactual remain unresolved. Inc., 1986 TIME, January 23, 1984
  • Documented application with adverse control — Compaq: [documented behavior · strong] Rosen held the chair role for 18 years and participated in two CEO removals; he later called DEC a mistake. Active governance supplied influence, not immunity from strategic or succession error. Compaq proxy CHM, printed pp. 28–31
  • Documented bounded application — Lotus: [documented behavior · moderate] Rosen served as founding director but left when IBM discussions conflicted with his Compaq chairmanship; Kapor credited his value mainly to credibility and access. Kapor oral history, printed pp. 11–13 CHM, printed pp. 15–18
  • Documented application — Ansa: [investor-stated · moderate] Rosen became non-executive chair after a broad product search; exact ownership, interventions, and proceeds are unknown. Practical Computing, printed pp. 80–81
• Evidence — The 1986 Q&A supplies the explicit lesson; Inc. supplies an independent contemporaneous account; SEC records establish the Compaq role; Kapor bounds Lotus value-add; and Rosen's later admission supplies an adverse governance result. 1986 Q&A Inc., 1990 SEC proxy Kapor
• Limits, failure modes, and exception ledger — Board activity is endogenous: weak companies demand more intervention. Seats can create overreach, founder conflict, groupthink, and reputational self-defense. Pfeiffer later argued his abrupt removal hurt Compaq, while critics said Rosen's dominance complicated recruiting; neither account proves causality. Computerworld adverse account Wired shareholder criticism
• Evolution and contradictions — The Osborne lesson hardens into active early ownership, but Compaq complicates it: Rosen favored separate chair/CEO roles, then endorsed Capellas holding both while emphasizing independent directors. This is a conditional governance model, not a constitutional rule. retirement email reproduction
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] Active governance transfers where ownership, risk, and expertise justify it; it fails in crowded cap tables where seats are ornamental or portfolio conflicts are pervasive. Signals: complete information, explicit reserved matters, independent directors, decision logs, and rehearsed succession; misuse: counting seats as value-add or using “founder-friendly” to avoid accountability. A hypothetical investor with seats at competing AI applications should pre-agree information barriers, recusals, and independent committees. The example is hypothetical; Rosen's Apple and Lotus recusals are the historical controls. Apple recollection CHM, printed pp. 15–18

7. Match leaders to the failure mode, then separate strategy from execution

• Rule — Define the job's dominant failure mode, select leaders using evidence that they understood and managed that risk before it became obvious, and evaluate strategy and execution on separate dated metrics.
• Evidence status — [investor-stated · moderate] across Rosen's Compaq meeting remarks, public-policy writing, and retirement email; [documented behavior · strong] for Compaq and the Met Opera. Venture-selection application beyond those roles is provisional.
• Provenance and origin — The earliest recovered operating use is Compaq governance after Canion's 1991 removal: Rosen defended a slower overhaul and named the new strategy rather than promising an instant fix. Washington Post, May 15, 1992 By 2009 he made the selection test explicit in another domain: ask whether a proposed leader had recognized leverage, capital, incentive, and innovation risks before crisis. “One Mistake,” five-part Secretary test
• Mechanism and reasoning — A generic “great CEO” score hides job differences: creation, scaling, turnaround, integration, and succession require different evidence. Pre-event decisions are harder to manufacture than post-event confidence. Once hired, separating thesis from operating cadence prevents a sound strategy from excusing slow execution—or weak strategy from being blamed only on people. [researcher inference · strong] Wired, April 22, 1999
• Historical and vehicle context — Rosen applied the model as a long-tenured public-company chair and civic-board participant, roles with authority and multi-year metrics. Venture boards often have less control, shorter runway, and incomplete executive alternatives. His independent-chair preference was later relaxed for Capellas, proving that board structure and job context matter. retirement email reproduction
• When to use — Use for founder-to-scaler transitions, turnarounds, integration, senior hiring, and board reviews. Do not use hindsight outcome alone as a leader score; do not demand prior incumbency where unconventional creation is the job; and do not replace product evidence with résumé pattern-matching.
• How to apply it — (1) Write the next 18-month failure modes; (2) collect dated examples of the candidate diagnosing analogous risks; (3) assign strategy and execution metrics separately; (4) define decision rights and review cadence; (5) precommit succession triggers. IC questions: What must be true even with excellent execution? Which miss is strategic versus operational? Kill criteria: no role-specific evidence, refusal to expose past errors, goals without dates, or board inability to act. Evidence threshold: two reference-verified episodes, a 30/90/180-day plan, and three operating indicators; next milestone is the first board scorecard.
• Examples —
  • Documented application with contested outcome — Compaq: [documented behavior · strong] Rosen backed Pfeiffer's strategy after Canion, then said in 1999 that strategy had not been executed well. Shareholder hostility and later DEC regret prevent a clean board-success narrative. Washington Post, 1992 Wired, 1999
  • Documented institutional application — Metropolitan Opera: [documented behavior · strong] Rosen describes defining the operating problem, hiring Peter Gelb for production/marketing capability, allowing transition, and monitoring multiple dated measures. His board role makes the narration interested. Met Opera post
  • Researcher-applied public-policy illustration — auto leadership: [researcher inference · moderate] Rosen argued that a turnaround leader needed industry knowledge spanning product, manufacturing, and distribution. This is not a portfolio hiring decision and conflicts with using outsiders where the missing capability is reinvention. “It's Time for a Change,” auto-management section
  • Exception — Kapor: [retrospective witness · moderate] evidence of rare product creation overrode Rosen's normal executive pattern, showing that the role score must not become pedigree matching. CHM, printed pp. 17–18
• Evidence — Contemporary Compaq reports show stated strategy and execution diagnoses; the Met post shows a separate selection/scorecard; the 2009 essay supplies the pre-crisis evidence test; CHM supplies the unconventional-founder exception. Washington Post Wired One Mistake CHM, printed pp. 17–18
• Limits, failure modes, and exception ledger — Prior foresight can be rhetorical rather than operational; references select favorable stories; board chairs can redefine strategy after an execution miss. Canion and Pfeiffer were distinct decisions inside one company, not independent proof. The public record lacks candidate lists, board packets, and predefined success metrics for those removals, while Pfeiffer's adverse account disputes the intervention. Computerworld
• Evolution and contradictions — Rosen moved from defending a strategic turnaround in 1992 to admitting execution failure in 1999, then endorsed combined chair/CEO authority under Capellas despite preferring separation. retirement email The durable principle is independent evaluation against the job, not one immutable governance form.
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] The role/failure-mode test transfers; what fails is treating big-company AI or growth titles as evidence for zero-to-one creation. Signals: dated incident decisions, recruiting quality, explicit strategy assumptions, and operating leading indicators; misuse: post hoc “execution” blame. A hypothetical industrial-AI company moving from pilots to deployment should hire for safety integration and field operations, with kill criteria tied to incident rate and repeat installations. The example is hypothetical, not Rosen behavior.

