Vinod Khosla

1. Snapshot and testable edge thesis

Vinod Khosla was a Sun co-founder and KPCB general partner before building Khosla Ventures from an initially own-capital arrangement into a multi-vehicle firm. The legal and economic actors include KPCB funds, KV main/seed/opportunity/strategic/AIV vehicles, personal and family entities, Khosla Impact, Khosla Labs, and SPAC sponsors; none is interchangeable. Final biography · capital silos [contemporaneous record · strong]

Testable edge thesis. In early company formation from the mid-1990s through 2026, Khosla's proposed repeatable advantage is a coupled search-and-building system: seek high-consequence discontinuities before categories are obvious; prefer first-principles founders; decompose fatal risks; spend first on discriminating evidence; recruit backward from the risk map; and use patient, assistance-heavy capital to preserve the successful tail. Black Swans, PDF pp. 2–6 · Risk Matrix, physical pp. 3–10 · Venture Assistance, decision sections [researcher inference · moderate]

The mechanism should be strongest when capability curves are nonlinear, incumbent incentives suppress a new architecture, uncertainty can be retired before irreversible scale, and one outcome can matter to the vehicle. It should be weakest when technical, construction, regulatory, financing and adoption risks are common-cause or when entry price, dilution and duration absorb the upside. Cleantech cases [researcher inference · strong]

The favorable thesis predicts dated category work before consensus, explicit risk registers, small first checks relative to evidence-backed follow-ons, risk-specific recruiting, and documented abandonment after failed milestones. It is falsified if these appear only in retrospective winners, ordinary outcomes do not improve when the mechanisms are present, follow-ons follow sunk cost, partner work explains the visible record, or audited economics disappear outside one or two outliers. Falsifiable profile [researcher inference · strong]

The searched denominator is 168 public leads: 27 sufficiently reconstructed, 134 unresolved and seven excluded; 27/161 = 16.8% is only an eligible-public-lead reconstruction rate. It is not career coverage, a hit rate or a return sample. Ledger reconciliation [researcher inference · strong]

2. Core philosophy

Khosla's own doctrine begins with consequence rather than consensus probability. A qualifying attempt should attack a material problem, create a non-incremental state, and have a path to scale economics; accepting a high probability of failure is defensible only when success changes the system. Black Swans, pp. 2–6 [investor-stated · strong]

The second layer makes audacity operational. Identify what kills the plan, quantify criticality and uncertainty, map milestones to risk reduction, and show the time and capital required. Risk Matrix, Steps 1–5 [contemporaneous record · strong] The artifact is KV-branded institutional guidance; its personal authorship and adoption denominator are unknown. [researcher inference · insufficient public record]

The third layer is human. Khosla repeatedly values learning velocity, recruits from organizations that solved each relevant subsystem, and asks investors to earn influence through assistance while preserving founder agency. Gene Pool Engineering, seven-step process · Pioneers of AI, team-risk H2 · board essay, decision-making passages [investor-stated · strong]

The synthesis is not “be reckless.” It is: buy asymmetric option value, but make every early dollar purchase information; raise burn only after uncertainty falls; preserve an abandonment path; and separate social consequence from LP return. Intelligent Failure, “90-percent” passage · Company Building 102, 00:05:49–00:10:25 [researcher inference · strong]

3. Sourcing

Problem and category first. The 1982 Sun plan starts with a cost-performance and OEM-system wedge, not a generic founder funnel. Later work searches infrastructure-scale health, energy, transport and AI problems where conventional supply cannot meet demand. Sun plan, pp. 1–3 / exhibit pp. 23–25 · Reinventing Societal Infrastructure, opening and “necessary, though not sufficient” [contemporaneous record · strong]

Instigators and outsiders. Familiar categories can be a warning because incumbents and experts often extrapolate from installed products. The investable interpretation is conditional: outsiders own the hypothesis, while experts test physics, regulation, safety and scale. Peak XV, 00:01:09–00:04:09 · Udacity, outsider/incumbent blocks [investor-stated · strong]

Access advantage. Sun, Stanford, KPCB, founders, LPs, advisers and the KV brand create privileged referral and recruiting networks. The 116-card current portfolio is curated marketing and cannot show whether category-first search generated a superior or repeatable funnel. Current portfolio denominator [researcher inference · strong]

Repeatability test. A complete source-channel ledger should distinguish founder/referrer, proactive thesis search, partner, Labs/Impact, personal and inbound routes; match them by stage, sector and vintage; then compare pass and outcome distributions. No such ledger is public. Attribution retrieval boundary [researcher inference · insufficient public record]

4. Picking

The core selection pair is a founder with high learning velocity and an important future category that others do not yet recognize. Khosla follows capability curves and talent migration rather than demanding a finished market map. Pioneers of AI, learning-rate and OpenAI H2s · Peak XV, 00:03:31–00:09:30 [investor-stated · strong]

Founder assessment is incomplete without a risk map. The founder must identify the five ways the plan dies, learn faster than those risks change, recruit missing capability, and change tactics without abandoning the mission. Company Building 102, 00:02:00–00:10:25 [investor-stated · strong]

Technical novelty is necessary but not sufficient in health, climate, transport and regulated infrastructure. Diligence must include manufacturing, distribution, reimbursement or offtake, permits, safety, trust, security, supply chain and capital formation. Societal Infrastructure · KiOR decision freeze [researcher inference · strong]

Red flags are irreversible scale before integrated proof, milestones defined as activity, expert dismissal without a falsifier, teams hired from comfort rather than risk, mission rhetoric without unit economics, and a board that cannot surface bad news. Risk Matrix, Steps 1–5 · Coskata case [researcher inference · strong]

Decision rights remain incompletely public. Firm biographies and issuer filings identify selected partner and board roles, but IC vote, veto, allocation and dissent records are missing for most names. People map [researcher inference · insufficient public record]

5. Deal and portfolio mechanics

Khosla's public doctrine prefers many shots at fund-material outcomes over high-confidence incremental outcomes. Black Swans, p. 3 [investor-stated · strong] That is a power-law selection claim, not evidence of a literal 80/20 split or superior net return. Pareto boundary [researcher inference · insufficient public record]

A disciplined implementation separates four decisions. This is a researcher-designed transfer from the fatal-risk model and vehicle/decision-rights ledger, not a recovered Khosla IC checklist. [researcher inference · strong]

1. Entry: consequence if right, fatal risks, cheapest proof, price, ownership and check size. Fatal-risk decomposition · discriminating experiment · option sizing [researcher inference · strong]
2. Reserve: what milestone earns the next dollar and whether the fund can preserve meaningful ownership. Staged capital · fresh re-underwriting · fund constraint [researcher inference · strong]
3. Governance: who owns each decision, what information rights exist, and how conflicts are handled. Founder agency and earned assistance · capital silos and decision rights [researcher inference · strong]
4. Liquidity: company duration, vehicle life, distributions, secondary options and stakeholder incentives. Patient-capital boundary · vehicle economics [researcher inference · strong]

Public records expose only fragments. Rocket Lab's 13D traces aggregate Series A–E/E1 purchases; Square's S-1 shows KV III ownership; KiOR exposes fund, trust and VNK instruments; RingCentral separates VNK direct shares from Vinod's indirect control. None gives a complete decision-to-cash-flow chain. Rocket Lab 13D, Items 3/5/6 · Square S-1, Principal Stockholders · ownership controls [contemporaneous record · strong]

Price is not irrelevant. Consequence and option size must be separated from security, dilution, reserve burden and fund materiality. Public direct voice supports tolerance for uncertainty, but complete entry-price and IRR calculations are unavailable. Vator post-seed note · AI valuation note [investor-stated · provisional]

6. Vehicle and incentive fit

The strategy fits patient capital only if reserves, liquidity and governance are designed for long technical clocks. It misfits any vehicle whose life, concentration or stakeholder obligations force scale before proof. Vehicle economics and incentives [researcher inference · strong]

7. Team attribution and work with founders

Vinod has unusually strong formation evidence for Sun, Juniper and Cerent, but even those cases are team outcomes. Bechtolsheim, Joy and McNealy own distinct Sun roles; Pradeep Sindhu and Scott Kriens built Juniper; Cerent required founders, recruited executives, KPCB and Cisco. People map in biography [contemporaneous record · strong]

KV attribution is more distributed. NJ names Samir Kaul and David Weiden as co-founders; Sven Strohband is linked to Rocket Lab; Keith Rabois's official biography credits DoorDash, Affirm, Faire, Stripe and Opendoor. NJ memo, p. 4 · Sven/Rocket evidence · Keith biography [contemporaneous record · strong]

The claimed founder contract is high-candor assistance without shadow management: recruit, question, expose tradeoffs and let the founder decide. Venture Assistance · Upfront, 06:47–07:39 and 19:01–21:36 [investor-stated · strong]

Governance is the limit. View, KiOR and Nextdoor show why formal governance, disclosures, related-party financing, independent challenge and minority-holder treatment must be analyzed separately from the earned-influence philosophy. Adverse cases [researcher inference · strong] No public advice denominator shows what help was offered, accepted, rejected or causally useful. [researcher inference · insufficient public record]

8. Views on the craft

• Power law: a small number of consequential successes must justify many failures, but social impact, gross deal multiple, fund return and personal return are separate. Black Swans, p. 3 [investor-stated · strong]
• Fund size: an opportunity must be material to the vehicle; no public universal check-size or concentration formula is recovered. 20VC, risk/IRR blocks [investor-stated · moderate]
• Access versus selection: brand and networks matter, but unfamiliar categories and ignored founders can create search advantage; no source-channel outcome study exists. Peak XV, 00:01:09–00:09:30 [researcher inference · moderate]
• Governance: boards should focus on unresolved questions and assistance, not reporting theater; this direct-voice source states an operating philosophy, not formal authority under governing law or corporate documents. Company Building 102, 00:33:45–00:40:57 [investor-stated · strong]
• Exits: create fundamental value rather than an early exit plan; vehicle duration and stakeholder liquidity still require explicit underwriting. Pioneers of AI, long-term-investor H2 [investor-stated · moderate]
• Great investor: searches nonconsensus futures, helps the team learn and recruits against risk. Whether that produced comparative alpha is not computable. talks framework [researcher inference · strong]

9. Signature bets and decision process

The decision process is best studied by freezing evidence before outcome. The method is a researcher audit protocol, not a recovered Khosla rule. Fourteen case freezes and exercises [researcher inference · strong]

• Juniper: architecture and customer-consensus challenge; formation/recruiting evidence is strong, economics unaudited. [researcher inference · strong]
• Cerent: explicit market design and team formation; acquisition value is not KPCB proceeds. [contemporaneous record · strong]
• Square / Block: adaptive plan and ownership evidence; “we” and board stories require actor controls. [researcher inference · strong]
• Impossible Foods: consequential mission and patient financing through the mapped March 2020 record; allocation and realization economics remain incomplete. [researcher inference · moderate]
• OpenAI: capability-curve conviction under extreme uncertainty; decision and closing clocks are retrospective. [retrospective witness · moderate]
• KiOR: the strongest scale-up, rescue-capital and governance stress test. [contemporaneous record · strong]
• Range Fuels: a created/financed scale-up case with incomplete vehicle and stop records. [researcher inference · moderate]
• Coskata: integrated-process and project-finance risk with Samir Kaul board attribution. [contemporaneous record · strong]
• View: governance/control and disclosure evidence, with the company order distinct from the dismissed CFO case. [contemporaneous record · strong]
• Nextdoor / KVSB: SPAC sponsor, PIPE, forward-purchase and public-holder incentives. [contemporaneous record · strong]
• Excite/Google: option-preservation miss with disputed remembered terms. [retrospective witness · moderate]
• Rocket Lab: patient follow-on and Sven Strohband attribution. [contemporaneous record · strong]
• RingCentral: ownership/control ontology without a recovered initial rationale. [contemporaneous record · strong]
• SKS Microfinance: personal/family capital and social-license risk, not a KV or Impact return. [contemporaneous record · strong]

All 14 cases end with invest/pass, kill criteria, ownership/terms and next-check prompts before the actual result. They are researcher-designed exercises, not recovered IC rules. Case-by-case list above [researcher inference · strong]

10. Mental models and maxims

The field guide reconciles the complete note and phase corpus into 18 distinct reusable rules. Twelve navigation families keep related mechanisms together without erasing different decisions, evidence thresholds or failure modes. Eighteen-model field guide [researcher inference · strong]

The index merges slogans that share a mechanism and discloses six candidate exclusions plus one researcher-only protocol rather than inflating a top-N list. Every retained model carries provenance, evidence status, mechanism, context, use conditions, procedure, grounded cases, an exception, evidence placements, evolution and a dated modern translation. Models 13 and 17 remain explicitly provisional one-stream rules; Model 18 is strong forecast practice but has unproved predictive validity. [researcher inference · strong]

11. Evolution and contradictions

The central contradictions remain failure tolerance versus capital discipline; founder agency versus board duty; unsubsidized economics versus catalytic policy; open access versus strategic restriction; abundance versus distribution; and autonomy forecasts versus clinical or safety evidence. Writings contradictions [researcher inference · strong]

12. Failures and limits

KiOR, Range and Coskata show that the rule fails when integrated scale, financing and policy risks cannot be separated. View and Nextdoor add disclosure, control and security-incentive limits. Excite/Google is a miss rather than a loss. SKS is personal capital and exposes social-license risk; it is not a KV or Khosla Impact return. Failure table [researcher inference · strong]

No named, dated KV policy change is public after these outcomes. “Intelligent failure” is therefore a doctrine and a testable interpretation, not proof that every loss bought useful information or improved later results. Uncertainty and retrieval ledger [researcher inference · insufficient public record]

Selection and survivorship bias are material: the current spotlight is a 116-card curated list, while 134 of 168 searched leads remain unresolved. Denominator The record cannot compute career hit rate, firm MOIC/IRR, Vinod-personal return, partner carry or dominant-outlier contribution. [researcher inference · strong]

13. Historical operating environment

MIT reports more than $25 billion entered cleantech startups in 2006–11 and investors lost more than half; NBER and Brookings add later cohort context. Those data help interpret capital intensity but are not information Khosla possessed in final form at decision dates. Historical/base-rate chapter [researcher inference · strong]

14. Modern VC translation (as of 2026-08-22)

Present-day examples remain case-specific: current Rocket Lab filings show long-duration follow-ons; 2025–26 AI/health works show recent task and workflow theses; neither proves that every translation produces excess return. [researcher inference · strong]

15. Comparison with completed peers

Arthur Rock is the closest company-formation and governance comparator; Doug Leone is the closest institutional multi-generation partner comparator; Brad Feld is the closest technical-founder, explicit-framework and founder-support comparator. All three packages are complete and have adversarial reviews. Rock philosophy and review · Leone philosophy and review · Feld philosophy and review [researcher inference · strong]

Khosla differs most in the breadth of sector forecasts and explicit tolerance for technical failure; Rock in concentrated formation governance, Leone in institutional continuity, and Feld in public mechanics and stopping architecture provide the most useful controls. These are dimension-matched comparisons, not a ranking of returns. Rock controls · Leone controls · Feld controls [researcher inference · strong]

16. The strongest case against greatness

The skeptical thesis is that the visible record can be explained by privileged access, era, team labor, founder execution, vehicle flexibility, marketing selection and a few outliers rather than repeatable personal selection alpha. Sun and KPCB supplied network and credibility; KV's own-capital start bought unusual freedom; later institutional funds, partners and brand widened opportunity. The current portfolio spotlights survivors, while 134 of 168 searched leads remain unresolved. Denominator [researcher inference · strong]

The Juniper result is reported: 2,500x (KPCB case, outcome passage, retrieved 2026-08-22) [retrospective witness · insufficient public record]; the OpenAI result is reported: >100x (SparX, 00:38:55–00:39:34, 2026) [retrospective witness · insufficient public record]. Those figures and Cerent's announcement value dominate the narrative but cannot be translated into audited KPCB/KV or personal economics. Economic ranking KiOR, Range and Coskata show that black-swan rhetoric can rationalize correlated capital-intensive loss; View and Nextdoor test governance and security incentives. Failure cases [researcher inference · strong]

The favorable thesis survives only if hidden records show its mechanisms before outcomes and across ordinary cases: nonconsensus category work, explicit fatal-risk tests, disciplined first checks, evidence-backed follow-ons, risk-specific recruiting and useful founder assistance. It fails if those records do not exist, if partner attribution dominates, if passes and failures met the same criteria, or if matched economics do not improve. Model register · cross-case process evidence [researcher inference · strong]

The deciding evidence is a dated funnel and pass ledger, IC and board records, person/vehicle attribution, complete cash flows, reserve/stop decisions, independent founder reports and an ex-outlier return test. Until then, Khosla is best described as an unusually explicit and consequential company-building investor whose comparative alpha remains unproved. Uncertainty and retrieval ledger · review falsifiers [researcher inference · strong]

17. Analyst study guide, glossary, and go deeper

Recommended reading path

1. Read the final biography for chronology and falsifiers.
2. Read the Sun plan, Black Swans, Risk Matrix and Venture Assistance.
3. Use the spoken framework synthesis to see repetition and memory controls.
4. Replay Juniper, OpenAI, KiOR and Excite/Google through their no-hindsight exercises.
5. Apply the mental models, then attack them with the failure and evidence-gap ledgers.

Investor-specific glossary

• Black Swan: Khosla's prospective label for a low-probability technology whose success changes scale or economics; not Taleb's unforeseeable-event usage. Black Swans, p. 3 [investor-stated · strong]
• Chindia price: cost low enough for India/China-scale adoption without a permanent green premium. FRONTLINE/World, Chindia-cost passage [investor-stated · strong]
• Instigator: founder who attempts a discontinuity rather than extrapolating an incumbent plan. 12 Instigators, opening definition [investor-stated · strong]
• Intelligent failure: bounded experiment that produces decision-relevant learning; not any ambitious loss. Intelligent Failure, definition and 90-percent passages [investor-stated · strong]
• Risk matrix: KV method that weights failure-mode criticality, uncertainty, milestone risk reduction, time and spend. Risk Matrix, matrix-field definitions [contemporaneous record · strong]
• Venture assistance: claimed post-investment product of questions, recruiting and context while the founder decides. Venture Assistance, board-role and decision sections [investor-stated · strong]
• Main/Seed/Opportunity/Strategic/AIV: different KV legal/mandate vehicles, not one economic pool. Vehicle map [researcher inference · strong]
• KFT/VNK/VK Services: family, investment and control entities exposed in selected filings; not interchangeable with KV funds or direct personal ownership. Capital-silo map [researcher inference · strong]
• R/U/X: sufficiently reconstructed, unresolved or excluded lead in the public deal ledger. Ledger method [researcher inference · strong]
• No-hindsight exercise: researcher-designed invest/pass, kill, ownership and next-check prompts shown before the sourced outcome. Juniper no-hindsight example [researcher inference · strong]

Reusable sourcing checklist

The five rules below are a researcher-designed transfer from future-back search, problem-first sourcing, portfolio materiality and forecast clocks; they are not a Khosla-authored checklist. [researcher inference · strong]

• Define the high-consequence problem and the incumbent assumption that may be wrong. Future-back search · problem-first sourcing [researcher inference · strong]
• Record proactive thesis, founder/referrer and partner route before the outcome is known. Problem-first sourcing · people and route ledger [researcher inference · strong]
• Map capability, adoption and financing clocks separately. Capability and adoption clocks [researcher inference · strong]
• Ask what source-channel advantage exists and whether it is repeatable without brand. Problem-first sourcing · searched denominator [researcher inference · strong]
• Preserve passes, ordinary outcomes and failures beside wins. Many-shot portfolio · comprehensive ledger [researcher inference · strong]

Reusable diligence and IC checklist

The seven rules below operationalize the risk matrix, discriminating experiment, staged capital, risk-backward hiring and deployment tree; they are researcher-designed IC controls. [researcher inference · strong]

• Name the five fatal risks and their dependencies. Fatal-risk decomposition [researcher inference · strong]
• Specify the cheapest discriminating test, threshold, time, spend and decision consequence. Discriminating experiment [researcher inference · strong]
• Separate technical proof from manufacturing, regulation, distribution, trust and unit economics. Deployment prerequisite tree [researcher inference · strong]
• Identify which hire owns each risk and test learning velocity with work and references. Risk-backward hiring [researcher inference · strong]
• Model security, valuation, ownership, dilution, reserves, duration and ex-outlier fund materiality. Option sizing · vehicle economics [researcher inference · strong]
• Write kill criteria and the next re-underwriting date before investing. Staged capital · fresh re-underwriting [researcher inference · strong]
• Label every rationale by attribution—investor-stated, contemporaneous record, documented behavior, retrospective witness, or researcher inference—and by support—strong, moderate, provisional, or insufficient public record. Central-claim audit [researcher inference · strong]

Reusable governance and follow-on checklist

The six rules below are researcher-designed controls derived from founder agency and earned assistance, fresh follow-on underwriting and the capital-silo map, not a universal statement of law or firm policy. [researcher inference · strong]

• Document founder, board, manager, sponsor, LP and minority-holder rights. Capital silos and decision rights [researcher inference · strong]
• Keep personal, trust, fund, Impact, Labs, AIV and SPAC capital separate. Capital silos and decision rights [researcher inference · strong]
• Log advice offered, founder decision, dissent, conflict, milestone and outcome. Founder agency and earned assistance [researcher inference · strong]
• Require independent challenge for related-party finance, accounting and disclosure. Founder-agency boundary · adverse and legal records [researcher inference · strong]
• Re-underwrite every follow-on against current evidence and a counterfactual pass. Fresh follow-on underwriting [researcher inference · strong]
• Preserve a safe abandonment path and a stakeholder communication plan. Staged abandonment · patient-capital boundary [researcher inference · strong]

Ranked primary and secondary sources

1. Sun plan, PDF pp. 1–12 — best contemporaneous operating artifact.
2. Black Swans, PDF pp. 2–6 — clearest selection and portfolio doctrine.
3. Risk Matrix, physical pp. 3–10 — most explicit evidence-and-capital tool.
4. Venture Assistance, decision sections — founder/investor contract.
5. Senate Finance, pp. 1–5 and 29–30 — decision-era forecasts, policy and disclosed holdings.
6. NJ DOI memo, pp. 1–4 — strongest public vehicle, team and dated performance window.
7. Issuer and legal case ledger — strongest independent attribution/economics controls.
8. Pioneers of AI transcript — current direct voice with structured locators.