8. Separate technical proof from commercial scale and total capital

• Rule — Treat prototype, customer commitment, industrialization, unit economics, regulatory approval, channel integration, and financing-to-scale as separate gates; size the initial check and reserves to the next falsifiable milestone, not the eventual market story.
• Evidence status — [researcher inference · strong] reconstructed from Rosen Motors, Eclipse, Capstone, and Rosen's own industrial writing; Rosen directly distinguishes technical from commercial success, but no source proves that he consistently used this staged rule before investing.
• Provenance and origin — The strongest origin is the 1993–97 Rosen Motors experience: after nearly five years, reported capital of about $24 million, and 70 employees, the company closed without an automaker commitment. Los Angeles Times, November 19, 1997 Rosen later described the system as technically but not commercially successful. CHM, printed pp. 26–28
• Mechanism and reasoning — Deep-technology risk is conjunctive: a working component creates no scalable value if certification, supply, yield, cost, customer integration, or finance fails. Milestone tranches preserve information and stop options; a reserve policy determines whether early ownership survives later capital. The causal chain is technical validation → integrator/offtake → cost/yield → certification → financed scale. Passing one gate must not silently waive the others.
• Historical and vehicle context — Rosen Motors was a personal/family operating project rather than a Sevin Rosen fund investment, while Eclipse was a modest later personal check with no follow-ons. These vehicles lacked the disclosed reserve and governance machinery of a dedicated industrial fund. Capstone held both personal/family and separate firm interests, illustrating why vehicle, related-party, and salvage boundaries matter. Capstone prospectus, printed pp. 30, 49, 59–62
• When to use — Use for hardware, climate, aerospace, biotech tools, regulated products, and any model whose gross margin appears only at future volume. Do not impose industrial gates on low-capital software, and do not treat a large backlog, famous partner, or successful prototype as a substitute for binding, financeable adoption.
• How to apply it — (1) Draw every technical, regulatory, supply, customer, and financing gate; (2) price the capital and time to each; (3) assign an evidence owner and expiration date; (4) stage the check and reserves; (5) define salvage and stop rights. IC questions: Who signs the integration? What unit cost at audited yield? How much cash reaches commercial release? Kill criteria: no paid/binding integrator by deadline, cost curve outside the adoption envelope, certification beyond funded runway, or financing needs beyond the reserve ceiling. Evidence threshold: independent technical validation, production quote, binding pilot/offtake, and full cash-to-launch model; next milestone is customer acceptance at target economics.
• Examples —
  • Documented failure — Rosen Motors: [documented behavior · strong] technical work proceeded without a committed automaker, and closure followed. Capstone's later $590,000 related-party asset purchase and assumed leases do not establish recovery of the reported $24 million consumed. LA Times Capstone prospectus, Certain Transactions, printed p. 59
  • Documented failure — Eclipse: [investor-stated · strong] aircraft certification, engine supply, scale economics, and follow-on finance were coupled; Rosen made no follow-ons, expected severe dilution, and later recorded bankruptcy. Eclipse post bankruptcy follow-up
  • Researcher-applied mixed illustration — Capstone: [contemporaneous record · strong] the prospectus documents product-direction, Rosen/Jaggers board roles, dual ownership paths, and salvage transactions, but not original checks, cash returns, or proof that microturbine industrialization validated Rosen Motors. Capstone prospectus, printed pp. 30, 49, 59–62
  • Counterexample to overstrict screening — Kindle: [investor-stated · moderate] Rosen initially saw a weak problem and later admitted the judgment was wrong; customer-need gates must allow timing and category uncertainty rather than institutionalizing investor taste. Potpourri, Kindle discussion
• Evidence — The closure report independently documents capital, staffing, and missing OEM commitment; CHM supplies Rosen's own technical/commercial distinction; the Eclipse posts document dilution and bankruptcy; the Capstone prospectus provides primary ownership, governance, and related-party details. LA Times CHM, printed pp. 26–28 Eclipse Capstone filing
• Limits, failure modes, and exception ledger — Milestones can create false precision, delay learning, and reject nonlinear discoveries. Binding customers can disappear; production quotes can exclude yield and service; reserves can throw good money after bad. Rosen Motors' technical salvage may have option value, but no capitalization or recovery ledger permits a return conclusion. Eclipse's exact check, security, dilution path, and Rosen's governance role remain unknown.
• Evolution and contradictions — Rosen Motors supplied the commercial failure by 1997; the 2007 retrospective emphasized industry adoption constraints; Eclipse then showed that early access without follow-on policy can erase retained exposure. “Car of the Future,” automaker and failure sections The record does not show a later Rosen investment process changed as a result, so evolution is lesson articulation rather than demonstrated behavior change.
• Modern VC translation (as of 2026-08-23) — [researcher inference · strong] The gates transfer directly to climate and industrial technology; what fails is assuming a strategic investor logo guarantees integration or capital. Signals: contracted milestone funding, yield, certified performance, binding offtake, service readiness, and financing headroom; misuse: valuing pilot announcements as commercial scale. Grounded current example: Rivian's Q2 2026 filing describes Volkswagen milestone funding alongside continuing JV, supplier, factory, and additional-capital risks—the kind of coupled gate set this model requires. Rivian Q2 2026 Form 10-Q, Liquidity and Capital Resources; Part II, Item 1A, Volkswagen JV and capital risks

9. Preserve options and update at evidence thresholds

• Rule — When commercial decisions have no uniquely correct answer, keep plausible alternatives alive only while uncertainty is cheap to resolve; predefine the evidence that expands, changes, or kills each option, and update rather than defend the original narrative.
• Evidence status — [investor-stated · moderate] for business decision uncertainty, competing-architecture tests, and the 80386 update; [retrospective witness · moderate] for preserving the Compaq team after rejecting its first proposal. Performance support is moderate because option costs and rejected alternatives are not reconstructed.
• Provenance and origin — The earliest observed use is the Compaq return invitation after the disk-drive rejection in 1981–82. CHM, printed pp. 13–16 The earliest explicit general statement recovered is Rosen's 1990 foreword: business decisions, unlike many engineering problems, often lack one derivable correct answer. Raising Money review, printed p. 39
• Mechanism and reasoning — Irreversible commitment under model uncertainty destroys learning; unlimited optionality destroys focus. A cheap, time-bounded test converts disagreement into information. The causal chain is alternative hypotheses → discriminating test → threshold → allocation update. The model assumes tests are informative and decision-makers will honor the threshold rather than reinterpret failure.
• Historical and vehicle context — Early PC/software prototypes could often be tested with modest teams, making return invitations and software experiments relatively cheap. Aircraft, automotive, and semiconductor options could require much more capital; vehicle reserves and partner vetoes therefore determine whether optionality is real. Rosen and Sevin's two-veto Fund I structure supplied dissent but could also block unconventional opportunities. CHM, printed pp. 11–16
• When to use — Use when two or more hypotheses are plausible, a reversible experiment can discriminate them, and delay does not forfeit the market. Do not use where safety requires a hard stop, a test cannot resolve the core risk, option maintenance consumes most runway, or “wait for data” is avoidance.
• How to apply it — (1) Write mutually exclusive hypotheses; (2) define the cheapest discriminating test; (3) cap time and capital; (4) precommit expand/change/kill thresholds; (5) log the update and residual uncertainty. IC questions: What do we believe differently? Which observation changes the decision? What does delay cost? Kill criteria: non-discriminating experiment, threshold drift, no owner, or option cost above expected information value. Evidence threshold: one independently observable result tied to a dated decision; next milestone is the next capital release.
• Examples —
  • Documented application — Compaq: [documented behavior · strong] the investors rejected a disk-drive company but preserved the team option; the compatible portable proposal then crossed the new decision threshold. CHM, printed pp. 13–16 2019 Q&A
  • Documented forecast update — processor applications: [investor-stated · moderate] Rosen's published view changed after he saw what 80386-class capability enabled. The report identifies no security action, so it is an epistemic update rather than investment proof. Byte, printed p. 6
  • Researcher-applied field-test illustration — vehicle architectures: [investor-stated · moderate] Rosen proposed comparing competing electric/hybrid architectures through service evidence rather than declaring one winner from specifications. Potpourri, vehicle-architecture section
  • Counterexample — Borland pass: [investor-stated · strong] no small validation option is documented; later success shows the cost of a false negative but does not prove a small check was rational on the missing pitch. Caltech report, printed p. 18
• Evidence — CHM and the 2019 Q&A independently preserve the Compaq iteration; the 1990 foreword states the uncertainty premise; the 1987 report records an update; the 2008 post records a comparative test preference. CHM, printed pp. 13–16 foreword excerpt, printed p. 39 Byte, printed p. 6 Potpourri
• Limits, failure modes, and exception ledger — Options can rationalize indecision, tiny non-informative checks, or portfolio sprawl. A milestone can be gamed; a fast market can close while the investor experiments. The original Compaq and Borland alternatives, prices, and dissent logs do not survive. Rosen Motors and Eclipse show that preserving a technical option without a commercial or reserve threshold can become expensive rather than prudent.
• Evolution and contradictions — The observed Compaq behavior precedes the 1990 uncertainty statement; the 1987 software update shows willingness to revise; the later DEC admission shows why decision logs matter when memory cannot reconstruct the original rationale. CHM, printed pp. 28–30 The tension is with conviction: Rosen still wanted companies large enough to justify the work, so options must expire into concentration or a pass.
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] Fast software experimentation makes options cheaper, but AI benchmarks and pilots can be non-discriminating; industrial tests remain slow and capital-intensive. Signals: preregistered metrics, independent user behavior, clear decision dates, and budget released by milestone; misuse: “optionality” as excuse for no thesis. A hypothetical AI agent investment starts with a capped pilot and expands only if retained task completion survives a model/vendor swap. The example is hypothetical, not a documented current investment.

10. Use denominators and forecast ledgers to reason under uncertainty

• Rule — Before ranking risk or claiming foresight, define the exposure denominator, record the forecast, horizon, alternatives, and falsifier, then score it after resolution without converting one vivid outcome into a general law; in structured systems, also trace leverage, opacity, maturity mismatch, custody, and payoff incentives.
• Evidence status — [investor-stated · moderate] for Rosen's denominator, conventional-wisdom, and forecast-scoring arguments; [researcher inference · strong] for the investment checklist. No public forecast universe or complete deal denominator proves forecasting skill.
• Provenance and origin — The earliest recovered investment-adjacent premise is the 1990 foreword's distinction between determinate engineering problems and ambiguous business choices. Raising Money review, printed p. 39 In 2008–09 Rosen made the method explicit through election-scorekeeping, football frequencies, and transportation exposure denominators. Reflections, football section Fear of Flying, fatality comparisons
• Mechanism and reasoning — Salience, convention, and retrospective narrative overweight selected outcomes. A denominator makes the reference class explicit; a frozen forecast prevents moving goalposts; a falsifier forces learning. Because markets are adaptive systems, base rates inform but do not mechanically determine a case. The model fails when the denominator is chosen after the outcome or when aggregate correlations are treated as causal.
• Historical and vehicle context — Rosen's public portfolio record is precisely the adverse setting: 23 named opportunity paths, 45 reporter-framed financed entities by 1990, and no complete cash-return history. Famous wins and narrated failures are selected. deal denominator His late writing applied the method mostly to public policy and everyday risk, not a disclosed Sevin Rosen IC process.
• When to use — Use for market sizing, failure rates, reserve decisions, portfolio claims, forecasts, and any argument driven by a vivid anecdote. Do not import a historical base rate across different stage, geography, vehicle, or regime without adjustment; do not withhold action merely because the perfect denominator is unavailable.
• How to apply it — (1) Define numerator, denominator, unit, period, and inclusion rule; (2) freeze prediction, horizon, alternatives, and falsifier; (3) identify selection and survivorship bias; (4) update at a fixed date; (5) preserve unresolved outcomes. IC questions: Compared with what? Which cases are unobserved? What result changes our view? Kill criteria: denominator cannot be named, forecast has no horizon, metric changes after outcome, or causality rests on one anecdote. Evidence threshold: one defensible reference class plus case-specific evidence and a dated score date; next milestone is forecast review.
• Examples —
  • Documented non-investment application — transportation risk: [investor-stated · moderate] Rosen compared vivid air accidents with exposure-based alternatives. Statistics remain publication-dated and do not transfer mechanically to venture tails. Fear of Flying
  • Documented non-investment application — football convention: [investor-stated · moderate] Rosen used observed fourth-down frequencies and loss aversion to question received strategy; game-state and selection controls remain necessary. Reflections, football section
  • Documented forecast audit — 2008 election: [documented behavior · moderate] Rosen scored his prediction after resolution and discussed which assumptions missed. That is process evidence, not venture performance. “On... a Lot of Topics,” election section
  • Researcher-applied system-risk illustration — financial structures: [researcher inference · moderate] Rosen's crisis writing asks who bears leveraged downside, where records and incentives become opaque, and whether a remedy reproduces moral hazard. The source is public-policy analysis, not evidence of a Sevin Rosen IC rule. “It's Time for a Change,” Stiglitz discussion
  • Researcher-applied portfolio boundary — Sevin Rosen: [researcher inference · moderate] removing Lotus and Compaq leaves no defensible public economic ranking, and 18/23 named-path reconstruction is not a firm hit rate. Pareto removal test denominator reconciliation
• Evidence — The foreword establishes uncertainty; two independent Rosen posts establish denominator and conventional-wisdom tests; a third records forecast scoring; the deal ledger demonstrates the selection problem in his own record. foreword, printed p. 39 Fear of Flying Reflections deal return controls
• Limits, failure modes, and exception ledger — Denominators can create false precision, and adaptive systems change reference classes. Rosen's transportation figures are time-bound; football averages are not causal; the election test is one forecast; no complete Rosen forecast log, screening funnel, pass set, loss schedule, or audited fund cash flow exists. The model disciplines uncertainty but cannot manufacture missing data.
• Evolution and contradictions — A 1982 category forecast lacked precise share definitions; the 1986–87 software forecast lacked a scored universe; the 2008–09 writing becomes more explicit about denominators and ex-post scoring. 1982 feature, printed pp. 23–24 1987 Byte, printed p. 6 His broad 2020 crisis forecast reverses some discipline by offering propositions with loose horizons, so it belongs in a prediction ledger rather than as proof of a timeless model. 2020 essay, PDF p. 1
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate] Denominator discipline transfers; what fails is using stale SaaS or venture base rates for AI, climate, or regulated fintech without stage and regime matching. Signals: frozen cohort definitions, exposure-adjusted losses, calibration curves, and preregistered follow-ups; misuse: base-rate vetoes that ignore case evidence. Grounded current example: the FDIC's proposed custodial-account recordkeeping rule identifies beneficial-owner and reconciliation risk across intermediaries, showing why a fintech IC must map the actual account denominator rather than trust a partner-bank logo. Federal Register, RIN 3064-AG07