Go deeper by obtaining dated IC/pass/follow-on memoranda, the reported OpenAI LP letter, company allocation and cash-flow ledgers, LPAs and side letters, partner attribution/carry, board packets, full rejected/failed portfolio exports, independent founder surveys, and unchanged forecast scorecards. Uncertainty ledger

Pareto 80/20 — what drove the record

The public record is consistent with a power-law franchise, but it does not support a literal “80% from 20%” calculation. Company transaction values, IPO ownership snapshots, fund marks, and retrospective multiples use different numerators and clocks. The three rankings below answer different questions and never collapse KPCB, Khosla Ventures, personal/trust capital, Khosla Impact, Khosla Labs, SPAC sponsors, or partner-led cases.

Economic contribution — ranked by defensible evidence, not computed proceeds

Economic rank is therefore a research priority order, not a contribution waterfall. Source date, measurement date, metric convention, gross/net status, realized/unrealized split, and missing cash-flow inputs are stated in each row; none supports aggregating a personal or firm track record. case ledger; retrieval boundary [researcher inference; strong]

Historical importance — ranked by effect on the investing franchise

Decision-learning value — ranked by information gained from mistakes and constraints

Dominant-outlier sensitivity

A defensible exclusion test would recompute fund and career returns after removing Juniper, Cerent, OpenAI, Square, and Impossible one at a time, then jointly. That requires each vehicle’s cash-in, follow-ons, dilution, cash-out, residual value, fees/carry, and time-weighted clocks. Those inputs are absent. Dominant-outlier contribution is not computable, and no marquee anecdote is converted into a portfolio multiple. case ledger; public-LP windows [researcher inference; strong]

Evidence boundary and searched denominator

Among the 161 eligible public leads, 27/161 = 16.8% meet only this scoped public-lead reconstruction threshold. It is not historical coverage, a hit rate, return coverage, or a Vinod-personal denominator. Complete career count, historical/person coverage fraction, firmwide return, and Vinod-personal return are not computable. ledger reconciliation; source-map retrieval boundary [researcher inference; strong]

The spotlight snapshot was retrieved 2026-08-22: 116 cards, 116 outbound routes, nine category pages, 131 assignments, and 15 cross-classified identities—Upstart, Rubrik, Rocket Lab, QuantumScape, Okta, Instacart, Oscar, DoorDash, BlackBerry/Cylance, Affirm, Guardant, Blue River Technology, Block, GitLab, and Nutanix. The curated page is selection- and survivorship-biased. Aliases stay within one issuer; follow-ons, conversions, distributions, listings, and acquisitions are events, not new deals. Contrail is a Juniper acquisition relationship, not another Khosla investment. Nextdoor sponsor and PIPE instruments remain separate events within one target identity. source-map portfolio reconciliation [contemporaneous record; strong]

Evidence labels use two axes: type—investor-stated, contemporaneous record, documented behavior, retrospective witness, researcher inference, or insufficient public record—and support—strong, moderate, provisional, or insufficient public record. Repetition across Khosla/KV appearances is one interested stream, not independent corroboration. The underlying route map is sources.md; direct written and spoken evolution is in writings.md and talks.md. method boundary [researcher inference; strong]

Historical market, regulatory, and base-rate context

Retrospective base-rate context is severe but dated. MIT’s 2016 study says more than $25bn entered cleantech startups in 2006–11 and investors lost over half (study landing) (contemporaneous record / strong; retrospective base-rate use). NBER’s 2000–20 sample covers about 149,358 startups and reports clean-energy share falling from 4.9% before the 2009 shock to 3.7% after, while finding demand and outlier-potential differences (paper abstract/PDF) (contemporaneous record / moderate; retrospective base-rate use). Brookings reports 2011–16 dollars falling from $7.5bn to $5.24bn, deals from 649 to 455, sector VC share from 16.8% to 7.6%, and 87% of 2016 dollars in late stage (numbered findings) (contemporaneous record / moderate; retrospective base-rate use). No numeric base rate is manufactured for eras without a defensible sample.

Vehicle economics and incentives

Capital silos and decision rights

The 2026-05-20 Form ADV has 31 Schedule D §7.B.(1) private-fund schedules. They enumerate containers and GAV, owner counts, and related-person percentages. Those values are not added: AIVs/SPVs and strategies may overlap, and GAV is neither committed capital, realizable NAV, nor return. The form does not disclose reserves, investment periods, liquidity, distributions, carry splits, or key-person rights. (contemporaneous record / strong)

Current SEC Form ADV private-fund schedules — 31

The 2026-05-20 Other-Than-Annual Amendment, Schedule D §7.B.(1), reports the following containers. GAV is the filing’s gross asset value; owners are approximate beneficial-owner counts; related is the approximate adviser/related-person ownership percentage. These values must not be summed as AUM or performance because AIVs, SPVs, and strategies can overlap. [contemporaneous record; strong]

Every schedule reports a routine outside-investor minimum commitment of $100,000. That field is not the actual commitment of every LP. The ADV does not disclose fund-level fee/carry, investment period, reserves, liquidity, key-person clause, LPAC rights, or distributions. ADV Schedule D §7.B.(1) [contemporaneous record; strong]

Selected Form D formation/offering windows

Each filing row below is contemporaneous formation/offering evidence; every interpretation-limit cell is a researcher boundary, not a filed fund term. source-map Form D routes [contemporaneous record; strong]

A 2026 Vinod Khosla SEC power of attorney supplies the broadest legal-name reconciliation: KV I–IX; IV CF; VIII/IX Strategic; Seed, Seed Side, Seed B/B CF/C/D/E/F/G; Opportunity I–III; Excelsior. It is an authorization/inventory route, not a complete fund ledger and omits the ADV’s transaction-specific SPVs/OAI. [contemporaneous record; strong]

Fees, carry, and alignment

The 2023 New Jersey memo, pp. 1–2 reports these proposed terms: Seed F target $400m, 2.5% management fee and 30% incentive allocation; VIII about $1.5bn, 2% and 30%; Opportunity II about $1bn, 1% and 20%; each has 100% management-fee offset. It reports a minimum $200m GP commitment across the proposal, with allocation and capital source undisclosed. These are incentive terms, not returns, and the memo is not the LPA. (contemporaneous record / strong; KV-supplied through public LP)

Public-LP windows — never pooled

New Jersey, as of 2022-06-30 (KV-supplied, not represented as audited):

Source: NJ diligence memo, pp. 1–4. Young-fund IRR/TVPI is mark-sensitive; “not available” is not zero. (contemporaneous record / strong; KV-supplied through public LP)

New Jersey, as of 2026-04-30:

Source: NJ April 2026 Director’s Report, pp. 18–19. Private values generally use manager latest-quarter marks—mostly December 2025 here—adjusted for later cash flows. Young zero-distribution funds are not realized outcomes. (contemporaneous record / strong)

CalPERS version clocks: the current extract, whose exact valuation quarter is not exposed, reports KV III $200m commitment/cash in, $362,846,119 cash out, $369,852,581 total value, 9.7% net IRR and 1.8x; Seed $60m commitment/cash in, $12,484,329 cash out, $71,183,741 total value, 1.4% net IRR and 1.2x. The 2024 report reports III 1.88x/9.9% and Seed 1.53x/3.9%. The 2017 report reports III $195m contributed, $165,090,556 distributed, $254,007,653 total, $103,963,600 NAV, 1.30x and 5.1%; Seed $60m contributed, $10,504,328 distributed, $50,242,151 total, 0.84x and -3.4%. These clocks and methods must not be averaged. (contemporaneous record / strong)

North Dakota: the 2023-12-31 report lists “Khosla Ventures Fund VII Composite,” vintage 2023, $35m commitment, $2.765m paid in, zero distributed, $2,386,792 value, 0.86x, IRR not meaningful. The vehicle/vintage mismatch remains unresolved; an immature mark is not a realized loss. (contemporaneous record / strong)

A registered-fund N-CSR exposes one Opportunity III LP position—$1.5m cost, $1.474m fair value, $3.5m unfunded at 2026-03-31, no redemption, ten-year term plus possible extensions. It does not establish every LP’s rights or another vehicle’s duration. (contemporaneous record / strong)

People, capital, and influence map

The current team index lists five managing directors: David Weiden, Keith Rabois, Samir Kaul, Sven Strohband, and Vinod Khosla. A current title is not an IC charter.

For the 14 material cases, the public role map is deliberately sparse:

Comprehensive deal ledger

The ledger is a searched-lead register, not a track record. Current-card membership is fixed by the official 2026-08-22 spotlight snapshot; each current row preserves that card's outbound identity route, while outside rows use the strongest mapped evidence. The 116 outbound company routes are identity locators and are deliberately not duplicated as separate research-source rows in sources.md; this route-class exception does not apply to substantive evidence. source-map portfolio reconciliation [contemporaneous record; strong] R means sufficiently reconstructed for a bounded narrative, U unresolved material fields, and X excluded from completed-investment coverage. M/S is research materiality, not outcome quality.

Current official spotlight — 116 unique identities

Consumer & Retail — 13

Enterprise — 22

Fintech — 11

Frontier — 18

Sustainability — 14

Digital Health — 13

MedTech & Diagnostics — 12

Therapeutics — 13

Outside current spotlight — 52 public leads

These rows do not form a complete historical portfolio. KPCB, early KV, personal/trust, Khosla Impact, Khosla Labs, and SPAC cases remain distinct.

Ledger reconciliation

The current spotlight tables contain 12 R + 104 U = 116. The outside table contains 15 R + 30 U + 7 X = 52. Combined: 27 R + 134 U + 7 X = 168. Current-card materiality is 47 M + 69 S = 116. Cross-category assignment duplicates do not create new ledger rows. current ledger; outside ledger [researcher inference; strong]

The 13-company 2007 Senate Appendix B has 12 companies outside today’s spotlight; LanzaTech is the one current-card overlap and is counted once. Khosla Impact’s named leads are discovery coverage, not a complete denominator. Khosla Labs project names remain incubation leads until financing/ownership is evidenced. Senate Appendix B, PDF pp. 29–30; official portfolio snapshot [researcher inference; strong]

Case 1 — Juniper Networks (KPCB, 1995–96)

Decision freeze

Decision date. Andy Bechtolsheim introduced KPCB to Pradeep Sindhu in 1995; after roughly six months of work, KPCB invested in June 1996. The filing records a June 11, 1996 sale of 1,578,418 Series A shares to a group of private investors at $1.00 per share; a separate registration-rights exhibit identifies KPCB VII and affiliated vehicles across common and Series A/B holdings, without allocating the group sale's cash by investor. (KPCB case, “A Vision for the Future”/“Research”/“KPCB’s Decision”/“An Entrepreneur Emerges”; 1999 S-1/A, Business/Related-Party/Principal Stockholders; Peak XV, 00:01:51–00:03:48 and 00:15:30–00:16:45) [contemporaneous record; strong]

Known then. Internet-service providers were growing, routers were a critical bottleneck, Cisco was the dominant incumbent, Sindhu proposed a purpose-built packet-forwarding architecture, and @Home supplied a realistic prospective-customer diligence route. The filing says Juniper was incorporated in February 1996; it later described a purpose-built ISP-router market and Cisco displacement. (KPCB case, “A Vision for the Future”/“Research”/“KPCB’s Decision”/“An Entrepreneur Emerges”; 1999 S-1/A, Business/Related-Party/Principal Stockholders; Peak XV, 00:01:51–00:03:48 and 00:15:30–00:16:45) [contemporaneous record; strong]

Unknown/disputed then. Whether a start-up could produce carrier-grade reliability; whether customer surveys reflected real willingness to switch; time to silicon and operating software; Cisco's likely response; and whether the Internet growth forecast would convert into Juniper revenue. KPCB's later “2,500x” is not decision-date evidence. (KPCB case, “A Vision for the Future”/“Research”/“KPCB’s Decision”/“An Entrepreneur Emerges”; 1999 S-1/A, Business/Related-Party/Principal Stockholders; Peak XV, 00:01:51–00:03:48 and 00:15:30–00:16:45) [researcher inference; strong]

Thesis and alternatives. Khosla says he preferred a technical architecture and founder capable of changing the router cost/performance frontier despite discouraging customer research. Alternatives were pass, fund an adjacent component vendor, or wait for a working system—each reduced technical risk but surrendered the architecture/control position. (Peak XV interview note, Juniper at 00:01:51–00:03:48 and 00:15:30–00:16:45; KPCB case, “KPCB’s Decision”) [investor-stated; moderate]

Risks/kill criteria visible then. Product tape-out/software integration, carrier reliability, inability to recruit a commercial CEO, customer unwillingness to qualify a second vendor, and Cisco price bundling. The six-month diligence and recruitment of Scott Kriens are documented; formal IC kill criteria are not. (KPCB case, “A Vision for the Future”/“Research”/“KPCB’s Decision”/“An Entrepreneur Emerges”; 1999 S-1/A, Business/Related-Party/Principal Stockholders; Peak XV, 00:01:51–00:03:48 and 00:15:30–00:16:45) [retrospective witness; moderate]

Terms, ownership, governance, attribution. The registration-rights exhibit identifies KPCB VII and affiliated vehicles—not Khosla personally—as holders of common and Series A/B securities; it does not allocate the group Series A sale's cash by investor. Khosla and Brook Byers/Hearst-related KPCB governance must not be collapsed into a personal check; Khosla served on Juniper's board from February 1996. Exact partner carry, reserves, later sales, distributions and dilution-adjusted cost are absent. (KPCB case, “A Vision for the Future”/“Research”/“KPCB’s Decision”/“An Entrepreneur Emerges”; 1999 S-1/A, Business/Related-Party/Principal Stockholders; Peak XV, 00:01:51–00:03:48 and 00:15:30–00:16:45) [contemporaneous record; strong]

Outcome and process lesson

Juniper went public in June 1999. KPCB later described the investment as 2,500x, but no audited deal cash-flow series was located. The researcher-derived lesson is not “ignore customers”; it is to test whether customer answers are constrained by the incumbent product definition, then replace survey confidence with technical milestones and a design partner. (KPCB case, outcome/2,500x passage; 1999 S-1/A, prospectus cover) [researcher inference; moderate]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Juniper KPCB case, Research/Decision [researcher inference; moderate]

1. Would you invest then? Decide using only founder/architecture evidence and the @Home diligence—not the IPO.
2. What would kill the deal? Specify the date by which forwarding silicon, operating-system stability, and a carrier-design partner must be demonstrated.
3. What ownership/terms fit the risk? Price a tranche that preserves meaningful KPCB ownership but releases later capital only against system-level tests.
4. What evidence unlocks the next check? Working prototype; independent reliability test; recruited CEO; first paid carrier deployment.

Actual decision/result: KPCB invested in 1996, helped recruit Kriens, and Juniper completed a 1999 IPO. The multiple remains a firm claim, not an audited return. (KPCB case, outcome/2,500x passage; 1999 S-1/A, prospectus cover) [documented behavior; strong]

Case 2 — Cerent (KPCB company formation, 1996–99)

Decision freeze

Decision date. Late 1996, when Khosla conceived a metropolitan optical-networking company and recruited the first engineers. The Wall Street Journal reported that KPCB paid about $8m for a 30% stake; it did not disaggregate financing rounds or prove cash returned. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs) [contemporaneous record; moderate]

Known then. Metro fiber rings had to carry both telephone traffic and fast-growing data traffic; telecom carriers faced a transition from circuit to packet networks; Khosla could recruit technical founders and temporarily perform the CEO role. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs) [contemporaneous record; moderate]

Unknown/disputed then. Product-market timing, carrier qualification cycles, incumbent response, optical-capex durability, and whether an IPO or strategic sale would dominate. The record is retrospective at announcement, not an IC memo. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs) [researcher inference; moderate]

Thesis and alternatives. Build a company around a specific network-architecture gap instead of waiting for an inbound pitch. Alternatives were a passive investment in an existing optical vendor, incubation with fewer operating duties, an IPO, or sale. By August 1999 Cerent's board considered the IPO route and accepted Cisco's 100m-share offer as financially compelling. (Forbes, “High-stakes poker in networking land”) [retrospective witness; moderate]

Risks/kill criteria. Recruiting a complete team, constraining feature creep, shipping on carrier schedules, and avoiding dependence on a single consolidator. A founder later said too many Khosla-proposed features could have slowed launch—a useful counterweight to the “venture assistance” narrative. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs) [retrospective witness; moderate]

Terms, ownership, governance, attribution. Khosla was described as concept originator, recruiter, CEO for 1.5 years and chairman. The security terms, KPCB fund identity, later dilution, partner allocation and proceeds are not public. The 30%/$8m figures are press-reported, not an issuer filing. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs) [contemporaneous record; moderate]

Outcome and process lesson

Cisco announced on August 26, 1999 that it would exchange 100m Cisco shares for Cerent shares it did not already own. Using Cisco's August 25 close yielded about $6.9bn; Cisco's combined press headline included Monterey and $7.4bn, so neither number should be called KPCB proceeds. (Cisco release) [contemporaneous record; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Cerent archived WSJ, lines 9–35 [researcher inference; moderate]

1. Would you invest then? Decide whether company formation compensates for the absence of an established founding team.
2. What would kill the deal? Team assembled, carrier-grade prototype, and first design win by fixed dates; stop if feature expansion repeatedly delays the core path.
3. What ownership/terms fit the risk? Demand governance matching the incubation work while budgeting founder/employee dilution.
4. What evidence unlocks the next check? CEO handoff, optical-system validation, carrier trial, IPO-readiness versus strategic bids.