11. Write conflict, authority, and attribution boundaries before acting

• Rule — Before investing, publishing, governing, or narrating an outcome, write which vehicle owns the exposure, who may decide and receive information, who must recuse, and which person or team deserves each part of the credit.
• Evidence status — [retrospective witness · moderate] for the Apple, VisiCorp, and Lotus conflict rationales; [documented behavior · moderate] for the Rosen Research divestiture, public roles, and later disclosure; [investor-stated · strong] for firm-versus-person attribution. Compliance documentation and economic allocation remain unavailable, so the prescriptive rule remains moderate.
• Provenance and origin — Rosen says he declined a private Apple placement and sold a small VisiCorp holding because securities ownership compromised analyst independence. CHM, printed pp. 9–11 He then sold the research business when venture investing created the same structural conflict. The conduct is retrospectively reported, not reconstructed from a compliance file.
• Mechanism and reasoning — Written boundaries protect decision quality, lawful information flow, trust, and accurate learning. Vehicle labels prevent personal, family, fund, company, and successor-firm economics from being mixed; role labels prevent sourcing or board work from becoming sole-investor mythology. Recusal costs access or upside, but hidden conflicts corrupt both action and historical attribution.
• Historical and vehicle context — Rosen crossed analyst, publisher, GP, public-company chair, civic trustee, personal investor, and family-company roles. Fund I's shared vetoes and later partner specialization further divide credit. These overlapping roles make boundaries a necessary operating system rather than a disclosure footnote. people and decision-rights map vehicle analysis
• When to use — Use before every security purchase, board appointment, competitive portfolio decision, publication, related-party transaction, follow-on, and retrospective case. Do not use disclosure as permission for an unmanageable conflict; recuse or decline where information barriers and independent decision rights are insufficient.
• How to apply it — (1) Name the legal vehicle and beneficial owner; (2) assign source, sponsor, diligence, approval, negotiation, board, follow-on, exit, and narration roles; (3) map competitive and publication conflicts; (4) document recusal and independent authority; (5) report economics only at the evidenced level. IC questions: Whose capital and authority are these? Who cannot see or vote? Which claim would mislead if “we” became “I”? Kill criteria: undisclosed personal exposure, no independent approver, incompatible board duties, or attribution unsupported by records. Evidence threshold: signed conflict schedule, role map, cap-table/vehicle confirmation, and recusal minutes; next milestone is quarterly refresh.
• Examples —
  • Documented conduct — Apple and VisiCorp: [investor-stated · moderate] Rosen says he declined the offered Apple security and sold VisiCorp to preserve analyst independence. The opportunity cost is real but unknowable; the compliance record is unavailable. 2011 first-person account CHM, printed pp. 9–11
  • Documented structural remedy — Rosen Research: [documented behavior · moderate] he sold the newsletter/conference business to Esther Dyson after becoming a venture investor. An unbylined 1983 Rosen-branded issue is therefore not assigned to Ben. Dyson finding aid 1983 note
  • Documented recusal/exit — Lotus: [investor-stated · moderate] Rosen left the board when IBM-related discussions conflicted with his Compaq chairmanship. The act limits information and influence rather than proving perfect conflict management. CHM, printed pp. 15–18
  • Documented disclosure — HP/Compaq: [investor-stated · moderate] when defending the merger, Rosen said he had not participated and held no HP stock; the essay is analysis, not transaction authority. merger post, disclosure
  • Attribution control — EA and SGI: [investor-stated · strong] Rosen named them as firm successes while saying Sevin Rosen did not lead and he held no board seat; Brook Byers is the evidenced EA director. CHM, printed pp. 22–24 Kleiner Perkins EA history
• Evidence — Rosen's CHM account supplies multiple role boundaries; the Dyson archive independently fixes custody transfer; Kapor independently bounds Rosen's Lotus contribution; SEC/company records distinguish board tenure and ownership where available. CHM, printed pp. 9–24 Dyson finding aid Kapor, printed pp. 11–13 Compaq proxy
• Limits, failure modes, and exception ledger — Most conduct evidence is Rosen's retrospective account; no conflict register, partnership agreement, board minutes, or personal transaction ledger survives publicly. Disclosure does not neutralize influence, and formal recusal may not erase prior knowledge. Excessive avoidance can forgo learning and returns; the rule requires proportional safeguards, not ceremonial paperwork.
• Evolution and contradictions — Analyst-era conflicts led to declined or sold exposure and then structural divestiture; Lotus produced a board-level exit; later writing disclosed no HP role or ownership; CHM explicitly separated firm wins from his lead roles. Yet Capstone's personal/family and firm interests and its related-party Rosen Motors transaction show how complex the map can remain even when filings disclose it. Capstone prospectus, printed pp. 49, 59–62
• Modern VC translation (as of 2026-08-23) — [researcher inference · moderate; analyst-created hypothetical] The rule transfers directly, while AI-era competitive overlap, shared models, expert networks, SPVs, and personal angel checks multiply conflicts. Signals: vehicle-level cap tables, written information barriers, independent approvals, recusal logs, and role-specific deal retrospectives; misuse: a generic portfolio-conflict disclosure with no operational restriction. A hypothetical multi-stage fund owning competing AI agents should assign separate teams, prohibit cross-company confidential sharing, and use independent committees for follow-ons. The example is hypothetical; no current named fund is being accused.

Evidence gaps and candidate models

The inventory reconciles all philosophy signals in the 29 source notes, the 17-item deep writing corpus, 12 direct-communication families, and 23-path deal ledger. Eleven models survive after synonyms and nested rules are merged; the exclusions below are explicit so a memorable phrase or famous outcome cannot silently become a twelfth rule. writing reconciliation talk reconciliation deal reconciliation

• “If you have no losses, you are not taking enough risk.” Rosen states the thought in 1986, but the public record lacks a loss-rate denominator, allocation rule, or evidence that more risk improved expected return. It remains a temperament slogan, not a decision-ready model. Practical Computing, printed p. 81
• “Only pursue very large companies.” Rosen links company scale to the fixed work of investing, but no minimum market, ownership, price, probability, or fund-return threshold survives. Scale is retained as a boundary inside Models 2 and 5 rather than inflated into an unsupported standalone allocation model. CHM, printed p. 16
• “Knowledge makes you luckier.” This causal phrase has no matched network cohort; its decision-ready content is already captured in Model 1's research/funnel procedure, with Borland as the false-negative control. CHM, printed pp. 25–26
• Always separate chair and CEO. The Capellas exception disproves an unconditional rule. Chair/CEO structure, independent oversight, intervention, and recusal are treated as boundary conditions inside Model 6. retirement email
• Self-obsolescence as a separate model. It is a mechanism within Model 5's successor-product rule; splitting it would duplicate the same renewal decision without adding a distinct trigger. Caltech lecture report, printed p. 26
• Acquisition complementarity as a separate model. Rosen's HP/Compaq analysis is retrospective, and DEC is an adverse case; capability fit is therefore an example and failure test inside Model 4, not an evidenced venture-M&A playbook. merger essay CHM, printed pp. 28–30
• Crisis automatically accelerates innovation. The 2020 essay is a broad late forecast with loose horizons and no allocation record. It belongs in Model 10's forecast ledger, not as proof of a reusable selection rule. 2020 essay, PDF p. 1
• Patient science funding as a venture model. Rosen supports upstream research and later climate work, but the sources provide no private-fund duration, selection, ownership, reserve, or exit rule. The idea remains macro context until an investment process or repeated cases surface. science and technology post 2019 Q&A, Energy Vault exchange
• Salvage component IP after system failure. Rosen Motors, Capstone, and Pentadyne create one linked salvage episode, not three independent decisions; ownership, proceeds, and repeat application are missing. It remains a useful postmortem question under Model 8. Rosen Motors post Capstone prospectus, printed p. 59

The attempted-route record is specific rather than generic: the project reviewed the complete preserved 2007–09 feed, the accessible 1979/1983/2011/2020 writing families, every mapped public direct-communication family, CHM/Kapor/L. J. oral histories, public-company filings, period press, and company/partner/adverse searches. Missing private IC memoranda, partnership agreements, board packets, LP reports, screening/pass logs, cap tables, and cash-flow histories are named in the deal retrieval ledger. source coverage assessment deal retrieval ledger

For the integrated operating system, continue to the investment philosophy; for decision freezes, counterfactuals, and no-hindsight exercises, use the deal cases.