Actual decision/result: Khosla/KPCB formed and operated Cerent; its board chose Cisco stock over the IPO. The transaction was spectacular, but KPCB's realized return is unavailable. (archived WSJ, lines 9–35; Forbes, board-deliberation paragraphs; Cisco completion release, opening) [documented behavior; strong]

Case 3 — Square / Block (KV III, 2009 onward)

Decision freeze

Decision date. November–December 2009 Series A. Contemporary reporting described a $10m round at a roughly $40m–$45m valuation, mostly/led by Khosla Ventures; these are press figures, not executed financing documents. (TechCrunch, 2009-12-01) [contemporaneous record; moderate]

Known then. Square began in February 2009 to let anyone accept card payments with a mobile device. The product made acquiring simpler for small sellers, but the public launch was not yet mature when the Series A was reported. The addressable wedge was card acceptance; Cash App did not yet exist. (TechCrunch, 2009 round/valuation/investor-attribution; Square S-1, Our Business/Risk Factors/Principal Stockholders) [contemporaneous record; strong]

Unknown/disputed then. Fraud/chargebacks, network and bank relationships, hardware reliability, merchant acquisition cost, regulatory obligations, unit economics, and whether the company could expand beyond a reader. Khosla's later Cash App narrative is outcome knowledge. (TechCrunch, 2009 round/valuation/investor-attribution; Square S-1, Our Business/Risk Factors/Principal Stockholders) [researcher inference; strong]

Thesis and alternatives. The strongest reconstructed thesis is founder quality plus a simple inclusion wedge into a vast, regulated network—not a 2009 thesis that Square would become a consumer bank. Alternatives: pass because payments infrastructure was crowded/regulatory; back merchant software without acquiring risk; or wait for loss-rate and activation cohorts. (TechCrunch, 2009 round/valuation/investor-attribution; Square S-1, Our Business/Risk Factors/Principal Stockholders) [researcher inference; moderate]

Risks/kill criteria. Seller activation and retention; fraud loss per GPV; gross-margin/payback after processing costs; bank/network continuity; product reliability. The 2015 S-1 reported a four-to-five-quarter seller payback and that payments/POS still produced about 95% of revenue; those validate metrics but were unknown in 2009. (TechCrunch, 2009 round/valuation/investor-attribution; Square S-1, Our Business/Risk Factors/Principal Stockholders) [contemporaneous record; strong]

Terms, ownership, governance, attribution. The S-1 records Khosla Ventures III, LP with 50,522,780 shares/17.3% before the IPO, controlled through VK Services. This is fund ownership, not personal capital. TechCrunch called the round Gideon Yu's first investment after joining KV; later KV pages associate Vinod with Block. Sourcing/sponsorship is therefore not cleanly Vinod-only. Later participation and board influence are incompletely public. (TechCrunch, 2009 round/valuation/investor-attribution; Square S-1, Our Business/Risk Factors/Principal Stockholders) [contemporaneous record; strong]

Outcome and process lesson

Square launched Cash in November 2013 and had transferred more than $1bn by its 2015 S-1; it completed its IPO in November 2015 and later became Block. The disciplined lesson is to underwrite the wedge and option value separately: fund card acceptance on its own merits, while using bounded experiments to discover adjacent businesses. Khosla later said the board resisted Square Cash and he supported testing it; that is self-reported governance, not board minutes. (Company Building 102, linked-video Cash discussion; S-1, Business and Square Cash) [investor-stated; moderate]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Square S-1, Our Business/Risk Factors [researcher inference; moderate]

1. Would you invest then? Underwrite mobile acquiring without assuming Cash App or banking adjacencies.
2. What would kill the deal? Fraud-adjusted gross margin, repeat GPV cohorts, network compliance, and seller payback outside an agreed ceiling.
3. What ownership/terms fit the risk? Set reserves for regulatory/product iteration and avoid confusing voting control with personal ownership.
4. What evidence unlocks the next check? Public launch, active-seller retention, fraud loss, bank-partner stability, then a separately budgeted consumer experiment.

Actual decision/result: Contemporary press reported that KV led the 2009 Series A; the 2015 S-1 records KV III's 17.3% pre-offering holding, and Square/Cash/IPO followed. The mapped record does not establish every intervening follow-on or a deal-level realized return. (TechCrunch, round/valuation/investor-attribution; Square S-1, Principal Stockholders/offering cover) (Square S-1, Square Cash/Principal Stockholders/offering cover; Company Building 102, 00:13:57–00:14:23) [documented behavior; strong]

Case 4 — Impossible Foods (KV, 2010–present)

Decision freeze

Decision date. Pat Brown first approached Khosla in 2010; company financing is reported in 2011. Brown says there was no company, deck, product or sample at the first meeting. Brown and Mallaby separately report an initial $3m commitment/seed financing. (KV “Our Debut” transcript, Brown segment; Our Debut transcript, Pat Brown/Mallaby sequence) [retrospective witness; moderate]

Known then. Brown was a respected scientist with a mission to displace animal agriculture; the environmental consequence could be enormous; technical work had not yet produced a commercial food. (Our Debut transcript, Pat Brown and “really improbable” sequences) [retrospective witness; moderate]

Unknown/disputed then. Product taste, texture, nutrition, regulatory treatment, production cost, manufacturing scale, consumer adoption, incumbent response and financing intensity. Exact initial security, valuation, ownership and board terms are unavailable. (Our Debut transcript, Pat Brown and “really improbable” sequences) [researcher inference; strong]

Thesis and alternatives. Khosla describes the decision as buying an option on an improbable but very large consequence and prioritizing founder ambition over a polished plan. Alternatives included a research grant, milestone-based seed, waiting for a prototype, or backing enabling ingredients rather than a branded food company. (Our Debut transcript, Pat Brown and “really improbable” sequences) [investor-stated; moderate]

Risks/kill criteria. A defensible ex-ante gate would require blinded sensory performance, ingredient/regulatory path, reproducible pilot process, cost-down curve and evidence that consumers repurchase without permanent subsidy. No public record proves these were KV's actual gates. (Our Debut transcript, Pat Brown and “really improbable” sequences) [researcher inference; moderate]

Terms, ownership, governance, attribution. This is a KV relationship publicly associated with Vinod. The $3m figure is repeated by Brown/Mallaby, but no closing document establishes security or ownership. A March 2020 company release says a $500m Series F included existing investor KV and brought cumulative funding since 2011 to nearly $1.3bn; it does not disclose KV's share. (company Series F release; 2020 SEC Form D) [contemporaneous record; strong]

Outcome and process lesson

By March 2020, Impossible had commercialized plant-based meat and reported cumulative financing near $1.3bn; the mapped record does not establish a public exit, realized return, or current KV ownership. The lesson is about structuring scientific option value, not declaring a financial win from financing marks. (Series F release, opening/investor list; 2020 Form D, offering/first-sale/related-person fields) [documented behavior; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Our Debut, Pat Brown/no-product sequence [researcher inference; moderate]

1. Would you invest then? Decide whether mission/founder quality justifies funding before a sample exists.
2. What would kill the deal? Blinded taste threshold, safety/regulatory path, pilot yield and cost-down milestones.
3. What ownership/terms fit the risk? Use staged capital with enough ownership for the scientific risk, while reserving heavily for manufacturing rounds.
4. What evidence unlocks the next check? Prototype, repeatable pilot batch, regulator engagement, paid launch and repeat purchase.

Actual decision/result: Retrospective accounts identify early KV capital, and the March 2020 issuer release names KV in the Series F; by that clock the company had commercialized products. The mapped record does not establish a later exit or financial outcome. (Series F release, opening/investor list; 2020 Form D, offering/first-sale/related-person fields) [documented behavior; strong]

Case 5 — OpenAI (KV, late 2018 / early 2019)

Decision freeze

Decision date. Khosla consistently dates the commitment decision to late 2018; TIME and later interviews report a $50m investment in early 2019. No executed closing record, security, valuation, ownership or side letter is public. The decision clock and closing clock must remain separate. (TIME interview, publisher intro and OpenAI question; SparX interview, 00:04:06–00:07:09) [investor-stated; moderate]

Known then. OpenAI was organized as a nonprofit research lab, had concentrated technical talent, required increasing compute, and was preparing a new legal/economic structure. On March 11, 2019 OpenAI announced a capped-profit LP: returns for first-round investors were capped at 100x, while governance remained tied to the nonprofit mission. Khosla Ventures was named as an investor. (OpenAI LP announcement, Structure/Returns/Why/named investor; SparX, 00:04:06–00:10:36; Peak XV, 00:02:19–00:03:01) [contemporaneous record; strong]

Unknown/disputed then. Whether scaling laws would continue; the route to revenue; compute and talent costs; governance stability; the exact security and cap applicable to KV; dilution; safety/regulatory shocks; and whether an investor could ever realize liquidity. Khosla describes the organization as nearly impossible to diligence and says he apologized to LPs for making a check about twice his normal maximum. The apology letter itself was not publicly recovered. (OpenAI LP announcement, Structure/Returns/Why/named investor; SparX, 00:04:06–00:10:36; Peak XV, 00:02:19–00:03:01) [investor-stated; moderate]

Thesis and alternatives. The reconstructed thesis was not a conventional SaaS forecast: concentrate exceptional research talent, treat compute as a scale input, and buy exposure to a radically nonlinear technical trajectory. Alternatives were pass because of nonprofit control, invest a smaller option-sized check, finance a commercial application company, or wait for product/revenue evidence. (OpenAI LP announcement, Structure/Returns/Why/named investor; SparX, 00:04:06–00:10:36; Peak XV, 00:02:19–00:03:01) [researcher inference; moderate]

Risks/kill criteria. Technical progress per compute dollar, talent retention, governance enforceability, commercialization rights, future capital intensity and safety/regulatory constraints. Traditional revenue milestones were a poor early fit, but legal rights and mission/control provisions were unusually important. No public record establishes KV's actual kill criteria. (OpenAI LP announcement, Structure/Returns/Why/named investor; SparX, 00:04:06–00:10:36; Peak XV, 00:02:19–00:03:01) [researcher inference; strong]

Terms, ownership, governance, attribution. OpenAI's announcement names Khosla Ventures and states the general first-round framework; separately, Khosla retrospectively describes making the decision and a reported $50m size. That does not establish personal capital, board rights or a standard equity security. The announced 100x cap applies to the first round generally, but without KV's contract it cannot be used to calculate KV economics. (OpenAI LP announcement, Structure/Returns/Why/named investor; SparX, 00:04:06–00:10:36; Peak XV, 00:02:19–00:03:01) [contemporaneous record; strong]

Outcome and process lesson

Khosla later called it a >100x outcome and projected much more value; those are retrospective, unaudited claims and may refer to marks rather than distributions. Current governance or valuation cannot be projected backward into the 2019 contract. The process lesson is that unconventional governance is itself an underwriting dimension: conviction in technology does not substitute for mapping control, caps and conversion rights. (SparX retrospective, 00:38:55–00:39:34) [investor-stated; provisional]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. OpenAI LP announcement, Structure/Returns [researcher inference; moderate]

1. Would you invest then? Decide whether talent/technical trajectory offsets the nonprofit-controlled and pre-product structure.
2. What would kill the deal? Define technical, governance and key-person failure states; do not force fake near-term revenue targets.
3. What ownership/terms fit the risk? Require an explicit cap, information rights, dilution/conversion treatment and mission-control map.
4. What evidence unlocks the next check? Measured capability progress, compute efficiency, talent retention, commercial structure and credible funding runway.

Actual decision/result: Khosla retrospectively dates his decision to late 2018 and a reported $50m investment to early 2019; OpenAI separately named KV on March 11, 2019. Later multiples are self-reported claims, not audited cash returns. (OpenAI LP announcement, named-investor paragraph; SparX, 00:04:06–00:07:09 and 00:38:55–00:39:34) [documented behavior; moderate]

Case 6 — KiOR (KV II/III, KFT Trust, VNK; 2007–14)

Decision freeze A: formation and early equity

Decision date. KiOR was incorporated in July 2007 as a joint venture between Khosla Ventures and BIOeCON. At formation/2007–08, filings show $4.4m Series A at $0.3625, then about $10m Series A-1 at $0.9725; a $15m 2009 KV note converted in 2010, and KV participated in the 2010 Series B. (2011 S-1, Organization and Related-Party Transactions) [contemporaneous record; strong]

Known then. The concept combined catalytic-cracking methods from petroleum refining with proprietary catalysts to convert cellulosic biomass to a renewable crude compatible with existing refining/distribution. Federal RFS2 created demand and credit value; Mississippi was willing to extend a $75m state loan for a first commercial plant. KiOR remained a development-stage, first-of-kind scale-up. (KiOR 2011 S-1, Business/technology/regulation/Mississippi loan/Risk Factors/Directors/Principal Stockholders; KV Q&A, Range/planned-failure passage) [contemporaneous record; strong]

Unknown/disputed then. Whether lab/demo yields would persist at a 500 bone-dry-ton/day plant; catalyst durability; uptime, throughput, product mix and cash cost; true capital needed; and whether RIN/offtake economics could compensate for process unreliability. These were not generic “market” risks but integrated-plant risks. (KiOR 2011 S-1, Business/technology/regulation/Mississippi loan/Risk Factors/Directors/Principal Stockholders; KV Q&A, Range/planned-failure passage) [contemporaneous record; strong]

Thesis and alternatives. Khosla's public strategy was to fund parallel biofuel approaches and select the winner after technical evidence. Alternatives were laboratory-only option capital, a smaller demonstration plant, project finance after independent engineering validation, licensing to an incumbent, or continued support for Range/Coskata rather than concentration in KiOR. (KiOR 2011 S-1, Business/technology/regulation/Mississippi loan/Risk Factors/Directors/Principal Stockholders; KV Q&A, Range/planned-failure passage) [investor-stated; moderate]

Risks/kill criteria. The decisive gates should have been continuous throughput, yield, on-stream percentage, specification mix, feedstock logistics, working-capital burn and financing without circular sponsor support. Public filings do not expose the original IC gates. (KiOR 2011 S-1, Business/technology/regulation/Mississippi loan/Risk Factors/Directors/Principal Stockholders; KV Q&A, Range/planned-failure passage) [researcher inference; strong]

Terms, ownership, governance, attribution. Samir Kaul joined the board in November 2007 and represented KV through multiple financings. Before/after the 2011 IPO, KV entities controlled the company through dual-class voting; this is both a Vinod/VK Services control route and a Samir governance route, not a lone-person decision. The 2011 filing records about 54% pre-offering beneficial ownership for KV II and related holdings. (KiOR 2011 S-1, Business/technology/regulation/Mississippi loan/Risk Factors/Directors/Principal Stockholders; KV Q&A, Range/planned-failure passage) [contemporaneous record; strong]

Decision freeze B: rescue capital after commercialization risk was visible

January 2012 KFT loan. KFT Trust, with Vinod Khosla as trustee, lent $25m alongside $50m from Alberta lenders. The secured loan bore 16% interest, amortized from August 2013 to February 2016, had prepayment premiums/end charge, warrants and a security interest in substantially all assets. It also conditioned Newton capex on new equity. This is trust capital, not KV-fund equity. (2012 8-K, Item 1.01, lines 75–102) [contemporaneous record; strong]

October 2013 VNK note. VNK Management directly held a $21.25m senior secured convertible note at $2.897/share; the filing states that Vinod sometimes recommended specific VNK investments but VNK disclaimed group status. This is a separate vehicle and a different risk/priority layer. (Form 3/A, derivative table and footnotes) [contemporaneous record; strong]

What was known by March 2014. Columbus had not reached steady state; structural bottlenecks limited feed; yields were low; hydrotreater reliability and product mix were poor; the plant had been idled; KiOR recorded a $185m plant impairment; 2013 product revenue was only $1.5m; and the company had no committed near-term finance beyond a conditional Khosla commitment up to $25m in monthly tranches tied to future milestones. (2013 10-K, Business lines 321–36; Risk Factors lines 721–97; liquidity lines 1696–1706) [contemporaneous record; strong]

Outcome, failure postmortem and process change

KiOR filed Chapter 11 on November 9, 2014; Khosla affiliates bid for assets and offered up to $15m DIP financing. The company expected no recovery for common shareholders. (bankruptcy exhibit, opening/no-common-recovery; Mississippi auditor, Failed Projects chart/KiOR narrative; SEC release, defendants/allegations/settlement; Mississippi opinion, facts/procedure/holding) Mississippi's later auditor report recorded a $76.25m public loss. The SEC's 2016 case named KiOR and former CEO Fred Cannon—not Khosla/KV—and the company settled without admission or denial; Mississippi litigation against Khosla had a procedural appellate history, not a merits liability holding. [contemporaneous record; strong]

Warning signs and capital response. Warning signs were not subtle by 2013–14: missed steady state, first-of-kind bottlenecks, adverse mix, limited revenues, repeated need for sponsor-linked secured finance, and going-concern doubt. The response moved from equity/control to high-cost secured trust/VNK/DIP instruments. That protected priority differently from common equity but also extended the experiment. (KiOR 2013 10-K, Business lines 321–336/Risk 721–797/liquidity 1696–1706; KFT loan, Item 1.01; VNK note, derivative table/footnotes; bankruptcy exhibit, opening/no-common-recovery) [documented behavior; strong]

Claimed lesson versus later fact. In a 2013 interview Khosla called KiOR a huge success and contrasted it with Range; KiOR failed the next year. That quote is a vivid warning against treating interim public marks/technical selection as realized success. (KV Q&A, KiOR/Range passage) [investor-stated; strong]

Process-change evidence. Later rhetoric stresses milestone-conditioned tranches and “intelligent failure,” but no public KV policy links a specific formal process change to KiOR. The March 2014 commitment itself contemplated monthly amounts conditioned on mutually agreed performance milestones, which is documented behavior but arrived after severe impairment. (KiOR 2013 10-K, Business lines 321–336/Risk 721–797/liquidity 1696–1706; KFT loan, Item 1.01; VNK note, derivative table/footnotes; bankruptcy exhibit, opening/no-common-recovery) [contemporaneous record; moderate]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. KiOR 2011 S-1, Business/Risk Factors [researcher inference; moderate]

1. Would you invest then? At formation, decide if the catalyst/FCC approach merits an option; re-decide separately before the 2012 secured loan.
2. What would kill the deal? Continuous nameplate throughput, minimum yield/product mix, cash cost before scale-up, and a stop if sponsor capital is the only available financing.
3. What ownership/terms fit the risk? Map KV II/III equity, KFT debt/warrants, VNK convertible debt and DIP priority separately; model no common recovery.
4. What evidence unlocks the next check? Independent mass-balance test, 30/90-day on-stream performance, customer-qualified fuel, working-capital need and third-party project finance.

Actual decision/result: KV formed and controlled KiOR, multiple affiliated vehicles followed on, the plant failed to reach steady state, and Chapter 11 left no expected common recovery. Exact vehicle-level recoveries remain unavailable. (KiOR 2013 10-K, Business lines 321–336/Risk 721–797/liquidity 1696–1706; KFT loan, Item 1.01; VNK note, derivative table/footnotes; bankruptcy exhibit, opening/no-common-recovery) [documented behavior; strong]

Case 7 — Range Fuels (KV-created/financed, 2006–11)

Decision freeze

Decision date. The 2007 Senate Appendix B lists Range/Kergy among Khosla's holdings, and the contemporaneous DOE review describes the woody-biomass project. The mapped record does not recover an exact investment date, signed Series A, KV amount, ownership, or decision authority. (Senate testimony, Appendix B, PDF pp. 29–30; DOE review, printed pp. iii–iv and 20–22) [contemporaneous record; moderate]

Known then. Woody biomass was abundant, federal policy targeted advanced-biofuel adoption, and the technology proposed gasification followed by catalytic conversion into alcohols. DOE selected Range for a grant up to $76m, and the project planned an initial commercial-scale Georgia plant. The 2008 DOE industry review simultaneously warned that advanced biofuels were unproven at commercial scale and hard to finance. (DOE 2007 Year in Review, PDF pp. iii, 20–22, Range profile) [contemporaneous record; strong]

Unknown/disputed then. Integrated gasifier/catalyst yield, alcohol product mix, continuous uptime, delivered feedstock cost, construction cost, qualifying-fuel status, and whether forecast production dates were credible. (Senate testimony, Appendix B, PDF pp. 29–30; DOE review, printed pp. iii–iv and 20–22; KV Q&A, Range/planned-failure passage) [researcher inference; strong]

Market/financing/regulatory context. The contemporaneous DOE review records a federal grant and a planned commercial plant; the later official project history records public guaranteed financing and default. Those public instruments did not remove technology or completion risk. Exact private-round sizes and KV allocation are not established by the mapped sources. (DOE review, printed pp. iii–iv and 20–22; DOE/LanzaTech record, project history) [contemporaneous record; strong]

Thesis and alternatives. Khosla later described Range as one of four technologies run in parallel, with an intent to stop the non-winners. Alternatives were continued pilot validation, a smaller modular plant, a strategic partner with operating expertise, or allocating more capital only after continuous fuel-grade output. (Senate testimony, Appendix B, PDF pp. 29–30; DOE review, printed pp. iii–iv and 20–22; KV Q&A, Range/planned-failure passage) [investor-stated; moderate]

Risks/kill criteria. DOE grant milestones, continuous syngas quality, fuel-grade alcohol yield, qualifying RIN generation, total installed cost, feedstock contracts and non-sponsor project finance. No actual KV IC kill sheet is public. (Senate testimony, Appendix B, PDF pp. 29–30; DOE review, printed pp. iii–iv and 20–22; KV Q&A, Range/planned-failure passage) [researcher inference; strong]

Terms, ownership, governance, attribution. The Senate appendix identifies Range as a Khosla holding, and Khosla later described it as one of several parallel bets. The mapped record does not establish source, signed security, KV ownership, board votes, reserve releases, or recovery; public grants/guarantees are not KV capital. (Senate testimony, Appendix B, PDF pp. 29–30; DOE review, printed pp. iii–iv and 20–22; KV Q&A, Range/planned-failure passage) [contemporaneous record; moderate]

Outcome, failure postmortem and process change

The official later project history says Range could not meet its obligations, defaulted, never reached full scale, and that LanzaTech acquired the facility in January 2012. It does not establish KV's loss or recovery. (DOE/LanzaTech record, project history/existing-facility sections) [contemporaneous record; strong]

Claimed lesson/counterevidence. Khosla later called Range an “almost planned failure” in a parallel race and said funding stopped when KiOR looked best. That explains portfolio logic, but it does not make the capital loss ex ante harmless—especially because KiOR subsequently failed. The stronger lesson is to precommit comparative milestones and a portfolio-wide stop rule before any contender becomes the favored narrative. (KV canonical Q&A, Range/planned-failure passage) [investor-stated; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. DOE review, pp. iii–iv and 20–22 [researcher inference; moderate]

1. Would you invest then? Decide whether parallel option value justifies financing Range independently of KiOR.
2. What would kill the deal? Fuel-grade continuous output, yield/cost, qualifying credits and a hard ceiling on public-plus-private capital before proof.
3. What ownership/terms fit the risk? Separate KV equity from grants and guaranteed debt; do not count taxpayer support as de-risked technology.
4. What evidence unlocks the next check? Integrated pilot, 30-day run, product certification, feedstock delivery, EPC budget and third-party project finance.

Actual decision/result: KV and others financed Range; public agencies added grants/guarantees; commercial claims were not achieved and the plant closed/sold. Exact KV loss is not publicly audited. (DOE/LanzaTech record, project history/existing facility; KV Q&A, Range/planned-failure passage) [documented behavior; strong]

Case 8 — Coskata (KV funds; Samir Kaul board, 2006–13)

Decision freeze

Decision date. Samir Kaul served as director from June 2006. In January/February 2008 Coskata sold about $20m of Series B at $5.26/share to investors including KV and ATV funds. This supports a KV-team/Kaul case; it does not establish Vinod as sole sponsor. (Coskata S-1/A, Business—Lighthouse/Flagship, Risk Factors, USDA Loan Guarantee, Series B, Principal Stockholders, Management) [contemporaneous record; strong]

Known then. Coskata combined gasification/cleaning with proprietary anaerobic fermentation. By 2012 its Lighthouse demonstration facility had more than 15,000 operating hours and produced ASTM-conforming ethanol; it had feedstock and offtake agreements and proposed a phased Alabama “Flagship” plant. (Coskata S-1/A, Business—Lighthouse/Flagship, Risk Factors, USDA Loan Guarantee, Series B, Principal Stockholders, Management) [contemporaneous record; strong]

Unknown/disputed then. The planned commercial plant substituted an indirect gasifier that had never been tested together with Coskata's cleaning system for fuels production. Coskata had no commercial-scale operating experience, limited revenue, continuing losses, feedstock/offtake and construction risks, and required project finance. The 15,000-hour figure therefore validated a different configuration, not the full Flagship system. (S-1/A, Risk Factors lines 563–610) [contemporaneous record; strong]

Market/financing/regulatory context. A June 2011 USDA conditional commitment proposed a 90% guarantee for $87.9m of debt. It required at least 50% project equity, feedstock/offtake, collateral coverage and other conditions; the filing said only feedstock/offtake were satisfied and the guarantee was necessary. (S-1/A, USDA Loan Guarantee lines 2088–92) [contemporaneous record; strong]

Thesis and alternatives. The thesis was feedstock flexibility plus biological conversion at potentially low cost, supported by modular pilot evidence. Alternatives were a natural-gas-first plant, licensing, a smaller integrated indirect-gasifier demo, or waiting until all USDA conditions and project-equity partners were signed. (Coskata S-1/A, Business—Lighthouse/Flagship, Risk Factors, USDA Loan Guarantee, Series B, Principal Stockholders, Management) [researcher inference; moderate]

Risks/kill criteria. Integrated commercial-configuration test; full mass/energy balance; $/gallon including co-products; 50% equity condition; firm EPC and feedstock/offtake credit quality. The company advertised a firm-price/date-certain construction strategy, but schedule certainty does not resolve untested-process risk. (Coskata S-1/A, Business—Lighthouse/Flagship, Risk Factors, USDA Loan Guarantee, Series B, Principal Stockholders, Management) [contemporaneous record; strong]

Terms, ownership, governance, attribution. The S-1 identifies KV funds as >5% holders, Series B shares/pricing, and Kaul's board role. Exact KV allocation within the $20m round, later purchases, ownership after offering, Vinod's role, and any exit proceeds are incomplete. (Coskata S-1/A, Business—Lighthouse/Flagship, Risk Factors, USDA Loan Guarantee, Series B, Principal Stockholders, Management) [contemporaneous record; strong]

Outcome, failure postmortem and process change

Coskata did not complete its proposed 2012 IPO; it withdrew the offering in July 2013 citing market conditions, and shifted emphasis toward natural gas. The Washington Post reported that it never made commercial biomass volumes. (Washington Post) [contemporaneous record; moderate]

The process lesson is unusually specific: diligence must validate the production configuration that consumes the growth capital, not a nearby pilot configuration with an impressive hour count. No public source shows a formal KV process change after Coskata. (Coskata S-1/A, Business/Risk/Series B/Directors; Washington Post, accessible Coskata outcome section) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Coskata S-1/A, Business/Risk Factors/USDA [researcher inference; moderate]

1. Would you invest then? Decide whether the demonstration record justifies funding before the new gasifier is integrated.
2. What would kill the deal? No Flagship construction until the actual gasifier/cleaning/fermentation stack completes a continuous integrated run.
3. What ownership/terms fit the risk? Tranche equity around technical validation and require financing commitments that do not depend circularly on an unfinished guarantee.
4. What evidence unlocks the next check? Integrated demo, independent mass balance, equity partner, USDA conditions, EPC certainty and binding customer credit.

Actual decision/result: KV participated and Kaul governed; the commercial biomass configuration did not scale, the IPO was withdrawn, and deal-level recovery is unavailable. (Coskata S-1/A, Business/Risk/Series B/Directors; Washington Post, accessible Coskata outcome section) [documented behavior; strong]

Case 9 — View / Soladigm (Sigma + KV, 2007 onward)

Decision freeze

Decision date. In 2007 Sigma Partners and KV jointly financed the Series A at a $5m pre-money valuation. Together—not KV alone—they initially held 50% fully diluted/62% outstanding and, after another purchase nine months later, 56% fully diluted/68% outstanding. (Nguyen v. View, lines 60–68) [contemporaneous record; strong]

Known then. The company developed electrochromic glass intended to reduce building energy use while improving comfort. A capital-intensive materials/manufacturing path and long building qualification cycles were inherent. Mike Scobey was to replace founder Paul Nguyen as CEO. (Delaware opinion, factual background—product/CEO/ownership/voting/authorization) [contemporaneous record; strong]

Unknown/disputed then. Manufacturing yield, field reliability, installation/channel adoption, full capital required, customer-payback evidence, and whether founder/VC governance could function after the CEO change. (Delaware opinion, factual background—product/CEO/ownership/voting/authorization) [researcher inference; strong]

Thesis and alternatives. The likely thesis was that a materials breakthrough could turn a passive building component into an energy-control system in a large market. Alternatives included licensing to glass incumbents, a pilot/JV rather than owned manufacturing, smaller tranches, or founder-led continuation. No public IC memo proves the chosen framing. (Delaware opinion, factual background—product/CEO/ownership/voting/authorization) [researcher inference; moderate]

Risks/kill criteria. Accelerated-life testing, warranty reserves, factory yield, installed economics, financing need, founder/IP continuity and governance authorization. These span technical and corporate-process risk; the latter became material before the later public-company accounting issue. (Delaware opinion, factual background—product/CEO/ownership/voting/authorization) [researcher inference; strong]

Terms, ownership, governance, attribution. The Sigma/KV voting agreement controlled corporate structure, board composition and CEO selection; the two investors controlled four of five board seats and had vetoes on major acts. The court opinion later described founder-consent and Series B authorization disputes, including an arbitrator's ruling that the Series B was void/invalid. These are findings/procedural history in the opinion, not proof that Vinod personally made each act. (Delaware opinion, factual background—product/CEO/ownership/voting/authorization) [contemporaneous record; strong]

Outcome, failure postmortem and process change

The Delaware opinion records later private financing while the governance dispute continued. A separate 2023 SEC order found that View materially understated 2019–21 warranty liabilities: disclosed $22m–$25m versus approximately $48m–$53m including installation. The company settled without admitting or denying the findings, self-reported, cooperated and improved controls. The SEC explicitly imposed no civil penalty. The commonly repeated “$28m penalty” is wrong; roughly $28m is the increase from the originally reported 2019 liability to the $53m restated amount. (SEC order, PDF pp. 1–6, especially ¶¶2–5, 24, 33–35) [contemporaneous record; strong]

The SEC's case against the former CFO was dismissed with prejudice in February 2026 after evidence review; that does not erase the separate company order and does not name Khosla as a charged party. (SEC dismissal release) [contemporaneous record; strong]

Process lesson. Technical ambition needs founder-rights hygiene, valid corporate authorizations and product-liability controls. No public evidence shows that KV adopted a specific post-View process change. (Delaware opinion, ownership/authorization history; SEC company order, pp. 1–6 ¶¶2–5/24/33–35; SEC former-CFO dismissal, disposition) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. View Delaware opinion, factual background [researcher inference; moderate]

1. Would you invest then? Decide if electrochromic performance justifies owning scale-up risk rather than licensing.
2. What would kill the deal? Accelerated failure rate, installed payback, factory yield, warranty accrual adequacy and valid governance approvals.
3. What ownership/terms fit the risk? Model Sigma and KV jointly; preserve founder/IP rights and require independent audit/warranty controls before later financing.
4. What evidence unlocks the next check? Pilot installations, reliability cohort, cost/yield, customer renewals, board authorizations and financing runway.

Actual decision/result: Sigma/KV took joint control; later financing, governance litigation, and a settled SEC company accounting order are documented. Deal-level KV recovery is unavailable. (Delaware opinion, ownership/authorization history; SEC company order, pp. 1–6 ¶¶2–5/24/33–35; SEC former-CFO dismissal, disposition) [documented behavior; strong]

Case 10 — Nextdoor via KVSB (SPAC sponsor/PIPE, 2021)

Decision freeze

Decision date. July 6, 2021 merger announcement; October 21 proxy; November 5 close. This is a Khosla Ventures Acquisition Co. II sponsor decision, not the original Nextdoor venture investment and not a generic KV fund deal. (Nextdoor/KVSB proxy, Summary—Business Combination; closing filing, opening/conversion) [contemporaneous record; strong]

Known then. Nextdoor's 2020 weekly active users grew about 37% to nearly 27m and ARPU rose from $4.23 to $4.62 during pandemic shelter-in-place behavior. Revenue was predominantly advertising; the investor case asserted a large monetization gap versus Twitter/Snap and projected over 40% annual revenue growth. The proxy defined active users to include people opening monetizable emails. (Nextdoor/KVSB proxy, Summary/metrics/WAU/Risk Factors/projections/redemptions/PIPE) [contemporaneous record; strong]

Unknown/disputed then. How much engagement was pandemic-pulled forward, whether email-open users generated comparable inventory, local-network retention after reopening, ad-market cyclicality, and the durability of >40% growth. The later complaint alleged misstatements; allegations are not facts. (Nextdoor proxy, Risk Factors/metrics; dismissal order, allegations at pp. 1–2) [contemporaneous record; strong]

Market/financing context. Public holders could redeem near $10. The transaction had a $400m minimum-cash condition, a $270m PIPE at $10/share, and a sponsor forward-purchase backstop in the maximum-redemption scenario. The proxy modeled up to 24,832,599 redemptions while preserving the condition; actual redemptions were much lower. (Nextdoor/KVSB proxy, Summary/metrics/WAU/Risk Factors/projections/redemptions/PIPE) [contemporaneous record; strong]

Thesis and alternatives. The published thesis was that a verified-neighbor network had durable engagement and material ARPU expansion. Alternatives were pass, delay until post-pandemic cohorts matured, reduce valuation, or structure more sponsor economics against realized user/ARPU targets. (Nextdoor/KVSB proxy, Summary/metrics/WAU/Risk Factors/projections/redemptions/PIPE) [contemporaneous record; strong]

Risks/kill criteria. Cohort-level app engagement excluding email opens, post-reopening WAU/DAU, advertiser retention, ARPU quality, valuation under no pandemic lift, redemption level and dilution. (Nextdoor/KVSB proxy, Summary/metrics/WAU/Risk Factors/projections/redemptions/PIPE) [researcher inference; strong]

Terms, ownership, governance, attribution. The sponsor bought 10m founder shares for $25,000: 5m Class B and 5m Class K. At closing they converted into 7,347,249 and 3,061,354 Class A shares, respectively. Sponsor-related investors also bought 750,000 PIPE shares for $7.5m; a conditional forward purchase could add 1m shares. Sponsor/PIPE economics must be evaluated separately from Nextdoor legacy holders and KV venture funds. (2022 resale prospectus, “KVSB Related Party Transactions—Founder Shares”) [contemporaneous record; strong]

Outcome, adverse record and process lesson

The combination closed on November 5, 2021. A plaintiff later alleged a 90% class-period value loss. On May 19, 2025 the federal court dismissed the claims with leave to amend: most statements for lack of plaintiff standing and the surviving May 2022 statement for failure to plead a material misstatement. It was not a merits exoneration of every business forecast and not a finding that the allegations were true. (closing filing, opening/conversion; dismissal order, PDF pp. 1–2 and 12–15) [contemporaneous record; strong]

Process lesson. A SPAC sponsor's near-zero-cost founder shares create a payoff unlike a $10 public share. Decision review should separately score target quality, public-share value, sponsor dilution and disclosure risk; pandemic cohorts need a reopening stress test. (2022 resale prospectus, Founder Shares; closing filing, closing/conversion; 2025 dismissal, pp. 12–15) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Nextdoor proxy, metrics/risk/redemptions [researcher inference; moderate]

1. Would you invest then? Decide at $10 using only July/October 2021 cohorts and explicit user definitions.
2. What would kill the deal? Post-reopening app engagement, ARPU quality, forecast miss bands, redemption ceiling and independent disclosure review.
3. What ownership/terms fit the risk? Put founder shares, Class K conversion, PIPE and forward purchase in a fully diluted waterfall.
4. What evidence unlocks the next check? Quarterly non-email WAU, advertiser retention, ARPU, cash burn and sponsor-share lockup/conversion.

Actual decision/result: KVSB sponsored and closed the merger; Nextdoor obtained public capital, and later securities claims were dismissed with leave to amend. Sponsor and fund economics remain distinct. (2022 resale prospectus, Founder Shares; closing filing, closing/conversion; 2025 dismissal, pp. 12–15) [documented behavior; strong]

Case 11 — Excite/Google acquisition miss (KPCB/Excite, 1999)

Decision freeze

Decision date. 1999 discussions among Google founders, Excite management and Khosla. No signed acquisition agreement existed. Price recollections conflict: Khosla later said $750,000; a 2014 Page/Brin/Khosla exchange discussed Khosla's $350,000 memory and a fake $1.6m “acceptance”; Excite CEO George Bell recalled $750,000 cash plus roughly 1% of Excite. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [retrospective witness; strong]

Known then. Excite was a major portal/search business with hundreds of engineers and its own technology. Google's founders had a demonstrably different ranking system but the perceived result difference was not yet dramatic to Excite. Khosla was an Excite backer/adviser, not the CEO with unilateral acquisition authority. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [retrospective witness; moderate]

Unknown/disputed then. Search's future profit pool, advertising auction economics, Google's eventual distribution/brand, and the cost of integrating or replacing Excite's stack. The exact cash/equity ask and meeting sequence remain disputed. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [retrospective witness; strong]

Thesis and alternatives. The simplified hindsight story is “buy Google cheaply.” Bell's named-participant account adds the actual organizational trade: Page required Excite to remove its search technology and install Google's, creating engineering/cultural switching cost. Alternatives included technology license, A/B traffic test, minority investment, staged acquisition, independent technical benchmark, or rejection. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [retrospective witness; strong]

Risks/kill criteria. Search-quality lift, latency/cost, revenue impact, talent retention, integration authority and cultural disruption. A staged live-traffic experiment could have converted disputed quality into data without accepting a full immediate replacement. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [researcher inference; strong]

Terms, ownership, governance, attribution. This was a missed corporate acquisition by Excite, not a missed personal/KPCB venture check. No security, ownership or cash flow was created. Khosla appears as advocate/negotiator; Bell retained management authority and says he rejected the replacement condition. (KV founders chat, 01:22–03:44; George Bell interview, search-quality/replacement/price passages) [retrospective witness; strong]

Outcome and process lesson

No deal occurred. Google became enormously valuable; Excite@Home later entered bankruptcy, but those outcomes were unknowable in 1999. The lesson is to capture asymmetric technical option value with a reversible experiment and to document non-price conditions. Calling it a “$750k miss” erases the integration decision and price disagreement. WIRED's 2002 Excite@Home postmortem supplies near-period corporate context but not a settled Google price. (WIRED postmortem) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. George Bell interview, replacement/price passages [researcher inference; moderate]

1. Would you invest then? Decide whether the quality evidence warrants a staged deal without knowing Google's future.
2. What would kill the deal? No degradation in retention/revenue/latency; key Google engineers remain; integration is reversible.
3. What ownership/terms fit the risk? Compare license, minority stake, earn-out and acquisition with explicit technology-replacement conditions.
4. What evidence unlocks the next check? Blind quality benchmark, 1% traffic test, commercial impact, integration plan and founder retention.

Actual decision/result: Khosla advocated; Excite management did not accept the deal/technology condition; no transaction closed. Exact price remains disputed. (KV founders chat, 01:22–03:44; George Bell interview, condition/price passages) [retrospective witness; strong]

Case 12 — Rocket Lab (Khosla Ventures V; partner-led control case, 2013–2025)

Decision freeze

Decision date. The surviving ownership filing says Khosla Ventures V began buying Rocket Lab preferred stock in August 2013. It aggregates purchases through May 2020 rather than isolating the first-check amount or valuation. (2025 Schedule 13D, Item 3) [contemporaneous record; strong]

Known then. Rocket Lab was pursuing a dedicated small-launch vehicle in a business where reliability, launch cadence, regulatory approvals and manufacturing scale all had to work. Sven Strohband—not Vinod Khosla—joined the board in August 2013 and remained the clearest documented KV governance actor. (2022 S-1/A, “Management—Board of Directors”) [contemporaneous record; strong]

Unknown/disputed then. The public record does not recover the 2013 investment memo, competing term sheets, initial valuation, ownership target, launch-failure tolerance, or whether Vinod personally sourced or approved the deal. Future launch cadence, government demand, unit economics and later SPAC availability were unknown. (Rocket Lab 13D, Items 3/5/6; 2022 S-1/A, Business/Risk Factors/Directors/Principal Stockholders) [researcher inference; insufficient public record]

Thesis and alternatives. A plausible thesis was that smaller satellites would value a dedicated, frequent launch service rather than waiting for rideshare capacity. That is a researcher reconstruction from the business model, not a recovered IC thesis. Alternatives were pass, milestone-based tranches, invest in satellite/software infrastructure, or wait for a successful orbital launch. (Rocket Lab 13D, Items 3/5/6; 2022 S-1/A, Business/Risk Factors/Directors/Principal Stockholders) [researcher inference; provisional]

Risks/kill criteria. Repeated launch failure, inability to reach orbital cadence, manufacturing bottlenecks, regulatory delay, insufficient contracted backlog, or capital needs that made ownership maintenance uneconomic. Exact KV kill criteria were not published. (Rocket Lab 13D, Items 3/5/6; 2022 S-1/A, Business/Risk Factors/Directors/Principal Stockholders) [researcher inference; moderate]

Terms, ownership, follow-on and attribution. KV V acquired Series A, B, C, D, E and E-1 securities from August 2013 through May 2020 for approximately $28.2m in aggregate; this is not an initial-check figure. At the 2021 SPAC closing, those securities converted into 115,004,795 common shares. A 2025 filing reports later distributions and 36,593,470 shares then deemed beneficially owned across KV V and VK Services; it also says Khosla held no shares directly and disclaimed beneficial ownership except any pecuniary interest. The filings support firm/vehicle ownership and Strohband governance, not a Vinod-solo deal narrative. (2025 Schedule 13D, Items 3, 5 and 6) [contemporaneous record; strong]

Outcome and process lesson

Rocket Lab completed the Vector Acquisition SPAC combination in 2021. The conversion and later distributions establish liquidity pathways, but they do not establish realized fund return: the public record lacks the complete distribution dates, prices, remaining cost basis, dilution and cash-flow ledger. (Rocket Lab 13D, Items 3/5/6; S-1/A, Directors) [contemporaneous record; strong]

Process lesson. This is a useful attribution control. A publicly prominent KV outcome can show a liquidity path while its economic contribution remains uncomputable; the public evidence points to a partner board role and a fund vehicle, not personal ownership or sole sponsorship. The investment also illustrates follow-on endurance, but whether that was planned reserve discipline or path dependence is not publicly resolved. (Rocket Lab 13D, Items 3/5/6; S-1/A, Directors) [researcher inference; moderate]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. Rocket Lab S-1/A, Business/Risk Factors [researcher inference; moderate]

1. Would you invest then? Decide in August 2013 before orbital proof, using only addressable launch demand, team capability and a capital plan.
2. What would kill the deal? Define failure-count, schedule, backlog, burn and regulatory milestones before funding.
3. What ownership/terms fit the risk? Specify initial ownership, milestone tranches, pro-rata reserves and dilution tolerance; do not substitute the later aggregate $28.2m for first-check economics.
4. What evidence unlocks the next check? Engine/vehicle tests, first orbit, cadence, contracted backlog, gross margin path and capital required per launch.