What the evidence supports

The record supports a narrower and more useful claim than “Ben Rosen picked the PC winners.” [researcher inference · moderate] From roughly 1979 through 1990, Rosen built an information surface through research, product use, conferences, proposals, and industry relationships; in selected PC/software cases he translated that information into an application, platform, or compatibility test; and in a few companies he coupled financing with long-duration governance. The frozen 1979 VisiCalc note supports application-led category observation, the 1982 feature supports the research/deal-flow loop, and the 1986 interview states people → unmet need → product plus early active-board preferences. VisiCalc letter, scans pp. 1–2 · Boston Computer Update, printed pp. 23–24 · Practical Computing, printed pp. 80–81

Selected behavior fits that process, but personal alpha does not follow. Compaq preserves the team after rejecting the disk-drive proposal and makes full IBM compatibility plus portability the wedge; Lotus shows an unconventional-founder exception and credibility/access value-add; Ansa shows a reported comparative search. Independent witnesses divide credit: L. J. Sevin confirms shared decision rights, while Mitch Kapor says Rosen brought legitimacy and Wall Street attention rather than operating or product strategy. Ben Rosen CHM, printed pp. 13–19 · L. J. Sevin CHM, printed pp. 20–24 · Kapor CHM, printed pp. 11–13

The economic record is much weaker. TIME reported Fund I paper values of roughly $70 million for a $2.1 million Lotus stake and $40 million for a $2.5 million Compaq position at their 1983 IPOs; those are contemporaneous firm-level marks, not distributions, audited multiples, or Rosen’s personal wealth. After removing both, no compatible public cash series supports a third-place ranking, a signed aggregate return, or a claim of repeatable personal alpha. TIME, January 23, 1984, Ben Rosen profile · top-outcome removal test

The package therefore teaches a bounded process, not a greatness verdict. Its strongest durable lessons are to freeze the customer job and platform constraint, preserve strong teams through a cheap idea revision, separate technical proof from commercialization capital, specify decision rights before entry, preserve partner/founder/vehicle attribution, and refuse to turn paper marks or company transaction values into investor returns. seven no-hindsight labs · eleven model inventory

Central-claim citation audit

The audit reopened every high-risk class required by the REVIEW gate. “Pass” means the resolved source supports the bounded adjacent claim and locator; it does not turn a public fragment into a complete deal or return record. Corrections made during REVIEW remain visible in the result column.

Ordinary-claim reproducible spread

The sample was fixed before reopening: the first bounded factual or corpus claim under the first substantive H2 of each of the seven pre-review core documents, then the next denominator/chronology claim in alternating documents, plus the first factual premise in this review after review.md existed. This produces 15 claims across all eight core documents rather than a convenient cluster.

What could falsify the thesis

The favorable thesis predicts dated pre-decision market maps or memos that name the user job and platform constraint; source/referral records showing that Rosen’s research surface preceded access; repeat preservation of strong teams through bounded idea tests; Rosen-specific roles concentrated in PC/software rather than partner-led sectors; and decision-relevant board interventions tied to the original risk. profile predictions

It weakens if original Compaq/Lotus/Ansa records show that the application, wedge, or people-first language arose only after outcomes; if a complete accepted/rejected cohort shows no advantage in Rosen’s mapped domain; if Ben-attributed deals do not outperform partner- or successor-led cases after vintage, stage, sector, ownership, access, and syndicate controls; or if board minutes show ceremonial rather than decision-relevant work. It fails as a personal-alpha thesis if full cash flows and role records show that founder execution, L. J./younger-partner judgment, syndicate access, era, and Lotus/Compaq concentration explain the result. skeptical case

The observed adverse set already constrains the story. Synapse is too opaque to teach; Osborne lacks a control counterfactual; Borland is a visible false negative; Rosen Motors failed commercially despite technical work; Eclipse lost retained exposure; Compaq’s DEC acquisition shows active governance can approve a bad strategic move; and five of 23 public paths remain unresolved. failure postmortems · denominator

Attribution and denominator audit

Independent underlying sources were used for more than five material deal/team checks. Compaq is controlled by Ben’s transcript, L. J.’s partner account, Kleiner Perkins, and the issuer proxy; Lotus by Ben and Kapor; Cypress/Crystal by Dataquest and L. J.; EA by Ben’s firm-level account and Kleiner Perkins’ board history; Ask Jeeves by Conway’s recollection and the issuer prospectus; Rosen Motors/Capstone by contemporary press and the issuer prospectus. These checks support shared work and vehicle separation, not sole-Rosen credit. decision-rights map · Compaq proxy · Capstone prospectus

Mechanical reconciliation passes: 65 source routes = 20 ★★★ + 41 ★★ + 4 ★; 59 processed and 6 metadata-only/unavailable/excluded; 29 notes = 17 writing + 12 communication; 23 public paths = 18 sufficiently reconstructed + 5 unresolved, with 8 excluded controls and 0 complete cash-return paths; the separate March-1990 census remains 45 reporter-framed entities, 14 confirmed/plausible names, 1 cutoff unresolved, and 30–31 unallocated. The 18/23 figure is named-set research coverage, not a portfolio hit rate; the firm coverage fraction is not computable. source reconciliation · deal reconciliation

The ledger now distinguishes source, sponsor, diligence, approval, negotiation, board, recruiting, follow-on, exit, and narration for each material case, preserving unknown instead of letting the best-known person absorb the role. Fund, personal, family, public-company, and successor-firm capital stay separate. ten-role assignments · vehicle analysis

Transfer limits for modern VCs

The transferable core is diagnostic, not stylistic. Research/convening transfers only when it produces lawful differentiated learning and a scored funnel. People-first transfers only with role-specific evidence and cheap falsifiable tests. Compatibility/wedge analysis transfers only after mapping incumbent control, complements, switching, and commoditization. Active governance transfers only with founder consent, named rights, board capacity, and an intervention audit. Industrial conviction transfers only with certification, yield, offtake/integrator, cost-down, total-capital, and reserve gates. dated transfer table

What does not transfer automatically is equally important: a 1981 small-fund bilateral veto is not a modern multi-stage IC; an IBM-compatible beachhead is not generic proof that compatibility creates a moat; product use is sample-biased; a board seat is not value-add; a first check without reserves is not retained exposure; and public content reach is not proprietary deal flow. The package’s named 2026 examples are factual context from filings/rulemaking, while every Rosen-to-modern application is labeled researcher inference or analyst-created hypothetical. model transfer tests

Unresolved evidence

No public route produced the Fund I–III partnership agreements, LP reports, audited cash flows, capital accounts, complete 45-company or lifetime lists, screening/pass logs, write-off and follow-on schedules, original Compaq/Lotus/Ansa memos, board/IC minutes, or disposition records. Those gaps prevent audited ownership, reserves, carry, DPI, TVPI, IRR, cash attribution, hit rate, and personal alpha. retrieval ledger

The direct record also remains incomplete: the 1973–82 newsletter run is fragmentary; several cataloged issues expose metadata or derivative quotation only; the complete Raising Money foreword and full 2019 Bloomberg interview are unavailable; Silicon Cowboys has no verified complete transcript in this environment; and private firm/board archives remain closed or unlocated. These are precise access boundaries after logged routes, not content analyzed from titles. writing gaps · talk gaps

The most decision-relevant remaining retrieval is mundane rather than celebratory: ordinary deals, passes, losses, votes, labor attribution, dilution, distributions, and stop decisions. Any of those could materially change the edge thesis, the outlier test, or the favorable-versus-skeptical balance.