Actual decision/result: KV V invested across six preferred series, Strohband served on the board, and the stake converted in the 2021 SPAC. Later share distributions occurred; an audited realized multiple is not public. (Rocket Lab 13D, Items 3/5/6; S-1/A, Directors) [documented behavior; strong]

Case 13 — RingCentral (VNK ownership-control case; initial decision not recovered)

Decision freeze

Decision date. Not recovered from the searched public record. The earliest decision date, security, price, valuation and initial vehicle allocation remain unknown; the surviving evidence is a later beneficial-ownership filing. (RingCentral 13G/A, Items 4–5 and reporting-person footnotes; source-map retrieval boundary) [researcher inference; insufficient public record]

Known then. No contemporaneous investment memo or dated announcement was recovered, so the memo does not backfill a cloud-communications thesis from RingCentral's later success. (RingCentral 13G/A, Items 4–5 and reporting-person footnotes; source-map retrieval boundary) [researcher inference; insufficient public record]

Unknown/disputed then. Sourcing, sponsor, diligence, approval, check size, competing alternatives, governance rights, follow-on decisions and the original risk model are all unresolved. This absence is itself the reason for including an ordinary-case control: public survivor stories systematically contain more detail than non-iconic holdings. (RingCentral 13G/A, Items 4–5 and reporting-person footnotes; source-map retrieval boundary) [researcher inference; strong]

Thesis and alternatives. No investor-stated ex-ante thesis was recovered. A generic “cloud communications” rationale would be title/sector inference and is therefore not used as evidence. The decision alternatives and rejected terms are also unknown. (RingCentral 13G/A, Items 4–5 and reporting-person footnotes; source-map retrieval boundary) [researcher inference; insufficient public record]

Risks/kill criteria. No contemporaneous KV kill criteria were recovered. Relevant business risks cannot be represented as the firm's actual checklist without a dated memo, testimony or board record. (RingCentral 13G/A, Items 4–5 and reporting-person footnotes; source-map retrieval boundary) [researcher inference; insufficient public record]

Terms, ownership and attribution. A February 2018 Schedule 13G/A reports VNK Management, L.L.C. as the direct owner of 1,030,033 shares and Vinod Khosla as the indirect beneficial owner of 1,032,033 shares, or 1.6%, while disclaiming beneficial ownership except to the extent of pecuniary interest; Khosla Family Trust reported zero shares. It also says the reporting persons ceased to be above 5% on November 6, 2017. This is a VNK-position record, not proof of personal title, Khosla Ventures fund ownership or sole deal sponsorship. (2018 Schedule 13G/A, Items 4–5) [contemporaneous record; strong]

Outcome and process lesson

The public filing shows a listed-company position and a drop below the 5% reporting threshold. It does not disclose sale proceeds, distributions, original cost, dilution or realized return, so no multiple can be calculated. (RingCentral 13G/A, Items 4–5 and footnotes; source-map retrieval boundary) [contemporaneous record; strong]

Process lesson. Denominator discipline requires incomplete, ordinary cases. If thesis quality can only be narrated for the famous winners and failures, the sample cannot support claims about a repeatable selection process. The next research priority is the earliest 13D/G chain and any pre-IPO offering document identifying original securities. (RingCentral 13G/A, Items 4–5 and footnotes; source-map retrieval boundary) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. RingCentral 13G/A, Items 4–5 [researcher inference; moderate]

1. Would you invest then? Mark “insufficient evidence” rather than inventing a historical decision packet.
2. What would kill the deal? Require a dated product, retention, sales-efficiency and financing plan before evaluating a hypothetical investment.
3. What ownership/terms fit the risk? Recover the first security, price, valuation, vehicle and dilution waterfall; the 2018 balance is not a substitute.
4. What evidence unlocks the next check? Earliest ownership filing, prospectus principal-holder table, board history and cash-distribution ledger.

Actual decision/result: The filing reports Vinod with a 1.6% indirect beneficial position, while VNK directly held 1,030,033 shares; the reporting group had fallen below 5%. The initial investment decision and audited economics remain unrecovered. (RingCentral 13G/A, Items 4–5 and footnotes; source-map retrieval boundary) [contemporaneous record; strong]

Case 14 — SKS Microfinance (personal Vinod Khosla investment; 2006–10 public-record case)

Decision freeze

Decision date. The SKS prospectus records Vinod Khosla's first personal allotment on March 27, 2006, followed by additional personal allotments. It must not be relabeled as a Khosla Ventures or Khosla Impact fund investment. (SEBI SKS prospectus, “History and Certain Corporate Matters—Share capital history” and “Principal Shareholders”) [contemporaneous record; strong]

Known then. SKS was scaling group-based microcredit to underserved Indian borrowers. The decision combined a large addressable financial-inclusion need with a model whose economics depended on repayment discipline, field execution and political permission. The prospectus later documented Khosla's personal ownership; it does not recover his 2006 memo or diligence record. (SEBI prospectus, Our Business/Risk Factors/Share capital/Principal and Selling Shareholders; Stanford microfinance report, paragraphs 3–6 and 9–10; official microfinance deck, slides 4–21) [contemporaneous record; moderate]

Unknown/disputed then. The investor's precise social-versus-financial objective, pricing assumptions, borrower-welfare safeguards, governance influence, liquidity plan and response thresholds are not public. The later Andhra Pradesh policy shock and industry crisis were not known outcomes, although regulatory and mission-drift risks were foreseeable categories. (SEBI prospectus, Our Business/Risk Factors/Share capital/Principal and Selling Shareholders; Stanford microfinance report, paragraphs 3–6 and 9–10; official microfinance deck, slides 4–21) [researcher inference; moderate]

Thesis and alternatives. The defensible high-level thesis was that scaled, professionally financed microcredit could expand access beyond charity-sized programs. Alternatives included nonprofit funding, slower state-by-state scaling, bank partnerships, stronger customer-protection covenants, or pass. The exact Khosla thesis and alternative set are not recovered. (SEBI prospectus, Our Business/Risk Factors/Share capital/Principal and Selling Shareholders; Stanford microfinance report, paragraphs 3–6 and 9–10; official microfinance deck, slides 4–21) [researcher inference; provisional]

Risks/kill criteria. Over-indebtedness, coercive collections, political backlash, state concentration, rapid loan-officer growth, funding mismatch and mission drift. A no-hindsight committee should require borrower-outcome and collections metrics alongside repayment and growth; there is no evidence these were Khosla's written 2006 criteria. (SEBI prospectus, Our Business/Risk Factors/Share capital/Principal and Selling Shareholders; Stanford microfinance report, paragraphs 3–6 and 9–10; official microfinance deck, slides 4–21) [researcher inference; strong]

Terms, ownership, governance and attribution. The prospectus reports 4,238,866 shares held personally by Vinod Khosla, 6.60% before the IPO. The public record searched here does not establish his board seat or approval authority over operating policy; the prospectus identifies him as not an IPO selling shareholder, while any later sale quantity and cash realization remain unknown. A prospectus mark is not a realized return. (SEBI SKS prospectus, “Principal Shareholders”) [contemporaneous record; strong]

Outcome and process lesson

CGAP reported SKS as India's first microfinance IPO, with about 5.8m clients, roughly 13-times subscription, and an approximately $1.5bn valuation. That establishes market reception, not Khosla's proceeds. The mapped sources do not provide a sufficiently complete post-IPO operating or realization record, so later crisis figures are not used here. (CGAP IPO/ownership/impact analysis; source-map retrieval boundary) [retrospective witness; moderate]

Process lesson. Social-impact underwriting needs two scorecards: enterprise economics and beneficiary/political legitimacy. High reported repayment is insufficient if customer harm, concentration, or regulatory permission can invalidate the operating model. Whether Khosla changed a later process because of SKS is not publicly documented. (SEBI prospectus, Share capital/Principal and Selling Shareholders; CGAP IPO/ownership/impact analysis) [researcher inference; strong]

No-hindsight exercise

The four prompts below are researcher-designed from the cited decision predicate; they are not recovered IC rules. SEBI prospectus, Business/Risk Factors [researcher inference; moderate]

1. Would you invest then? Decide in March 2006 using growth, repayment, borrower outcomes, state concentration and governance—not the later IPO.
2. What would kill the deal? Cap customer over-indebtedness, coercive collections, state concentration, funding mismatch and complaints; specify independent audits.
3. What ownership/terms fit the risk? Separate personal shares from any firm vehicle and require governance/customer-protection rights plus a dilution and liquidity plan.
4. What evidence unlocks the next check? Borrower-outcome sampling, repeat-loan burden, complaints, collection conduct, state mix, capital adequacy and funding tenor.

Actual decision/result: Khosla invested personally beginning in 2006, held 6.60% pre-IPO, and was not an IPO selling shareholder; the IPO was heavily subscribed. Later personal sales, cash flows, and realized return are not public. (SEBI prospectus, Share capital/Principal and Selling Shareholders; CGAP IPO/ownership/impact analysis) [documented behavior; strong]

Portfolio-level forensics

What the observed denominator can and cannot say

The most important control is an ordinary-case sample. Sort eligible non-marquee leads deterministically by normalized issuer name, exclude the 14 teaching cases, and select every k-th row until 20–30 companies are reached; then reconstruct the same fields—date, stage, vehicle, source/sponsor, ownership, reserves, governance, outcome, and cash flow. Repeat with a second deterministic offset. This can test whether lessons survive outside famous anecdotes. It must never be promoted into a fund-return estimate because the historical denominator and selection probabilities remain unknown.

Cross-case process evidence

• Square/Block documents tolerance for a bounded product experiment after the initial thesis changed; it does not show unconditional support for every pivot. Square case; Company Building 102, 00:13:57–00:14:23 (documented behavior / strong)
• Khosla describes parallel technical bets and willingness to accept failure. KiOR/Range are consistent with that portfolio design, but no public record proves either failure produced a tighter later reserve, governance, or stop-loss rule. KV Q&A, Range/planned-failure passage; retrieval boundary (retrospective witness / moderate)
• Juniper/Cerent have unusually strong Vinod operating evidence; Rocket Lab points to Sven; SKS is personal; Nextdoor mixes sponsor/PIPE; View has joint control; KiOR mixes fund, trust, and VNK instruments. people and influence map; case-role map (contemporaneous record / strong)
• OpenAI’s later value, Google’s later dominance, Impossible’s later financing, and Rocket Lab’s public liquidity are outcomes, never inputs to their decision freezes. OpenAI case; Impossible case; Rocket Lab case (researcher inference / strong)
• No complete rejected-deal or middling-holdings ledger exists. RingCentral shows how thin an ordinary public case can be, so broad claims about selection skill remain provisional. RingCentral case; source-map retrieval boundary [researcher inference; strong]

Failure, miss, and anti-portfolio postmortems

These postmortems are not moral verdicts or proof of negligence. The original thesis, contemporaneous warnings, later outcome, and legal posture are separated. Martins Beach is excluded from venture performance and appears only in profile governance context.

Uncertainty and retrieval ledger

Public discovery included current and historical KV pages; all nine portfolio category pages; current IAPD/ADV; EDGAR Form D indexes/documents and issuer registration, ownership, current-report, merger, bankruptcy, and litigation exhibits; NJ, CalPERS, North Dakota, and registered-fund reports; SEC releases/orders/complaints; Delaware, Mississippi, California, and federal opinions; DOE/USDA records; SEBI; KPCB/company histories; Khosla Impact and Labs entity/name variants; and named adverse, return, write-off, board, founder, acquisition, and bankruptcy queries. No archive or person was contacted. source-map inventory and coverage [researcher inference; strong]

The remaining conclusions are literal. They follow from the ledger reconciliation, public-LP comparability limits, and source-map retrieval boundary, not from assumed private data. [researcher inference; strong]

• Complete career investment count: not computable.
• Historical and person-level coverage fraction: not computable.
• Firmwide hit rate, loss rate, gross/net MOIC, IRR, and total value: not computable.
• Vinod-personal return and partner carry: not computable.
• Juniper 2,500x, OpenAI “more than 100x,” Cerent KPCB proceeds, and cleantech aggregate multiple: not independently computable.
• Realized versus unrealized contribution of marquee cases: not computable.
• Scoped eligible-public-lead reconstruction is only 27/161 = 16.8%.
• Source and method details remain auditable in sources.md, while thesis evolution and repeated interested narratives are bounded in writings.md and talks.md.

Asymmetric experiments, operating assistance, and auditable limits

Across six decades, Khosla’s writing converges on a decision system: seek non-incremental technical outcomes, expose fatal risks early, buy information with bounded experiments, stage capital behind evidence, and help founders recruit or decide without confusing advice with operational control. Eighteen-model synthesis [researcher inference · strong]

This synthesis tests that system across 62 distinct works = 44 ★★★ core + 18 ★★ supplementary, supported by 48 full + 13 partial + 1 unavailable work records and 62 unique notes. Six ★ announcement or version routes do not create additional works. Source-map inventory and access reconciliation [contemporaneous record · strong] Access permits analysis of the declared text; it does not make company claims audited, prove returns, or assign firm activity to Vinod personally. The unavailable Innovator’s Ecosystem remains bibliographic only, and no missing or paywalled body is reconstructed. [researcher inference · strong]

Source-by-source analysis

1. Sun Microsystems Business Plan (1982)

Argument, reasoning, and evidence. The plan begins with a narrow OEM wedge: modular, networked CAD/CAM workstations built from open standards, sold at a fraction of incumbent prices, then extended into a compatible family. It joins architecture, university visibility, cost reduction, staffing, patents, manufacturing, and financing in one dated operating model. Its 500-unit/$4 million first-year and 1,500-unit/$10 million second-year figures are forecasts, not results. Primary work; PDF page and exhibit printed page (PDF pp. 1–12 / exhibit pp. 23–34) · pinpoint evidence note [contemporaneous record · strong]

Assumptions and decision implications. It assumes standards-based integration can beat vertically integrated incumbents and OEM distribution can deliver scale without surrendering product advantage. The durable diligence rule is to test deep technology as a coupled system—technical edge, cost-performance, channel, working capital, hiring, and milestones—rather than underwrite invention alone. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Later writings preserve open architectures, cost advantage, and talent density, but add power-law risk and board assistance. Sun’s later fame must not backfill every 1982 projection as correct; actual plan-versus-result reconciliation remains missing. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full image-only HBS exhibit was inspected. The landing labels Khosla as author, while the plan itself has no sole byline and assigns Bechtolsheim product/engineering, Khosla management/finance/strategy, and McNealy operations. Use it as an early-team record, with Khosla’s role bounded accordingly. Source note

2. The Entrepreneurial Roller Coaster (first version reported 1986; composite public deck through 2002)

Argument, reasoning, and evidence. Khosla’s early operating model combines people, leverage, an implemented customer value proposition, urgency, managed conflict, tolerance for mistakes, and transition from exploration to execution. The startup sequence is technology change → market wave → diverse team → rule change → leveraged wedge → long-term defensibility; the KPCB appendix adds unfair advantage, “risk up front,” and home-run ambition. Primary work; slide number · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. Diversity and argument are assumed to improve risk coverage, while an early wedge is assumed to compound into an asset. Investors can translate this into a risk-to-hire map and explicit wedge-to-moat test; “organized chaos” should end where evidence, budget, and execution discipline begin. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The deck foreshadows intelligent failure, gene-pool engineering, and founder assistance. Its tolerance-for-error slogans lack loss budgets and stopping rules, a gap later partly addressed by the Risk Matrix but never resolved at fund-denominator level. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The accessible deck is complete but chronologically composite: Khosla later dates its origin to 1986, the landing says 2001, and the appendix contains end-2002 KPCB statistics. Those statistics are firm-level self-report, not Vinod-only return evidence. Source note

3. RISC: A Model for Implementing the Bicycle Commute Economy (2003)

Argument, reasoning, and evidence. Khosla and Atanu Dey propose about 5,000 rural hubs within bicycle reach, aggregating demand for power, broadband, buildings, markets, finance, education, health, and government services. The mechanism is minimum viable density: shared infrastructure lowers unit cost; competing public, private, and nonprofit providers populate a service layer. Primary work; PDF printed page numbers 1–39; landing by paragraph under title · pinpoint evidence note [contemporaneous record · strong]

Assumptions and decision implications. The authors explicitly call RISC necessary but insufficient and concede that an entrepreneurial 5% generating gains for the other 95% is arbitrary. Modern infrastructure diligence should test utilization, travel radius, willingness to pay, inclusion, and unit economics before extrapolating development impact from platform density. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is an early version of Khosla’s resource-multiplier thesis, later expanded to health, AI, and societal infrastructure. It resists technology-only development claims, yet offers no implementation results showing the proposed centers improved income or distribution. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full 39-page paper is available through the current official route; the landing is only partial and misspells Dey’s name. Attribution is joint, the November 2003 paper predates the 2006 landing, and the 2024-regenerated PDF is a recovered version—not a new work. Source note

4. The Device That Used to Be Your Phone (landing 2004; later-revision evidence)

Argument, reasoning, and evidence. The deck treats mobile as a platform across transport, ads, payments, software portability, media, diagnostics, translation, and machine-to-machine services rather than as a handset. Its strongest reasoning identifies who pays and what constrains adoption: data pricing and battery life can block otherwise plausible products. Primary work; slides 1–45; landing/revision clock warning · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. Forecast categories mix installed services, near-term products, and speculation. A modern investor should score direction, timing, economics, bottleneck resolution, and value capture separately; a correct prediction that a product appears does not identify its winning company. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The platform-and-bottleneck approach recurs in later AI and climate maps. Directional hits such as mobile money cannot validate uncited market-size estimates or obscure misses on schedule and ownership. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The complete current official CDN deck spans slides 1–45 and uses a KV email, but the 2004 landing conflicts with 2006–07 data in the asset. It is a later revision, not a clean 2004 forecast set, and its examples reveal no vehicle, ownership, or return. Source note

5. Microfinance: Enabling the Power of Ideas and Entrepreneurial Energy for the Other Half (2004)

Argument, reasoning, and evidence. Khosla frames microfinance as institution building: professional boards, audits, ratings, market-priced funding, incentives, securitization, and technology should make entrepreneurial lending scalable rather than permanently subsidized. SHARE, Grameen, SKS, and campaign data illustrate client growth, leverage, and financial self-sufficiency. Primary work; slides 1–36; saved official landing-to-CDN link · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. Repayment and scale are assumed to coexist with beneficiary welfare, but Khosla explicitly flags exclusion of the poorest, mission drift, loss of services, and outside control. Impact underwriting should therefore put sustainability, borrower outcomes, local governance, pricing, and securitization incentives in one scorecard. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The later Architecture of Audacity adds network reporting and a clearer correlation-not-causation warning. Subsequent SKS controversy means the 2004 “do well by doing good” frame is a hypothesis requiring outcome records, not validation. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The complete official CDN deck spans slides 1–36, identifies Khosla through a KPCB email, and predates Khosla Impact; it must not be assigned to that later vehicle. The canonical landing returned 404 on final live check, but its saved official HTML preserves the live CDN asset link. Company/campaign figures are mixed-source and targets for 2004–08 are plans, not achieved results. Source note