Twenty-point acceptance evidence

Reader conclusion

Ben Rosen’s historical importance is better supported than a claim of audited, repeatable investment alpha. The evidence-backed curriculum is an information-and-decision discipline: create a useful research surface, identify the customer job and platform constraint, separate people from the first proposal, acquire rights proportional to the intended work, map every contributor and vehicle, and keep an ordinary-case denominator. Its strongest safeguard is the one Rosen’s legend most needs: do not let Compaq and Lotus erase Synapse, Osborne, Borland, Rosen Motors, Eclipse, DEC, the unresolved 30–31-company historical census, or the difference between paper value and cash.

Direct writing and archive corpus

• ★★★ | Rosen Electronics Letter VisiCalc reaction | http://www.bricklin.com/history/rosenletter.htm | mid-1979; two-page scans | reproduced direct writing; reliability A/B for custody; access full | Earliest recoverable Rosen thesis that application software could pull PCs into business use; scan date must control, not HTML metadata.
• ★★★ | Through Rosen-Colored Glasses RSS archive | http://www.brosen.com/files/blog.xml | 2007-11 to 2009-12; 43 items | first-party metadata and abridged excerpts over invalid TLS; reliability A/B; access partial feed plus full individual HTML permalinks under TLS exception | Forty-one substantive Ben-authored posts plus two admin notices; 13 investing, technology, company-operations, biographical or decision-relevant posts deep-processed and 28 substantive posts routed.
• ★★ | The Merger That Worked | https://www.huffpost.com/entry/the-merger-that-worked-co_b_95873 | 2008-04-09 | first-person post; reliability A/B; access full | Compaq/HP complementary-assets thesis, management execution diagnosis and explicit stock/nonparticipation disclosure.
• ★★ | Memories of Steve | https://www.huffpost.com/entry/steve-jobs-remembered_b_1026789 | 2011-10-24 | first-person post; reliability A/B; access full | Apple II/VisiCalc, analyst conflict and Jobs correspondence; retrospective and memorial selection limits.
• ★★ | A Miserable Old World, A Better New World | https://corp-realty.com/wp-content/uploads/2021/02/Rosen_Recipe.pdf | 2020 | first-person essay; reliability A/B; access full PDF | Late-life systems and public-policy forecast, not a venture memo.
• ★★ | Rosen Electronics Letter, January 6, 1983 | https://archive.org/details/release-1.0-01-83 | 1983-01-06 | organizational newsletter; reliability A/B; access full | Rosen-branded but unbylined after Dyson transition; processed as archive evidence, not Ben-authored prose.
• ★★ | Rosen Electronics Letter, March 31, 1980 | https://www.princeton.edu/~ota/disk3/1983/8314/831409.PDF | cited in OTA report | metadata/limited quotation; reliability C; access partial | “Quality” manufacturing piece; original issue unrecovered.
• ★★ | Rosen Electronics Letter, July 14/15, 1980 Semiconductor Forum | https://www.jec.senate.gov/reports/97th%20Congress/International%20Competition%20in%20Advanced%20Industrial%20Sectors%20-%20Trade%20and%20Development%20in%20the%20Semiconductor%20Industry%20%281079%29.pdf | 1980 source cited in congressional report | derivative metadata; reliability C; access partial | Original issue and exact authorship unrecovered.
• ★★ | Rosen Electronics Letter, August 24, 1981 IBM PC issue | https://www.mcjones.org/halpern/catalog.html | 1981-08-24 | archive-catalog metadata; reliability C; access unavailable | High-value archival target; box number may have changed after transfer.
• ★ | June/November 1981 PC issues | https://museum.syssrc.com/artifact/exhibits/140/ | 1981 | museum paraphrase; reliability C; access unavailable | Xerox 820 and IBM PC review leads; no scan used.

Interviews, oral histories and speeches

• ★★★ | CHM Oral History of Benjamin “Ben” Rosen | https://archive.computerhistory.org/resources/access/text/2019/07/102738545-05-01-acc.pdf | recorded 2017-09-12; 34 pp. | direct transcript; reliability A/B; access full | Core chronology, process, decisions, failures, attribution and self-critique; late memory slips are marked.
• ★★★ | Practical Computing interview | https://device.report/m/61ab47729c4f894642cfa32d0d5c96b0a83d8267a213271227352a618c09cec4.pdf | 1986-04; printed pp. 80–81 | contemporaneous Q&A scan; reliability A/B; access full | People/need/product hierarchy, active boards, Osborne lesson, Ansa screening and software forecasts.
• ★★★ | Risky Business, Inc. | https://www.inc.com/magazine/19900301/5070.html | 1990-03-01 | reporter profile plus Q&A; reliability A for visible answers/B for editorial quantities; access partial legacy body ending after third question | Two Rosen answers support outlook/authority transition; fund sizes, 45-entity denominator and early-failure process are reporter-framed, not audited or direct answers.
• ★★ | Boston Computer Update interview | https://dtvgroup.com/BCS/ComputerUpdate.pdf | 1982-04; printed pp. 23–24 | contemporaneous interview; reliability A/B; access full | PC market adoption, business/home uses, falling cost and ease-of-use forecasts.
• ★★ | Two Brief Conversations with Ben Rosen, Byte | https://www.worldradiohistory.com/Archive-Byte/80s/Byte-1987-01.pdf | 1987-01; printed p. 6 | reported conversations; reliability B; access full OCR | Rosen updates his software opportunity view after the 80386.
• ★★ | Four Million Transistors lecture report | https://calteches.library.caltech.edu/3592/ | 1988 | report of four separate public lectures; reliability B; access full report/partial Rosen remarks | Hyper-growth, founders/technology/capital and admitted Borland pass; not a panel or verbatim transcript.
• ★★ | Raising Money foreword excerpt | https://calteches.library.caltech.edu/614/2/Books.pdf | 1990; printed p. 39 | reproduced direct excerpt; reliability A/B; access partial | Business decisions lack a unique engineering-style answer; complete foreword missing.
• ★★ | Compaq retirement email | https://www.theregister.com/2000/09/29/compaqs_rosen_steps_down/ | 2000-09-28/29 | third-party email reproduction; reliability B; access full | Succession and independent-board rationale for combined chair/CEO.
• ★★ | Compaq Promises Better Days | https://www.wired.com/1999/04/compaq-promises-better-days/ | 1999-04-22 | speech report/direct excerpts; reliability B; access partial speech | Execution, bureaucracy and recovery priorities amid shareholder criticism.
• ★★ | Compaq: New Strategies Will Pay Off | https://www.washingtonpost.com/archive/business/1992/05/15/compaq-new-strategies-will-pay-off/428ea2d4-12c7-48ef-b29a-732e3e76f0ff/ | 1992-05-14 page/dateline | annual-meeting report; reliability B; access partial speech | Post-Canion strategy and governance defense; URL date differs from controlled page date.
• ★★ | Bloomberg Markets/Yahoo interview | https://www.yahoo.com/news/early-tech-backer-journey-steve-090014165.html | 2019-10-04 | Q&A syndication; reliability A/B; access partial | Timing, networks and Apple recollections; canonical Bloomberg body blocked.
• ★★ | Silicon Cowboys | https://www.imdb.com/title/tt4938484/ | 2016 | documentary appearance; reliability C until acquired; access unavailable in research environment | Appearance confirmed, stable full transcript not found; no content claims rely on it.

Biography, people and institutional controls

• ★★★ | Caltech official biography | https://board.caltech.edu/board-members/mr-benjamin-m-rosen-chair-emeritus | current; accessed 2026-08-23 | institutional biography; reliability A/B; access full | Degrees, employment, GP dates, portfolio examples and civic roles.
• ★★★ | Caltech 1954 commencement | https://campuspubs.library.caltech.edu/2503/1/June_11%2C_1954.pdf | 1954; printed p. 5 | contemporaneous institutional record; reliability A; access full | Full name Benjamin Maurice Rosen and degree.
• ★★ | Caltech 1988 profile | https://calteches.library.caltech.edu/607/1/ES52.1.1988.pdf | 1988 | institutional profile; reliability B; access full | Board roster and Borland pass admission.
• ★★ | Caltech 1989 profile | https://campuspubs.library.caltech.edu/2438/1/1989_04_23_02.pdf | 1989 | institutional profile; reliability B; access full | Newsletter timing and trustee context.
• ★★ | Caltech trustee-chair announcement | https://hr.caltech.edu/news/sally-ride-david-lee-named-caltech-trustees-ben-rosen-named-trustee-chair-440 | 2001 | institutional announcement; reliability A/B; access full | Trustee and chair chronology.
• ★★ | Caltech Millikan Medal | https://associates.caltech.edu/news/caltech-awards-millikan-medal-ben-rosen-bs-54-84219 | 2018 | institutional announcement; reliability A/B; access full | Award and retrospective biography.
• ★★ | Caltech Rosen Bioengineering Center | https://rosen.caltech.edu/about | current | institutional page; reliability A/B; access full | 2008 family philanthropy, not investment performance.
• ★★★ | L. J. Sevin CHM oral history | https://archive.computerhistory.org/resources/access/text/2013/05/102702194-05-01-acc.pdf | recorded 2009; pp. 20–25 | partner witness; reliability A/B; access full | Fund sizes, veto practice, Compaq check recollection and retirement.
• ★★★ | Jon Bayless preserved firm biography and independent deal controls | https://www.srfundstest.com/jon-bayless/ | current preserved firm biography; accessed 2026-08-23 | successor-firm retrospective; reliability B; access full | Direct firm attribution that Bayless led early Ciena and Landmark investments and chaired both; independently controlled by Ciena's 1998 proxy, Proposal 1 and Landmark founder Roice Nelson's financing and start-up account.
• ★★★ | Mitch Kapor CHM oral history | https://archive.computerhistory.org/resources/text/Oral_History/Kapor_Mitch/Kapor_Mitch.oral_history.2006.102657943.pdf | 2006; pp. 11–13 | founder witness; reliability A/B; access full | Why Kapor chose Rosen and limits of Rosen's operating contribution.
• ★★ | Mitch Kapor workshop oral history | https://archive.computerhistory.org/resources/access/text/2012/04/102658198-05-01-acc.pdf | 2004; pp. 4–5 | founder witness; reliability A/B; access full | Lotus seed and founder contribution.
• ★★ | Tommy Unterberg CHM oral history | https://archive.computerhistory.org/resources/access/text/2021/05/102792231-05-01-acc.pdf | oral history; p. 136 | peer witness; reliability A/B; access full | Fundraising/underwriting relationship and Compaq context.
• ★★ | Esther Dyson papers finding aid | https://archive.computerhistory.org/resources/access/text/finding-aids/102733962-Dyson/102733962-Dyson.pdf | 2015 | archive metadata; reliability A; access full | Rosen Research acquisition and highest-value undigitized archive route.
• ★★ | Morgan Stanley Apple history | https://ourhistory.morganstanley.com/stories/new-horizons-with-new-challenges/story-1980-apple | institutional retrospective | witness-based company history; reliability B | Rosen's Apple II demonstration and banker introduction.
• ★ | ETHW biography | https://ethw.org/Benjamin_M._Rosen | current | community biography; reliability C; access full | Discovery/identity cross-check only.

Deals, vehicles and outcome controls

• ★★★ | Compaq 2001 proxy | https://www.sec.gov/Archives/edgar/data/714154/000089155401501345/d25040_def14a.htm | 2001 | SEC filing; reliability A; access full | Chair tenure 1982–2000 and retirement.
• ★★★ | Compaq 1999 10-K | https://www.sec.gov/Archives/edgar/data/714154/000091205700008116/0000912057-00-008116-d2.pdf | 2000 filing | SEC filing; reliability A; access full | Interim executive role and 200,000 option grants; no personal return inference.
• ★★★ | Kleiner Perkins Compaq retrospective | https://www.kleinerperkins.com/perspectives/compaq/ | firm archive | co-investor retrospective; reliability B; access full | Product/channel thesis and shared creation.
• ★★★ | Kleiner Perkins Electronic Arts retrospective | https://www.kleinerperkins.com/perspectives/electronic-arts/ | firm archive | co-investor retrospective; reliability B; access full | KPCB sourcing/board attribution; prevents Rosen over-credit.
• ★★★ | Dataquest I.C. Start-ups 1987 | https://archive.computerhistory.org/resources/access/text/2013/04/102723181-05-01-acc.pdf | 1987; pp. 122–36, 244 | contemporary industry report; reliability B+; access full | Cypress/Crystal financing and LSI anti-attribution control.
• ★★★ | Ask Jeeves 424B4 prospectus | https://www.sec.gov/Archives/edgar/data/1054298/000095014999001225/0000950149-99-001225.txt | 1999-07-01; printed pp. 60–65 | issuer SEC filing; reliability A; access full | Ben's four direct purchases, beneficial ownership and IPO-price paper mark; separates CPQ Holdings corporate capital and leaves dispositions unresolved.
• ★★★ | Capstone Turbine 424B1 prospectus | https://www.sec.gov/Archives/edgar/data/1009759/000095015000000546/e424b1.txt | 2000; Product Development milestone table, printed p. 30; director biographies, printed p. 49; Certain Transactions, printed p. 59; Principal Shareholders table and notes (5), (6), and (11), printed pp. 60–62 | issuer SEC filing; reliability A; access full | Ben/Harold's 1993 investment and contemporaneous board roles; Ben's separate 3,492,621 beneficial shares; Sevin Rosen's 4,120,967 shares represented by John Jaggers; and the $590,000 Rosen Motors related-party asset purchase.
• ★★ | TIME, Lotus paper value | https://content.time.com/time/subscriber/article/0,33009,949963-3,00.html | 1984-01-23 | contemporaneous report; reliability B; access full | Reported 1983 IPO marks of $2.1m-to-$70m for Lotus and $2.5m-to-$40m for Compaq; not audited proceeds.
• ★★ | Ansa acquisition report | https://www.techmonitor.ai/technology/ben_rosen_to_join_board_of_borland_under_agreement_for_acquisition_of_ansa/ | 1987-07-12 | trade report; reliability B; access full | Reported acquisition and Rosen board role.
• ★★ | Sevin Rosen preserved portfolio | https://www.srfundstest.com/portfolio/ | current archive | successor-firm retrospective; reliability B; access full | Firm association only; sponsor, vehicle and economics need separate proof.
• ★★ | IBM personal-computer history | https://www.ibm.com/history/personal-computer | institutional history | reliability A/B; access full | Market inflection context, not Rosen decision evidence.
• ★★ | CHM 1981 timeline | https://www.computerhistory.org/timeline/1981/ | institutional history | reliability A/B; access full | Contemporaneous category context.
• ★★ | 1981 SEC venture-capital speech | https://www.sec.gov/news/speech/1981/061281friedman.pdf | 1981 | regulator speech; reliability A; access full | Regulatory/fundraising context, not Rosen-specific conduct.
• ★★ | Federal Reserve Staff Study 168 | https://www.federalreserve.gov/pubs/staffstudies/1990-99/ss168.pdf | historical study | government research; reliability A/B; access full | Capital-gains/ERISA and venture-market context.
• ★ | 2026 IPO equity reconstruction | https://www.startup-book.com/wp-content/uploads/2026/03/20260228-Equity-List-Lebret.pdf | 2026 | secondary reconstruction; reliability C/B-; access full | Compaq/Lotus cap-table lead contains a suspicious Lotus date; excluded from computed returns.