6. The Architecture of Audacity (2007)

Argument, reasoning, and evidence. Susan Davis and Khosla analyze the Microcredit Summit Campaign as an organization and movement: an audacious numeric goal, priorities for the poorest and women, financial sustainability, convening, tools, and public reporting mobilized an ecosystem. Survey, interview, and campaign records support increased participation and legitimacy, while the authors explicitly state that correlation is not causation. Primary work; PDF pp. 1–22 / printed pp. 159–180 · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The causal claim depends on an unobservable counterfactual and potentially selective reporting. The transferable ecosystem rule is measurable shared goals plus reporting and network bridges, paired with beneficiary-level metrics and an explicit test of what would have happened without the coordinating institution. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Compared with the 2004 deck, the paper is more cautious about commercialization crowding out client transformation. It still calls the campaign highly successful despite conceding limited impact validation, so later borrower evidence and crises must remain outside the authors’ 2007 window. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The complete current official CDN paper spans PDF pp. 1–22 / printed pp. 159–180. The byline is Susan Davis then Vinod Khosla, and Grameen Foundation roles/support materially shape the perspective. It is not sole Khosla authorship, independent campaign validation, or a Khosla Impact investment record. Source note

7. A Simpler Path to Cutting Carbon Emissions (2010)

Argument, reasoning, and evidence. Khosla argues a weak carbon price would entrench incrementalism; five years of technology development and technology-neutral performance standards could instead produce electricity, fuel, vehicle, and efficiency solutions with materially lower emissions. He proposes an updating fee-bate and a 2020 electricity target, treating policy as insurance against climate risk. Primary work; HTML paragraph by unique opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The approach assumes radical cost declines arrive within the innovation window and that performance standards preserve architectural competition. Investors should test whether a company can meet an outcome threshold without permanent category privilege, while modeling permitting, infrastructure, financing, and near-term lock-in. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay says government should stop picking winners while advocating standards, credits, guarantees, and fee-bates. The coherent distinction is neutrality among qualifying technologies, not absence of intervention; the unresolved risk is emissions lost if breakthroughs arrive late. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full KV mirror preserves Khosla’s Washington Post op-ed. Its $40 carbon threshold, emissions shares, and 2020 target lack reproduced methods, and no portfolio company, vehicle, or ownership is disclosed. Source note

8. Innovation and Innovators (2010)

Argument, reasoning, and evidence. The 65-slide deck begins with expert forecast failures and assumption inversion, then favors hard technical problems, early iteration, dissent, and new capabilities over top-down market sizing, customer wish lists, or stage-gate discovery. “Being right” is insufficient if a company merely joins a crowded bandwagon; the investable case needs a differentiating contribution. Primary work; slides 1–65 · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. Khosla assumes technical discontinuity plus winner-take-most structure can dominate conventional forecast error. The usable rule is to map consensus assumptions, specify the measurable break, and fund cheap tests; rejecting NPV during discovery cannot waive unit economics and capital discipline after uncertainty falls. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This deck anticipates intelligent failure and black-swan selection. It criticizes expertise while relying on external charts and selected companies, so the proper reading is adversarial technical diligence—not anti-expert intuition. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The complete current official CDN deck supports locators across slides 1–65, but many charts are thinly sourced. Passif, Square, SKS, Soraa, and others are live examples without vehicles, ownership, partner credit, or outcomes; SKS is a required counterexample, not an impact proof. Source note

9. Corn Ethanol: Time to Move On (2010)

Argument, reasoning, and evidence. Khosla argues corn-ethanol tax support had completed its incubation job and should expire. His general rule funds technologies with societal benefit, remaining technical/scale risk, and a path to competitiveness for roughly five to seven years, then sunsets support as a class reaches scale. Primary work; unique paragraph openings and footnotes (1)–(3) · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. Subsidy becomes a temporary bridge rather than durable revenue. Investors should model the cliff, learning curve, externality, public additionality, and lobbying response before treating protected demand as company value. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay predicts several KV biofuel companies could reach unsubsidized economics, while allowing that some would fail. KiOR, Range, Amyris, and other later outcomes make the December 2010 predictions scoreable and expose the difference between policy logic, technical deployment, and LP return. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full mirror names HCL, KiOR, Coskata, Lanza, Range, Gevo, LS9, and Amyris as a firm portfolio. Khosla expressly acknowledges that reallocating support away from corn could benefit his interests; disclosure does not independently validate lifecycle, cost, or public-value claims. Source note

10. Foreign Policy: Long Shots (2010)

Argument, reasoning, and evidence. Khosla defines climate innovation as a portfolio of high-failure-probability technologies with enormous potential cost and emissions discontinuities. A few Google-like outcomes can dominate thousands of failures; current technologies may be deployed for learning, while economics—not permanent subsidy—must drive global adoption. Primary work; HTML paragraph by unique opening phrase · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The model assumes independent shots, bounded experiment costs, extreme positive skew, and eventual unsubsidized scale. A modern portfolio needs technical milestones, capital-at-risk limits, ownership math, and kill criteria so “chaos” does not rationalize unbounded follow-ons. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the concise precursor to the 2011 black-swan paper and 2020 instigator thesis. It preserves a real policy tension: limited near-term deployment may conserve innovation capital but creates emissions and lock-in risk if breakthroughs miss their schedule. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. Full mirror access supports the argument, but market forecasts and examples were not methodologically reproduced. The “10 of 10,000” line is illustrative, not KV’s disclosed denominator, and unnamed portfolio claims cannot support personal attribution or returns. Source note

11. Black Swans Thesis of Energy Transformation (2011)

Argument, reasoning, and evidence. Khosla defines an energy “black swan” as a technology that can compete without subsidy at global scale and at prices relevant to China and India. Because most candidates will fail, the paper argues for thousands of technically distinct attempts so that a small number of extreme winners can materially change deployment; incremental technologies still matter as components and learning steps, but are not the portfolio’s intended endpoint. Primary work; PDF printed pages 1–40; landing by unique paragraph phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The model assumes power-law technical outcomes, sufficiently independent experiments, and disciplined follow-on capital. Its practical test is ex-ante: ask whether a proposal can reach unsubsidized cost, scale, and material emissions impact, then size an early experiment cheaply enough that a roughly 90% failure expectation is survivable. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the clearest early statement of the doctrine later applied to climate, health, and AI, but it repurposes Taleb’s retrospective term as a prospective search strategy. The paper supplies neither an audited portfolio denominator nor realized return evidence; later company outcomes and Khosla’s dated 2020–25 forecasts must therefore score the framework rather than be used as proof of it. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official paper is fully accessible and attributed to Vinod Khosla; the current PDF appears regenerated while retaining a 2011 work date. Portfolio examples and outcome characterizations are interested evidence from an investor advocating the strategy, not independent validation. Source note

12. The Innovator’s Ecosystem (2011/2012 route)

Argument, reasoning, and evidence. No substantive argument can be responsibly reconstructed: the canonical PDF and the discovered dated route are unavailable, and only bibliographic traces remain. The title suggests an ecosystem thesis, but that inference is not evidence and is deliberately excluded from the analytical corpus. Primary work; no body locator available; legacy filename/version route only · pinpoint evidence note [researcher inference · insufficient public record]

Assumptions and decision implications. The decision-relevant lesson is evidentiary, not thematic. A bibliography or filename can establish that a work was referenced, but cannot support claims about its reasoning, recommendations, examples, or relationship to Khosla’s investment process. Source-note evidence · open questions [researcher inference · insufficient public record]

Evolution and contradictions. The official filename points to 2011-12-19 while a legacy dated route places the work in 2012, so both clocks remain unresolved. Live retrieval yielded no body from either route; the analysis should change only after recovering a complete artifact or reliable archive copy. Source-note contradiction and version ledger [researcher inference · insufficient public record]

Access, authorship, and conflicts. This is the corpus’s one unavailable work and a bibliographic-only recovery target. No quotation, thesis, or personal-authorship claim is drawn from the missing body. Source note

13. Guide for Performance Management (2012)

Argument, reasoning, and evidence. The guide treats performance management as a continuous evidence system: align expectations, collect feedback upward, downward, and across peers, compare it with self-assessment, and deliver themes close enough to the behavior to enable correction. Its templates emphasize purpose-specific cadence, documented disagreement, and avoiding end-of-cycle surprises. Primary work; PDF printed pages 1–3 and named sections · pinpoint evidence note [contemporaneous record · strong]

Assumptions and decision implications. It assumes multi-rater observation improves calibration and that managers can preserve candor without turning anonymous feedback into politics. For company-building diligence, the useful tests are whether expectations are explicit, feedback is timely, conflicting accounts are reconciled, and development actions are actually followed. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The operating specificity anticipates Khosla Ventures’ later “venture assistance” materials, but the document offers no causal evidence that its process improves retention or performance. Anonymous feedback can introduce bias or retaliation risk, and the embedded legal caution is historical language rather than current legal advice. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official landing and document are fully accessible as institutional Khosla Ventures guidance. The landing associates the material with Vinod, while the PDF lacks a clear personal byline and carries Deborah Biondolillo metadata; authorship should therefore remain firm-level unless independently resolved. Source note

14. Do We Need Doctors or Algorithms? (2012)

Argument, reasoning, and evidence. Khosla proposes a staged medical-automation path: systems first assist clinicians, then offer second opinions, become decision support and first opinions, and ultimately refer only hard cases to humans. The reasoning is comparative—software can integrate more data than the median practitioner—while preserving a role for the best 10–20% of doctors during a projected ten-to-fifteen-year transition. Primary work; unique section phrases and intact paragraphs; exclude the corrupted passage identified below · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The thesis assumes data quality, prospective validation, workflow adoption, and regulation advance with model capability. An investable version should stage autonomy behind measurable clinical evidence, define escalation and liability boundaries, and separate task substitution from the much stronger claim that whole professions disappear. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. This is the early outline later elaborated in the 2014 reinvention essay and the 2016/2019 “20 Percent Doctor” works. The timetable and top-percentile claims were not operationally defined, while recent joint work with Neal Khosla places greater weight on risk-bearing deployment and system incentives. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The migrated official HTML is only partially usable because one passage is corrupted and contains unrelated links; that passage is excluded. Vinod is the named author, and his Jawbone interest is disclosed without resolving whether it was a personal or fund vehicle holding. Source note

15. Will We Need Teachers or Algorithms? (2012)

Argument, reasoning, and evidence. The essay argues that software can make mastery the fixed variable and time the flexible one, taking over instruction and adaptation while teachers become mentors and coaches. CK-12, Khan Academy, low-cost online experiments, and learner telemetry are offered as mechanisms, while peer interaction and social development remain explicitly human. Primary work; named section phrases and unique paragraph openings · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. It assumes usage telemetry maps to genuine learning and that personalization can scale without worsening access or privacy problems. Investors should require learning outcomes rather than engagement proxies, compare cohorts against credible baselines, and test whether software complements or merely shifts work onto teachers. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay extends Khosla’s medical-automation logic to education, but its “top 20%/80%” framing is undefined and the examples do not establish causality. Later AI-abundance writing broadens the access thesis while making governance, connectivity, and institutional trust more visible constraints. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod as the third TechCrunch installment. CK-12 is connected to his family through his wife’s organization, a material relationship when weighing the example. Source note

16. Maintain the Silicon Valley Vision (2012)

Argument, reasoning, and evidence. Khosla distinguishes founders pursuing a durable mission from teams built primarily for acquisition, arguing that acquisition should remain an option rather than the founding objective. Retrospective Sun, Juniper, and Cisco counterfactuals support a broader claim that ambitious companies can create platforms and ecosystems beyond the original product. Primary work; KV mirror by unique paragraph opening; original New York Times publisher context only partly exposed · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The thesis assumes founder motivation is legible and predictive. In diligence, it is better tested through repeated resource choices, recruiting, product scope, and willingness to endure setbacks than through mission rhetoric alone; an investor should also model when a sale is rational despite authentic ambition. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The missionary-versus-mercenary distinction recurs in later founder-mode and recruiting work, but the evidence is survivor-selected and the Juniper/Cisco stories are retrospective. A company can hold a large vision and still destroy capital, while a disciplined acquisition can be the highest-value outcome. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The New York Times publisher context is only partially available, but the visible official mirror preserves the authored opinion. John Doerr’s views and deal anecdotes remain participant accounts rather than contemporaneous transaction records. Source note

17. Do You Need to Be a Jerk to Be a Successful Entrepreneur? (2012)

Argument, reasoning, and evidence. The essay separates demanding standards from personal denigration: critique the work, preserve fairness and ethics, and remain obstinate about vision but flexible about tactics. Team quality and loyalty are presented as strategic assets, supported by Sun’s equal-equity origin story and Khosla’s account of repeated recruiting dinners. Primary work; unique paragraph openings and numbered Steve Jobs examples 1–4 · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. It assumes boards and teams can distinguish productive candor from abuse before damage compounds. A diligence process should therefore examine attrition, complaint handling, promotion fairness, reference patterns, and whether pressure improves decisions rather than relying on a founder’s self-description. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The argument complements later recruiting and leadership work, yet its personality labels are undefined, its Steve Jobs discussion is speculative, and it offers no governance mechanism or causal evidence. High standards can coexist with misconduct, making observable controls more reliable than archetypes. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. The Sun anecdotes are retrospective self-report and should not be treated as independent evidence about team dynamics or ownership fairness. Source note

18. If, When and How to Avoid Hiring a CEO (2012)

Argument, reasoning, and evidence. Khosla favors keeping a capable founder as CEO, surrounding that person with complementary operators, 360-degree feedback, and coaching. He distinguishes exploratory markets, where founder judgment matters most, from known operating playbooks, while reserving CEO hiring and firing as the board’s exceptional decision. Primary work; named sections and unique paragraph openings · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The model assumes role fit changes as product, commercialization, and organizational complexity change. Boards should diagnose missing capabilities explicitly, test whether they can be added around the founder, and define milestones that would make a leadership transition necessary rather than importing a generic “professional CEO.” Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This founder-first doctrine connects to the later CEO-signals essay and 2024 hiring guidance, but it provides no base rates or decision thresholds. Its narrow operating description is not a complete account of formal board authority, governing law, or corporate and financing documents. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. Lookout, Square, and personnel examples involve Khosla Ventures relationships; firm, team-member, and Vinod-specific contributions must remain separated. Source note

19. Signs It Might Be Time to Consider Hiring a CEO (2012)

Argument, reasoning, and evidence. The follow-up identifies organizational signals for a transition: a hub-and-spoke founder structure may work below roughly 30–40 people, but coordination, orchestration, and commercialization eventually become distinct jobs. It advises hiring before the problem becomes acute, prioritizing shared values and complementary skill, and treating the choice as reversible. Primary work; growth stages 1–3, named section “Avoiding Common Pitfalls Of Management Transitions,” and unique paragraph openings · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The framework assumes stage and task, not prestige, determine fit; it also cites an approximately 30% outside-hire failure rate without supplying a denominator. A board can convert the heuristics into a dashboard of decision bottlenecks, executive gaps, sales repeatability, attrition, and founder learning rate. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay makes the prior founder-first stance more conditional and anticipates the 2024 CEO-quality checklist. Thresholds remain heuristic and sector-dependent, while LanzaTech and Soraa are selected illustrations rather than comparative evidence. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. Statements from participating executives are evidence of their views, not Khosla’s independent verification of company performance or hiring outcomes. Source note

20. The Big Green Opportunity (landing 2012-11-27; PDF 2012-11-24)

Argument, reasoning, and evidence. Khosla argues that climate technologies must become competitive “main tech,” not subsidy-dependent niches, with costs relevant to China and India. The portfolio method is diversification plus staged capital: make small early bets, fund follow-ons only as technical and market risks fall, and abandon projects that cannot establish durable differentiation. Primary work; HTML section heading and paragraph; linked PDF page · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The method assumes technical milestones are informative, follow-on discipline survives sunk-cost bias, and public policy does not mask weak unit economics. Investors should separate science, scale-up, financing, and adoption risks; predefine kill criteria; and test unsubsidized cost rather than extrapolating a laboratory result. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The presentation is a bridge from the 2011 black-swan thesis to later climate forecasts, contrasting projects such as Range Fuels, KiOR, Ausra, and Stion. Its self-reported clean-tech return comparison and claims about eleven financings and roughly $500 million are unaudited, while later company outcomes provide counterevidence to promotional framing. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official landing and PDF are fully accessible, with a live landing date of 2012-11-27 and PDF date of 2012-11-24 despite catalog-date conflict. Examples and economics are firm-level, interested claims; they do not establish Vinod’s personal capital, sourcing, or realized proceeds. Source note

21. Technology Will Replace 80% of What Doctors Do (2012)

Argument, reasoning, and evidence. Khosla decomposes medicine into tasks and argues that data-rich systems can replace much routine diagnosis and treatment while humans remain the patient-facing layer. He expects outsiders to build the disruptive stack and accepts a high innovator failure rate, using examples such as Lifecom to illustrate components rather than a complete autonomous system. Primary work; named sections and unique paragraph openings · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The argument assumes clinical work is sufficiently decomposable, machine evidence can generalize, and patients and regulators will accept new accountability structures. Investors should underwrite specific tasks, prospective outcomes, escalation rules, and workflow economics instead of treating “80% of doctors” as a measurable product milestone. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This sharper occupational claim follows the earlier staged algorithm essay and precedes the more detailed v0–v7 framework. The rhetorical profession-level forecast conflicts with later task-level caveats, and the cited products do not establish the safety, adoption, or timing of wholesale substitution. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official mirror is accessible and attributed to Vinod. Portfolio-company references are selected interested examples; a lost migration asterisk appears to have carried investment disclosure, so the surviving page cannot support precise vehicle attribution. Source note

22. Venture Assistance: What Boards Should and Should Not Do (2013)

Argument, reasoning, and evidence. Khosla defines a board as an active source of recruiting, strategy, introductions, and difficult debate, while leaving operating decisions to management except for selecting or removing the CEO. The essay is unusually procedural: assistance should be requested and useful, disagreement should be explicit, and directors should avoid converting advice into control. Primary work; HTML section and distinctive opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The model assumes founders can reject advice without hidden financing penalties and directors can distinguish influence from authority. Its diligence tests are concrete: record who proposed major decisions, whether alternatives were debated, whether help produced observable results, and whether management retained real agency. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the governance core of later “venture assistance” marketing and the 2026 Q&A. Khosla’s claim that he has never needed to vote against a founder is self-report. The essay's operating/non-operating division is philosophy, not a complete account of formal board authority, governing law, or contractual rights. Primary essay, “Decision-making should be a sharp line” · source-note contradiction and version ledger [researcher inference · strong]

Access, authorship, and conflicts. The full official HTML essay is accessible and attributed to Vinod. It describes an intended board philosophy, not audited evidence that every Khosla Ventures director or portfolio relationship followed it. Source note

23. Designing the Future: Making Technology More Accessible (2014)

Argument, reasoning, and evidence. The paper treats design as an operating capability that converts technical possibility into understandable, usable, and desirable products. Irene Au’s operating-partner role and a network of practitioner advisers illustrate a firm-level mechanism for embedding design expertise in portfolio work rather than reserving it for late-stage polish. Primary work; unique paragraph openings 1–6 · pinpoint evidence note [contemporaneous record · moderate]

Assumptions and decision implications. The framework assumes better user research and interface decisions translate into adoption and defensibility. Investors can test whether design is represented in product discovery, whether accessibility and usability have measured outcomes, and whether advisory support changes decisions rather than merely supplying branding. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. It broadens the firm’s assistance thesis from boards and recruiting to a functional center of excellence. The examples are selected and provide no counterfactual performance evidence; despite the title, “accessible” primarily means broad usability and market reach, not a documented disability-access program. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official work is fully accessible as Khosla Ventures material, but personal authorship is not established merely by its placement in Vinod’s archive. Adviser roles and company examples are firm-level relationships and should not be converted into Vinod-specific contribution claims. Source note

24. The Case for Intelligent Failure to Invent the Future (2014)

Argument, reasoning, and evidence. Khosla argues that consequential invention requires ambitious experiments whose individual failure probability can approach 90%, provided each failure is designed to produce information. He contrasts controllable execution risk with genuine invention risk and suggests retaining roughly 30–40% control over outcome variables while accepting uncertainty elsewhere. Primary work; HTML paragraph and distinctive opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. Intelligent failure requires bounded loss, explicit hypotheses, measurable milestones, and a rule for stopping or redirecting capital. A portfolio should therefore budget failure ex ante and distinguish learning that changes the next decision from retrospective storytelling that merely relabels a loss. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay generalizes the energy black-swan doctrine into a company-building philosophy and later recurs in AI and climate. It supplies no denominator showing which failures produced transferable learning, and a culture celebrating failure can weaken accountability unless kill criteria and follow-on discipline are observable. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. Portfolio examples and the 90% framing express his investment philosophy, not audited loss rates, ownership economics, or proof that unsuccessful investments were intentionally structured experiments. Source note

25. The Reinvention of Medicine: Dr. Algorithm v0–v7 and Beyond (2014)

Argument, reasoning, and evidence. The essay reframes medicine from practice toward data-driven science through successive “Dr. Algorithm” versions: narrow assistance, integrated decision support, increasingly autonomous recommendations, and eventual referral of exceptional cases. A five-by-five matrix of specialties and functions turns a sweeping thesis into investable wedges, with outsiders expected to move faster than incumbents. Primary work; HTML paragraph and v0–v7 transition passages · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The sequence assumes data access, validation, regulation, reimbursement, trust, and workflow integration improve together. Investors should fund bounded clinical wedges, specify the evidence required before each autonomy step, and price adoption and liability risk separately from raw model performance. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. It is more operational than the 2012 essays and becomes the template for the 2016 paper and 2019 deck. The version ladder creates scoreable forecasts, but many examples are demonstrations or portfolio claims rather than outcomes, and the timetable does not itself answer distribution shift or patient-safety questions. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official HTML is accessible and attributed to Vinod. Named health companies illustrate an investor’s thesis and may involve Khosla Ventures interests; their inclusion does not independently validate product efficacy or establish personal deal attribution. Source note

26. The Next Technology Revolution Will Drive Abundance and Income Disparity (2014)

Argument, reasoning, and evidence. Khosla predicts that machine expertise will make many goods and services abundant while reducing the market value of routine human labor, producing prosperity and inequality at the same time. The essay links automation across medicine, education, transport, and professional services to a political need for redistribution and transition support. Primary work; canonical suffixed-route title/byline/date/summary; complete unsuffixed-route body by unique paragraph opening; migrated-title warning · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The thesis assumes capability improvement converts into broad substitution, productivity gains lower prices, and ownership remains concentrated enough to widen income disparity. Investment cases should model sector-specific demand elasticity, adoption friction, labor complements, and who captures savings rather than inferring economics directly from technical capability. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Written before modern generative AI, it establishes the abundance-versus-displacement tension later developed in AI: Dystopia or Utopia? and A Roadmap to AI Utopia. Its occupational generalizations remain under-specified, and abundance can increase demand for human complements rather than reduce total employment. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. Paired official KV routes provide full work access: the canonical suffixed route preserves the correct title, Vinod byline, date, and summary, while the unsuffixed route preserves the complete body under the wrong migrated title. The version warning remains material. Vinod is the attributed author, but forward-looking social claims are forecasts rather than verified portfolio evidence. Source note

27. The Risk Matrix (2015)

Argument, reasoning, and evidence. The framework requires a company to enumerate failure modes, score probability and consequence, identify milestones that retire each risk, and graph the time and capital needed. It shifts planning from a single optimistic roadmap to an explicit portfolio of technical, market, team, regulatory, and financing uncertainties. Primary work; PDF physical pages 1–10 and printed labels “PAGE 1 of 9” through erroneous “PAGE 10 of 9”; Steps 1–5 · pinpoint evidence note [contemporaneous record · strong]

Assumptions and decision implications. The matrix assumes scores are directionally calibrated, risks can be decomposed, and milestones provide genuine information. Its strongest use is comparative: expose hidden assumptions, sequence the cheapest discriminating tests, reserve capital for correlated failures, and update scores when evidence arrives. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the operating counterpart to intelligent failure—failure is acceptable only when experiments retire named risks. Numeric scores can create false precision, dependencies can invalidate additive rankings, and the document supplies neither validation data nor proof that portfolio follow-ons consistently followed the method. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official guide is fully accessible as Khosla Ventures material, but personal authorship is not clearly established. Treat it as institutional process guidance and keep portfolio outcomes separate from the framework’s descriptive claims. Source note

28. Team 360 Reviews (2015)

Argument, reasoning, and evidence. The guide combines stakeholder interviews, self-assessment, cultural themes, and a value-versus-performance matrix to diagnose an executive team. It asks reviewers to convert patterns into a small set of actions and revisit progress, making the exercise a development and team-design tool rather than a one-time score. Primary work; PDF printed pages 1–3; named blocks “Who,” “What,” “Review,” “How,” “Key questions,” and page-3 matrix · pinpoint evidence note [contemporaneous record · strong]

Assumptions and decision implications. The process assumes respondents speak candidly and facilitators can distinguish signal from coalition politics. Useful controls include representative sampling, protected reporting, evidence for high-impact claims, an appeals path, and separation of development feedback from decisions with legal consequences. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. It extends the 2012 performance guide into a team-wide intervention and complements recruiting and CEO-transition work. Anonymity can protect candor but also conceal bias; the value/performance grid risks oversimplifying context, and no evidence is offered that the method predicts company results. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official document is fully accessible as institutional Khosla Ventures guidance; it does not establish Vinod as sole author or facilitator. The templates are historical operating material, not current employment or legal advice. Source note

29. The Art, Science and Labor of Recruiting (2015)

Argument, reasoning, and evidence. Khosla treats recruiting as a founder-led, high-volume search rather than delegation to a recruiter. The process broadens the funnel, evaluates actual contributions instead of titles, tests whether candidates can recruit others, and uses adversarial references to discover failure patterns; the Dynabook story illustrates the cost of underbuilding the team. Primary work; HTML heading and numbered interview prompts · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The approach assumes intensive reference work predicts performance and founders can allocate the required time without narrowing diversity through network bias. Investors can audit funnel breadth, evidence behind scorecards, reference independence, closing ownership, and post-hire outcomes by role. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay operationalizes the people side of “venture assistance” and feeds directly into gene-pool engineering and later hiring work. Its anecdotes are retrospective, the labor intensity may not scale, and no comparative hiring denominator shows which practices caused better outcomes. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. Dynabook and other examples are his recollections; they are useful process evidence but not independent proof of causation or company performance. Source note

30. Gene Pool Engineering for Entrepreneurs (2015)

Argument, reasoning, and evidence. Khosla asks founders to name the five risks most likely to kill the company and the five opportunities most likely to transform it, then recruit “centers of excellence” around both lists. Diverse experience is framed as a way to expand the organization’s problem-solving gene pool rather than merely fill a conventional org chart. Primary work; HTML heading and seven-step numbered process · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The method assumes talent gaps can be inferred from strategic risks and that diversity of relevant cognition improves decisions. A board can map each critical uncertainty to accountable expertise, test whether hires change milestone velocity, and avoid confusing recognizable pedigree with the needed capability. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This converts the risk matrix into a human-capital plan and extends the recruiting essay’s emphasis on contribution over title. The anonymized ePowersoft case limits verification, while analogies involving age, language, or “genes” can encode stereotypes unless selection criteria are explicit and job-related. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The official work is fully accessible and attributed to Vinod; HTML and document versions are the same underlying work. Company details are anonymized, so the account cannot be independently matched or used as outcome proof. Source note

31. Is Majoring in Liberal Arts a Mistake for Students? (2016)

Argument, reasoning, and evidence. Khosla argues for a “modern thinking” foundation—quantitative reasoning, scientific method, economics, psychology, and epistemic discipline—before broad cross-domain study. The target is not the humanities themselves but curricula that, in his view, leave students unable to test claims or adapt to technical change; personal education and hiring anecdotes supply most of the support. Primary work; named headings, unique paragraph openings, and “Additional Responses to Comments and Questions” prompts · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The argument assumes degree labels proxy for reasoning skill and that a common analytical core improves adaptability. Hiring or founder assessment should test actual learning velocity, numerical literacy, causal reasoning, and cross-domain synthesis rather than using major choice as a shortcut. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay connects Khosla’s systems-thinking style to later calls for AI-era reskilling, but it admits anecdotal grounding and never establishes the claimed median-performance gap. Its title risks a binary conclusion that the body partly retreats from by endorsing interdisciplinary breadth. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. The proposed curriculum and employer judgments are normative arguments, not controlled evidence about graduate outcomes. Source note

32. “20 Percent Doctor Included” & Dr. Algorithm (2016)

Argument, reasoning, and evidence. The long paper predicts a progression in which algorithms assist, learn from, amplify, and ultimately perform most clinical tasks, leaving roughly 20% of the doctor role. It combines a v0–v7 capability ladder, a specialty/function matrix, consumer control of health data, and a claim that outsiders—under clinician oversight—will build more radical systems. Primary work; printed page and PDF page; current official linked PDF asset · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The thesis depends on interoperable data, robust prospective evidence, regulatory permission, reimbursement, trust, and safe escalation. Investors should map each product to a bounded task and version, demand outcome evidence before expanding autonomy, and distinguish clinical utility from a profession-level substitution forecast. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the most complete early health-automation statement, extending the 2012 and 2014 essays and enabling comparison with the 2019 deck and 2025 joint article. Its timetable and “20%” endpoint remain rhetorical, while the later work gives more attention to incentives and deployment architecture. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The mapped legacy URL is the canonical route; a current official CDN copy supplied the full recoverable paper when the legacy route failed. Vinod is the author, and company examples are thesis evidence from an interested investor rather than independent clinical validation. Source note

33. AI: Scary for the Right Reasons (2017)

Argument, reasoning, and evidence. Khosla redirects AI risk from speculative machine consciousness toward nearer threats: cyberattack, automated persuasion, concentrated state power, geopolitical competition, and labor displacement. Capability diffusion can create both social abundance and asymmetric misuse, so the essay favors attention to concrete attack surfaces and transition costs. Primary work; HTML paragraph and distinctive opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The analysis assumes near-term deployment and human misuse dominate tail-risk scenarios. Investors should threat-model abuse, data access, political capture, labor exposure, and dual-use pathways alongside product performance, while demanding explicit evidence for claims that openness or restriction will reduce net risk. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. It anticipates the geopolitical framing of the 2023 AI-war essay and the risk taxonomy of AI: Dystopia or Utopia?. The preference for broad innovation and access later conflicts with calls for export controls and strategic restriction; the corpus never fully resolves when openness helps defense more than offense. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. It is a foresight argument, not a security audit or empirical probability estimate, and any relevance to portfolio companies must be established separately. Source note

34. Reinventing Societal Infrastructure with Technology (2018)

Argument, reasoning, and evidence. Khosla presents technology as a resource multiplier across health, education, finance, food, transport, and public services. Progress comes from “instigators” who make an initially implausible system work, after which institutions and incumbents can scale it; technology is described as necessary but not sufficient for social change. Primary work; HTML paragraph and distinctive opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The framework assumes a technical wedge can alter entrenched cost or access curves and that complementary policy, capital, trust, and distribution will follow. Investors should identify the bottleneck being multiplied, define the institutional complements, and test whether a startup changes system economics rather than simply digitizing an incumbent workflow. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This broadens sector essays into a unified infrastructure thesis and foreshadows the “instigator” framing in the 2020 climate paper and AI-for-India essay. Its breadth supplies a search map, not evidence that any specific intervention works, and it under-specifies political economy and distribution. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay and associated document are accessible and attributed to Vinod. It contains conceptual and illustrative claims rather than deal evidence; no company should be credited with impact without independent outcome support. Source note

35. 20 Percent Doctor Included — Exponential Medicine Deck (2019)

Argument, reasoning, and evidence. The deck updates the health thesis as a dated v0–v7 roadmap, placing v0 in 2017 and v7 in 2036, and maps portfolio examples across a five-by-five specialty/function grid. It continues to argue that outsiders can assemble an algorithmic care stack and that clinical work will migrate from assistance toward autonomy. Primary work; PDF slide number · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The dated versions make the thesis auditable: assess which capabilities arrived, which produced validated outcomes, and which remained demos. Portfolio construction should separate model capability, regulatory clearance, provider adoption, reimbursement, and patient trust rather than treating one company’s launch as proof of the system forecast. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Compared with the 2016 paper, the deck is more portfolio-specific and gives later forecast clocks, but slide compression reduces caveats. Its examples show investment exposure, not clinical success, and the endpoint can move if each version is redefined after the fact. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The canonical legacy URL is retained while a current official CDN copy provided full slide access. The deck is attributed to Vinod; portfolio placement and company claims are interested evidence, not personal-deal attribution or independent medical verification. Source note

36. A Few Critical Climate Technology Breakthroughs Multiplied by “Instigators” (2020)

Argument, reasoning, and evidence. Khosla argues that roughly twelve to fifteen breakthrough technologies, driven initially by entrepreneurial “instigators,” can reach cost parity and then be scaled by incumbents. He separates proof around 2025 from widespread adoption nearer 2040 and names sector-specific technical and cost targets. Primary work; HTML numbered examples and distinctive opening phrase · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The model assumes a small Pareto set of innovations can cover most emissions and that cost parity unlocks financing, infrastructure, and adoption. Each opportunity should therefore have a dated technical milestone, unsubsidized cost target, deployment bottleneck, incumbent handoff path, and explicit failure condition. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This translates the black-swan doctrine into a scoreable climate roadmap and later underlies fusion, geothermal, and anti-green arguments. The selected examples create portfolio bias, and reaching a prototype milestone does not establish manufacturing scale, project finance, permitting, or emissions impact. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod, with a 2020-07-15 date. Named technologies and companies may overlap Khosla Ventures interests, so forecast scoring must use independent deployment and economics evidence. Source note

37. Stop Asking Me About the Markets (2022)

Argument, reasoning, and evidence. The recoverable record supports only an authored opinion’s title, byline context, and a brief official summary indicating that Khosla prefers company-building questions to short-term market forecasting. The publisher body is not available, so its reasoning, examples, and any qualifications cannot be reconstructed. Primary work; publisher byline/paywall notice and KV landing summary; no article locator available · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The visible premise suggests a process preference for controllable operating variables over macro timing. It is usable only as a bounded orientation; it cannot support claims about valuation discipline, cycle behavior, portfolio construction, or what Khosla would do under specific market conditions. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The stance is consistent with founder-assistance writing and outlier selection, yet other works discuss financing markets, capital intensity, and geopolitical policy. Without the full text, apparent tensions cannot be resolved and silence cannot be converted into a doctrine. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. This is a partial official mirror of a 2022-01-31 The Information authored opinion, not an interview. The canonical KV URL is preserved, and no unseen publisher content is paraphrased. Source note

38. AI and Climate Investors Can Help Western Values Win Out (2022)

Argument, reasoning, and evidence. Only the title and short visible summary are recoverable, supporting a bounded claim that investment in AI and climate technology is framed as strategically important to western values. The article body is unavailable, so the causal chain, policy prescriptions, examples, and caveats remain unknown. Primary work; KV title, author chips, date, and summary paragraph; FT canonical UUID landing/paywall boundary · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. Any investment implication is provisional: geopolitical relevance does not establish commercial quality, and values alignment requires an explicit definition. Diligence should separate technical advantage, market economics, national-security exposure, and normative claims rather than using the title as a thesis. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The visible framing bridges the 2017 AI-risk essay and the 2023 techno-economic-war argument, but body-level evolution cannot be assessed. It may also conflict with global-access goals; the missing text prevents determining how the authors balance openness and strategic restriction. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The KV route is partial. CMS attribution chips name Vinod Khosla and Richard Waters while the underlying Financial Times byline/body was not recovered, so sole authorship is not asserted and no unseen text is reconstructed. Source note

39. Vinod Khosla Wants More Crypto Regulation—Now (2022)

Argument, reasoning, and evidence. The accessible material is a one-paragraph post-FTX excerpt calling for stronger and faster SEC action. It establishes the direction of Khosla’s view but not a detailed regulatory architecture, allocation of jurisdiction, treatment of tokens, or reasoning beyond the immediate failure context. Primary work; KV excerpt paragraph 1 and “Read the full article”; The Information landing title/byline/paywall boundary · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The excerpt assumes more enforcement or rulemaking would reduce harm without specifying tradeoffs. It can prompt diligence on custody, conflicts, leverage, disclosure, and regulatory perimeter, but cannot support a comprehensive Khosla crypto framework. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. This is a narrow exception to an archive otherwise focused on technology creation and AI/climate policy. Because only an excerpt survives, it is impossible to assess whether the position changed with later market structure or how it fits his general preference for entrepreneurial freedom. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The official KV page is partial and derives from a The Information item. Vinod is the attributed speaker/authorial voice of the excerpt; missing publisher material is not inferred. Source note

40. What Is at Stake in This AI-Based Techno-Economic War Between the West and China? (2023)

Argument, reasoning, and evidence. Khosla frames AI leadership as a contest over economic capacity and political values, advocating skilled immigration, domestic innovation, strategic export controls, and safety work that does not concede capability leadership. Automation of much routine cognitive work supplies the economic premise; state competition supplies the urgency. Primary work; sequential HTML paragraphs; paragraphs beginning “The nation,” “While this techno-economic war,” “What can the West do,” and “Are the risks” · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The case assumes national AI leadership is coherent, restrictions slow rivals more than domestic firms, and innovation can advance alongside meaningful safeguards. Investment analysis should test supply-chain exposure, talent mobility, compute dependence, dual-use risk, and policy sensitivity instead of treating geopolitical importance as a substitute for product economics. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay develops the 2017 misuse thesis and the 2022 western-values framing into explicit policy. Its restriction agenda conflicts with universal-access and open-innovation rhetoric, while the “80% of 80%” labor claim remains a forecast rather than measured adoption. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. It is advocacy by an AI investor with economic exposure to the sector; national-security and labor claims require independent evidence. Source note

41. Where AI Meets Copyright Law, Present and Future (2023)

Argument, reasoning, and evidence. Khosla argues that copyright analysis should focus on output similarity and substitution rather than the mere fact that a model learned from protected works. He analogizes training to human learning, supports crawler opt-outs, and places liability for infringing outputs on the relevant tool provider or user depending on control. Primary work; headings “Copyright Infringement in AI Training” and “AI as an Author and User Liability”; sequential paragraphs within each heading · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The proposal assumes training and memorization can be distinguished, opt-outs are technically enforceable, and output tests preserve incentives without blocking general learning. Investors should diligence dataset provenance, memorization rates, rights controls, indemnities, output filters, and exposure by jurisdiction rather than relying on a broad fair-use analogy. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay adds a concrete legal-policy layer to Khosla’s pro-innovation AI agenda. Its balance between open training and creator control remains unsettled by courts and legislation, and the human-learning analogy may fail where models reproduce market-substituting content at scale. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. It is policy advocacy, not a judicial holding or legal advice, and the author’s AI investment interests are material context. Source note

42. OpenAI’s Board Set Back the Promise of Artificial Intelligence (2023)

Argument, reasoning, and evidence. The available two-paragraph excerpt argues that OpenAI’s board structure and effective-altruist influence endangered AI’s promise during the 2023 leadership crisis. It also presents Khosla as the company’s first venture investor, but the missing publisher body leaves the detailed governance case and evidence unavailable. Primary work; two exposed paragraphs beginning “I was the first” and “OpenAI’s board”; publisher link after excerpt · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The excerpt assumes the board’s actions reflected structural ideology rather than information unavailable to outside observers. The investable lesson is to map control rights, mission constraints, removal procedures, conflicts, and capital-provider remedies before a crisis; it is not to infer that investor control is inherently superior. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The criticism is in tension with Khosla’s founder-autonomy board doctrine and shows that mission structures can shift power away from both founders and investors. Because the board’s contemporaneous account and full article are absent, causal judgments must remain provisional. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The canonical KV route is a partial excerpt of an authored The Information opinion. Vinod’s economic relationship with OpenAI is a direct conflict, and the first-investor claim is self-report rather than transaction-record proof. Source note

43. How AI Will Change Our Relationship With Computers (2024)

Argument, reasoning, and evidence. The recoverable paragraph predicts that low-latency voice interaction will become the primary computing interface and that devices will recede into supporting roles. It establishes a directional product thesis but supplies no architecture, adoption data, economics, or boundary conditions. Primary work; sole exposed paragraph beginning “There’s a lot of buzz”; publisher link immediately after excerpt · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The view assumes speech systems become fast, reliable, private, and socially acceptable across contexts. Investors should test latency, task completion, error recovery, privacy, multilingual performance, and whether voice expands use or merely replaces an efficient screen workflow. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The prediction updates the 2004 post-phone device thesis for generative AI, shifting the interface from mobile applications toward conversational agents. A one-paragraph excerpt cannot show how Khosla handles environments where visual precision, discretion, or accessibility favors other modalities. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The official KV page is a partial mirror of a The Information authored opinion. Vinod is attributed, but no missing publisher text is inferred and no portfolio product is treated as validation. Source note

44. The US Is Right to Target TikTok (2024)

Argument, reasoning, and evidence. Khosla supports forced divestiture rather than a content ban, arguing that AI-mediated persuasion creates a national-security risk beyond conventional data privacy. Foreign control of a mass recommendation system is the central concern, with ownership change presented as a narrower remedy than prohibiting the service. Primary work; sequential HTML paragraphs; paragraphs beginning “The US House,” “This TikTok divestiture,” “Some argue,” and “Second, a forced divestiture” · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The case assumes control can translate into covert influence, divestiture can produce genuine operational independence, and the remedy survives legal scrutiny. Platform diligence should examine governance, algorithm access, data localization, update control, influence operations, and jurisdiction-specific remedies. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The article applies the techno-economic-war framework to a consumer platform and favors restriction despite Khosla’s broader openness rhetoric. Claims about state influence and legal necessity require independent evidence; divestiture also may not eliminate technical or informational dependencies. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official mirror of the 2024-04-09 Financial Times opinion is accessible and attributed to Vinod. It is advocacy, not a court finding or verified account of Chinese government conduct. Source note