Failures, criticism, litigation and regulatory controls

• ★★★ | Berger v. Compaq | https://www.ca5.uscourts.gov/opinions/pub/00/00-20875.cv0.wpd.pdf | 2002 | official appellate opinion; reliability A; access full | Rosen named in allegations; reversal addressed class certification, not merits.
• ★★★ | HP 2002 10-K | https://www.sec.gov/Archives/edgar/data/47217/000104746903002015/a2096381z10-k.htm | 2003 filing | SEC filing; reliability A; access full | Company settlement and insurance; no admission or Rosen-specific finding.
• ★★★ | LA Times, Rosen Motors closes | https://www.latimes.com/archives/la-xpm-1997-nov-19-fi-55325-story.html | 1997-11-19 | contemporaneous report/direct comments; reliability B+; access full | $24m, 70 employees, no automaker commitment and commercial failure.
• ★★ | LA Times, Rosen brothers/Capstone | https://www.latimes.com/archives/la-xpm-1997-03-05-mn-35031-story.html | 1997-03-05 | contemporaneous profile; reliability B; access full | Family/team boundary and company context.
• ★★ | Inc., Osborne loss | https://www.inc.com/magazine/19860201/7497.html | 1986-02-01 | contemporaneous report/direct remarks; reliability B+; access full | First-person $400,000 loss language and active-board lesson; TIME assigns the same amount to the fund, so capital ownership remains unresolved.
• ★★ | Washington Post, Canion ouster | https://www.washingtonpost.com/archive/business/1991/10/26/compaq-ousts-president-who-founded-firm/69510cac-24ef-45d7-95af-b83ed96b9d83/ | 1991-10-26 | contemporaneous report; reliability B; access full | First CEO transition, loss/layoff context and surprise.
• ★★ | Washington Post, CEO search | https://www.washingtonpost.com/archive/business/1999/07/23/compaq-insider-tabbed-as-new-ceo/438f7b11-20af-438b-be5d-03daeb756f9a/ | 1999-07-23 | contemporaneous report; reliability B; access full | Governance criticism and successor search.
• ★★ | TheStreet, CEO search criticism | https://www.thestreet.com/technology/compaqs-quest-for-a-ceo-754202 | 1999 | contemporaneous commentary; reliability B/C; access full | Explicit concern that Rosen's continued role deterred candidates.
• ★★ | Pfeiffer criticism | https://www.computerworld.com/article/1371834/former-compaq-ceo-says-his-ouster-damaged-company.html | 1999/2000 | interested-witness interview; reliability B/C; access full | Adverse governance testimony, not adjudicated fact.
• ★★ | USC CEO appraisal paper | https://ceo.usc.edu/wp-content/uploads/1997/10/1997-15-g97_15-CEO_Appraisal.pdf | 1997 | governance study/interview; reliability B; access full | Rosen on separate chair/CEO and board appraisal.
• ★★ | 2006 canceled fund, NYT mirror | https://www.wcapgroup.com/wp-content/uploads/2013/07/NYTimes-October.7.2006.pdf | 2006-10-07 | contemporary report mirror; reliability B/C custody; access full | Later successor-firm event; explicitly outside Ben's 1981–95 active GP record.
• ★★ | Institutional Investor, No Exit | https://www.institutionalinvestor.com/article/2btfvr3z1yqmxt9f0swlc/home/no-exit | 2006 | contemporary report; reliability B; access full | Later fund-cancellation rationale and market context, not Ben attribution.
• ★ | SEC 1970 administrative decision | https://www.sec.gov/files/alj/aljdec/1970/id19700626web.pdf | 1970; printed p. 92 | regulator decision; reliability A; access full | Identifies Rosen as Coleman analyst/source, not respondent or wrongdoer.

Coverage reconciliation and open archive work

• Inventory: 65 annotated routes: 10 writing/archive, 12 interview/speech/direct-communication, 15 biography/people, 15 deal/context and 13 adverse/legal routes. Priority split: 20 ★★★, 41 ★★ and 4 ★.
• Access: 59 routes were processed within their stated full or partial boundary; six are metadata-only, unavailable or excluded from claim support. A found URL is never treated as a processed body.
• Writing denominator: the RSS archive contains 43 items: 41 substantive Ben-authored posts and two admin notices. Thirteen investing, technology, governance, operating, biographical or decision-relevant posts are deep-processed from individual HTML bodies; 28 substantive items are routed. Outside the feed, four standalone works are full (1979 VisiCalc scan, 2011 Jobs remembrance, 2020 essay and unbylined 1983 issue) and four archive leads are partial/unavailable. The HuffPost merger page is duplicate custody of a feed work.
• Talk denominator: 12 distinct interview, speech, feature, email, foreword or documentary families are mapped. Eleven have usable full or partial attributable text; Silicon Cowboys is metadata-only/unavailable. The full CHM oral history is the anchor; reports, excerpts and emails retain their form boundaries.
• Deal denominator: the public opportunity set contains 23 named paths: all were searched, 18 are attribution- or decision-bounded and five remain association-only. That set includes passes, personal capital, later deals and projects, so it is not the Inc. census. Within Inc.'s reporter-framed 45 entities financed by March 1990, 14 names are confirmed or plausibly contemporaneous, Citrix's cutoff is unresolved, and 30–31 historical identities remain unallocated. Nine named paths or controls are outside the 45: Apple, Borland, Eclipse Aviation, LSI Logic, Ciena, Capstone, Ask Jeeves, Energy Vault and Rosen Motors. No cash-return path is reconstructable. This is research coverage, not a hit rate.
• Identity and attribution controls: Ben/Benjamin M./Benjamin Maurice Rosen are included. Brother Harold's Navy record, unrelated same-name people, L. J.-led semiconductor deals, Bayless-led later deals, KPCB-led board work, successor-firm post-1995 events and Rosen-branded unbylined Dyson-era writing remain separated.
• Highest-value gaps: the 1973–82 Rosen Electronics Letter run; Rosen Research corporate files; Sevin Rosen partnership agreements, LP reports, IC and board minutes; complete opportunity/pass/write-off/follow-on and cash-flow ledgers; audited Fund I/II returns; personal capital accounts; original Compaq/Lotus/Ansa memos; full prospectuses; private settlements; and stable Silicon Cowboys transcript.
• Negative-search boundary: exact-name SEC, DOJ, FINRA, FTC, court, LP-suit and enforcement searches found no Rosen-specific enforcement disposition. That does not cover private, sealed, non-U.S. or unindexed records and is not legal clearance.

Source notes