45. The Generalist: Four Correspondences with Vinod Khosla (2024)

Argument, reasoning, and evidence. Across four edited correspondences, Khosla links Sun-era recruiting and repeated questioning to a “venture assistant” identity, founder tenacity, and willingness to pursue unpopular markets. The format develops operating principles through biographical episodes, including interactions with John Doerr and Mackenzie, rather than presenting a formal investment framework. Primary work; Part I–IV; “Mario’s letter,” “Vinod’s response,” subject, sender, recipient, and internal date · pinpoint evidence note [retrospective witness · moderate]

Assumptions and decision implications. The series assumes retrospective founder episodes reveal repeatable selection and assistance skill. Readers should extract falsifiable practices—reference depth, question quality, recruiting ownership, and response to setbacks—while keeping outcome stories separate from ex-ante evidence. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. It is a late synthesis of themes present since the business plan and 2012–15 operating essays. Winner selection, memory compression, and editorial construction can make the philosophy appear more coherent than it was at decision time; return anecdotes also lack transaction-level substantiation. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The partially accessible four-part series and its index are one underlying work, co-constructed by Mario Gabriele and Vinod Khosla. Editorial narration, Khosla’s direct voice, and third-party claims must remain distinct. Source note

46. Plausible Tomorrows (2024)

Argument, reasoning, and evidence. The evolving deck organizes fourteen forecasts across AI, health, climate, transport, robotics, and infrastructure as a search map for discontinuous change. It pairs each “plausible tomorrow” with technical direction and examples, encouraging portfolio breadth rather than confidence in one deterministic future. Primary work; landing paragraph and May-v2 PDF slide number · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The useful investment move is to translate each forecast into a prerequisite tree, dated milestones, counter-signals, and exposure limits. Plausibility alone is not probability; capital intensity, regulation, adoption, and value capture must be underwritten independently. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The work compresses decades of Khosla’s sector theses into a modern cross-domain roadmap and overlaps the TED predictions without being the same work. April and May versions evolved, but only the later recoverable deck was analyzed fully, so changes between versions cannot be characterized. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The official landing is partial as a version history even though the May v2 deck was accessible. Vinod is attributed; portfolio examples and forecasts are interested thesis material, not verified outcomes. Source note

47. How to Hire a CEO (2024-07-03)

Argument, reasoning, and evidence. Khosla proposes eight CEO qualities: founder alignment, a record of real value creation, skepticism toward polished search-firm packaging, technical-athlete learning, team building, sustained growth velocity, strategic synthesis, and fundraising capacity. The role is evaluated against the company’s next constraints rather than a universal executive archetype. Primary work; numbered headings 1–8 and unique paragraph openings · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The checklist assumes past contributions can be separated from institutional halo and that references reveal the candidate’s causal role. Boards should define the next value-creation state, score evidence behind each quality, use adversarial references, and precommit transition milestones and decision rights. Source-note evidence · open questions [researcher inference · insufficient public record]

Evolution and contradictions. The work updates the 2012 founder-versus-CEO essays by focusing on selection after the need for a CEO is established. Eight broad virtues can still invite confirmation bias, and the article offers no denominator comparing hires selected with or without the framework. Source-note contradiction and version ledger [researcher inference · insufficient public record]

Access, authorship, and conflicts. The current official landing is fully accessible and dated 2024-07-03; that date differs from older catalog metadata. Vinod is the attributed author; company and executive anecdotes remain participant evidence rather than verified performance attribution. Source note

48. How a California Bill Could Affect the US Tech Race With China (2024)

Argument, reasoning, and evidence. In the visible five-paragraph excerpt, Khosla opposes California’s SB 1047 and argues that frontier-AI safety should be handled federally because state rules could impair the US technology race with China. The excerpt establishes position and framing, but not the full statutory analysis or proposed alternative. Primary work; five exposed paragraphs beginning “A California bill,” “Senate Bill 1047,” “Commercial interests,” and “Because of national security”; publisher link after excerpt · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The position assumes the bill’s burden would materially reduce domestic capability and that federal action would be timely and better calibrated. Investors should model compliance scope, developer liability, compute thresholds, jurisdictional fragmentation, and geopolitical effects without treating competitiveness as dispositive of safety design. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The argument is consistent with the AI-war essay’s emphasis on leadership but presses against Khosla’s stated concern about near-term misuse. The partial text cannot reveal whether he distinguishes safety evaluation from liability or what evidence would justify stronger constraints. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The official page is a partial excerpt attributed to Vinod and dated 2024-09-10. It is policy advocacy by an AI investor; unseen material is not reconstructed and statutory claims require primary legal sources. Source note

49. AI: Dystopia or Utopia? (2024)

Argument, reasoning, and evidence. The long essay argues that AI will multiply access to expertise and produce abundance while also threatening jobs, political stability, security, and concentrated power. Khosla treats high venture failure as compatible with a few transformative successes and assigns policy the job of managing transition rather than stopping capability development. Primary work; numbered section and lettered subsection · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The thesis depends on rapid capability diffusion, falling service costs, sufficient demand elasticity, and workable redistribution and governance. Investors should model technical milestones, deployment complements, labor substitution versus creation, concentration, misuse, and policy response as separate uncertainties. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is the most comprehensive current synthesis of the 2014 abundance thesis, 2017 risk essay, and sector forecasts. Its repeated “80%” formulations remain weakly defined; the essay supports both broad access and strategic restriction, and it offers optimism about policy without an implementation denominator. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full roughly 13,150-word official essay is accessible and attributed to Vinod; a shorter summary is a derivative version, not another work. Portfolio examples and forward claims come from an investor with sector exposure and require independent scoring. Source note

50. Ride-Sharing Robo-Pods Are the Future of Cities (2024)

Argument, reasoning, and evidence. Khosla proposes small autonomous electric pods operating on dedicated lanes as a lower-energy, higher-throughput urban transport layer. He contrasts a claimed roughly 50 watt-hours per passenger-mile with 200–600 for alternatives and presents Glydways as an implementation pathway. Primary work; sequential HTML paragraphs; paragraphs beginning “Self-driving,” “The energy use,” “PRT pods,” and “A version of the pods” · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The model depends on public right-of-way, high utilization, station and lane economics, safety approval, network density, and credible energy comparisons. Investors should underwrite a corridor’s complete capital stack and counterfactual mode share, not only vehicle efficiency. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The essay updates his transport-abundance thesis from general automation to a concrete infrastructure design. Dedicated lanes can improve operations but create permitting and opportunity-cost barriers, while efficiency claims need common occupancy and system-boundary assumptions. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official essay is accessible and attributed to Vinod. He calls himself an investor in Glydways, but the text does not resolve the holding vehicle; company metrics and energy claims are not independently verified. Source note

51. How to Hire (2024)

Argument, reasoning, and evidence. Part I starts hiring from the company’s required state: value creation, value protection, strategy formation, or turnaround. Khosla then evaluates candidates against eight qualities and insists on evidence of actual contribution rather than title, using adversarial references and role-specific scorecards. Primary work; HTML Part I headings and numbered qualities; announced internal Parts II–III are outside this denominator work · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The process assumes the board can diagnose the state correctly and references can isolate causal contribution. A defensible search should write the value-creation mandate first, weight competencies before meeting candidates, sample disconfirming references, and track post-hire outcomes against the mandate. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The work consolidates the 2012 CEO essays and 2015 recruiting methods into a general selection architecture. Broad qualities can still be scored subjectively, and Parts II and III are internal/nonpublic, so the published work does not reveal the complete execution procedure. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The mapped denominator work is the fully accessible published Part I, attributed to Vinod. Announced internal Parts II and III remain access gaps outside this declared work rather than making Part I partial. Company examples are selected advisory experience rather than comparative hiring evidence. Source note

52. Venture Assistance Blog (2024 update)

Argument, reasoning, and evidence. The firm’s case-study compilation presents venture assistance as practical work across recruiting, pricing, sales, product, financing, and crises, including examples where founders accepted or rejected advice. It argues that investor value should be judged by company-building contribution rather than capital or brand alone. Primary work; named sections and company case-study headings · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The blog assumes selected founder accounts represent the wider portfolio and that advice can be causally linked to outcomes. A stronger assistance ledger would record the request, adviser, recommendation, founder decision, timing, cost, and measurable result—including cases where advice harmed or had no effect. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The compilation operationalizes the 2013 board philosophy but shifts from normative doctrine to marketing evidence. The inclusion of rejected advice is useful, yet selection bias and absent denominators prevent conclusions about frequency, causation, or whether the model travels beyond particular partners. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official firm post is accessible as a 2024 update, but it is institutional marketing rather than a solely authored Vinod essay. Its SVB personal-loan and operating-impact claims are self-reported and unaudited; trailing migrated code was excluded. Source note

53. A Roadmap to AI Utopia (2024)

Argument, reasoning, and evidence. Khosla describes a path from near-free expertise and automation toward broad abundance, while acknowledging labor disruption, concentration, security, and political instability. He proposes managing the transition through redistribution and public policy rather than slowing the underlying capability frontier. Primary work; sequential HTML paragraphs; opening steam-engine analogy; paragraphs beginning “For the next five to 10 years,” “Let’s continue this thought experiment,” and “Ultimately, the future” · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The roadmap assumes AI services become cheap and reliable, productivity gains pass through to users, demand expands, and governments can finance and administer transition support. Investors should model value capture, substitution, complements, distribution, and regulation separately from headline capability. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This is a more policy-oriented derivative of the 2024 utopia/dystopia thesis, with the familiar “80% of 80%” automation claim and greater emphasis on transition mechanisms. Deflation and abundance do not automatically produce equitable access, and the proposed social response lacks an implementation or fiscal denominator. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official 2024-11-11 essay is accessible and attributed to Vinod. It is a forward-looking policy argument by an AI investor; forecasts and portfolio examples require independent evidence. Source note

54. Is the World Ready for the Transformational Power of Fusion? (2025)

Argument, reasoning, and evidence. Bob Mumgaard and Vinod Khosla argue that advances at Commonwealth Fusion Systems can make commercial fusion plausible, pointing to SPARC operation targeted for 2027 and an early-2030s ARC plant planned around 400 megawatts. They link technical progress to abundant clean power and model large system savings. Primary work; author/affiliation block; opening scenario; headings “The race for the Wright Brothers' moment,” “From one power plant,” and “The fusion map of the world” · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The thesis depends on magnet and plasma performance, construction schedule, plant reliability, regulatory approval, supply chain, cost of capital, and repeatable project economics. Investors should gate physics proof, engineering integration, net-energy performance, plant cost, and bankability separately. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Fusion is a current instantiation of the 2011 black-swan and 2020 climate-breakthrough frameworks. A milestone announcement is not commercial validation, and modeled savings can be dominated by deployment timing, learning curves, grid alternatives, and financing assumptions. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full World Economic Forum article dated 2025-01-07 is jointly authored by CFS CEO Bob Mumgaard and Vinod Khosla. Both have promotional or economic interests in the company, so targets and savings are claims, not independently verified outcomes. Source note

55. The Opportunity for AI in Healthcare Is Here Now (2025)

Argument, reasoning, and evidence. Vinod Khosla and Neal Khosla argue that recent comparative benchmarks make AI-enabled care deployable now, with an “AI intern” progressing toward greater autonomy over five to seven years. They emphasize risk-bearing, value-based organizations as the deployment channel because those organizations can capture savings from better decisions. Primary work; Roman-numbered HTML section · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The case assumes benchmarks predict real clinical outcomes, workflow integration is safe, and incentive alignment overcomes provider resistance. Investors should demand prospective comparisons, calibrated escalation, liability allocation, cohort economics, reimbursement durability, and evidence that savings do not come from undertreatment. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The article updates the 2012–19 doctor-automation sequence with deployment incentives and a shorter near-term clock. Its company metrics, anonymized examples, and trillion-dollar opportunity estimate do not independently establish causation, generalization, or total addressable value. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official 2025-01-31 essay is jointly authored by Vinod Khosla and Neal Khosla; neither contribution should be silently assigned to the other. Portfolio and company claims come from interested parties and remain unverified. Source note

56. How the Anti-Green Agenda Could Help Climate Tech (2025)

Argument, reasoning, and evidence. The accessible Economist opinion excerpt argues that political hostility to “green” policy can force climate technologies to compete as industrial technologies, with China-and-India-relevant costs and offtake preceding subsidy exit. It invokes a portfolio of roughly a dozen breakthrough categories but does not expose the full argument. Primary work; KV excerpt paragraph and publisher link · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The visible thesis assumes policy withdrawal improves discipline without destroying the demonstration and infrastructure support needed for scale. Underwriting should separate temporary offtake, tax credits, regulated demand, learning-curve economics, and genuinely unsubsidized competitiveness. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. It reprises the 2011 and 2012 main-tech standard under a changed political frame. The argument can conflict with the capital and policy complements acknowledged elsewhere, and a short excerpt cannot establish which subsidies are catalytic versus permanently masking cost. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The canonical official page preserves a bounded excerpt of Vinod’s 2025-07-09 Economist op-ed, not an interview. His climate portfolio exposure is a direct conflict; unseen publisher text and company economics are not reconstructed. Source note

57. Are AI Valuations Bonkers? (2025)

Argument, reasoning, and evidence. The short excerpt concedes that most AI valuations are “bonkers” while arguing that a small number of outliers can justify selective high prices and cover many losses. It is a power-law defense, not a disclosed valuation method. Primary work; four exposed paragraphs beginning “To many investors,” “The lower valuations,” “Why did,” and “Increased valuations”; publisher link after excerpt · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The view assumes an investor can identify exceptional companies with enough edge and ownership for outlier outcomes to dominate. A usable decision rule still needs entry price, dilution, reserve strategy, probability-weighted exits, ownership, and fund-return contribution rather than category enthusiasm. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The excerpt applies the black-swan portfolio doctrine to an expensive consensus market, but it does not reconcile paying up with base rates or show whether prior success creates an institutional feedback loop. Without criteria or realized economics, selectivity is asserted rather than demonstrated. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The official KV page is a partial excerpt of Vinod’s 2025-09-10 The Information opinion. As an active AI investor discussing valuations, he has an economic conflict; no unseen article detail or return result is inferred. Source note

58. Mazama Energy Unveils the World’s Hottest Enhanced Geothermal System (2025)

Argument, reasoning, and evidence. The announcement reports a 629°F/331°C enhanced-geothermal result at 10,200 feet and presents high-temperature rock connectivity as a path toward lower-cost, scalable power. It pairs the technical result with company claims of tenfold improvement, major drilling-cost reductions, sub-five-cent power, and a 15-to-200-megawatt scale path. Primary work; paragraphs beginning “Almost all geothermal startups,” “Mazama Energy,” and “During the technical demonstration”; headings “A New Era for Data Center Energy,” “Thermal Lattice,” and “SuperHot 750 °F” · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The economics depend on repeatable reservoir creation, flow rate, decline curve, drilling learning, water and induced-seismicity management, permitting, and project finance. Investors should treat temperature as one milestone and require sustained flow, full-system cost, and replicated-site evidence before extrapolating. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. Mazama is a direct 2025 test of the climate-instigator framework: a startup pursues an extreme technical wedge intended for incumbent-scale energy markets. The announcement does not verify the projected cost curve or commercial scale, and demand from AI does not remove resource or financing risk. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official 2025-10-28 post is attributed to Vinod but functions as a Khosla Ventures portfolio-company announcement. KV says it incubated Mazama, making every company performance and cost claim promotional until independently corroborated. Source note

59. AI as an Enabler of Universal Access and Equity for India (2026)

Argument, reasoning, and evidence. Khosla argues that regional-language AI can widen access to education, health, law, government services, and entertainment in India, potentially through a personal Aadhaar-linked portal. The essay combines a 2030–35 capability roadmap with constraints around labor transition, compute, energy, institutional trust, and deployment. Primary work; H2 heading and labeled service paragraph · pinpoint evidence note [investor-stated · strong]

Assumptions and decision implications. The opportunity depends on multilingual quality, identity and consent safeguards, connectivity, service integration, grievance mechanisms, and public legitimacy. Investors should evaluate outcome equity, privacy, fraud, exclusion errors, unit economics, and accountability rather than using user reach as the impact metric. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. This localizes the abundance thesis and makes complementary infrastructure more explicit than earlier automation essays. Aadhaar integration can lower friction while increasing surveillance and exclusion risk, and universal access conflicts with geopolitical restriction unless data and governance boundaries are credible. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official HTML essay dated 2026-01-20 is accessible and attributed to Vinod; no separate document or media asset exists. Named service categories are forecasts, not deployed-outcome evidence. Source note

60. When SVB Fell, “Founder-Friendly” Was Put on Trial (2026 retrospective; 2023 embedded posts)

Argument, reasoning, and evidence. The 2026 retrospective uses preserved March 2023 posts to argue that investor alignment during Silicon Valley Bank’s failure required direct founder support, including an expressed willingness to make personal loans, while keeping LP capital separate. It also stresses contacting founders, avoiding blame, supporting government action, and signaling stability. Primary work; H2 heading and embedded status IDs 1634737634907127810, 1634742543534673920, 1634740918841651202, 1634740457069760513, 1634739215857782784, 1634391256028839936 · pinpoint evidence note [retrospective witness · moderate]

Assumptions and decision implications. The account assumes public commitments reduced panic and that personal assistance could be delivered quickly and fairly. A crisis protocol should predefine communication ownership, liquidity triage, legal and conflict review, the boundary between personal and fund capital, and an auditable record of offers and completed support. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The episode is presented as lived proof of venture assistance, but intention, outreach, loan execution, terms, and founder outcomes are different claims. The 2023 posts are contemporaneous evidence of statements; the 2026 narrative is retrospective interpretation and cannot independently verify wires or loans. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full official 2026-03-10 essay preserves embedded 2023 post text and dates. Vinod is the author of most posts; the joint stability statement has separate authorship, and neither the page nor social posts verify every claimed transaction. Source note

61. Venture Assistance: Entrepreneur Q&A Session (2026)

Argument, reasoning, and evidence. In a published anonymized Q&A excerpt, Khosla presses founders to express contribution margin in dollars, measure CAC payback and sales cohorts, manage runway and burn, and define competition from the customer’s perspective. The method turns strategy conversation into a sequence of falsifiable operating questions. Primary work; speaker label and unique opening phrase · pinpoint evidence note [investor-stated · moderate]

Assumptions and decision implications. The questioning assumes the disclosed metrics are comparable and that cohort behavior predicts scaling. Investors should reconcile reported definitions to source data, stress-test retention and working capital, distinguish gross margin from contribution margin, and record how answers change the next financing or operating decision. Source-note evidence · open questions [researcher inference · moderate]

Evolution and contradictions. The exchange is the most current direct example of the venture-assistance style described in 2013 and marketed in 2024. One edited session cannot establish the quality or frequency of assistance across the portfolio, and anonymization prevents outcome verification. Source-note contradiction and version ledger [researcher inference · moderate]

Access, authorship, and conflicts. The full published excerpt dated 2026-04-09 is accessible, but it is not the complete meeting; speaker labels and unique opening phrases provide locators. The company and other participants are anonymized, so claims and effects cannot be independently matched. Source note

62. Will Autonomous AI Exceed AI-Aided Physicians as the Best Medical Care? (2026)

Argument, reasoning, and evidence. The accessible opening challenges the emerging consensus that AI should remain only an aid to physicians and asks whether autonomous systems may eventually deliver better care. It marks a sharper current claim than simple augmentation, but the body’s evidence, qualifications, recommendations, and disclosures are unavailable. Primary work; publisher metadata; author/affiliation block; opening paragraph beginning “The prevailing view” · pinpoint evidence note [investor-stated · provisional]

Assumptions and decision implications. The question turns on comparative patient outcomes, calibrated uncertainty, escalation, accountability, distribution shift, equity, and cost. Investors should require prospective head-to-head evaluation and a defined operating boundary before using autonomy as either a product claim or valuation premium. Source-note evidence · open questions [researcher inference · provisional]

Evolution and contradictions. The Perspective brings the 2012–19 automation thesis into a 2026 clinical-policy forum, now framed explicitly against AI-aided physicians. Because only the abstract opening was recovered, it is not possible to determine how the authors revised earlier timelines or handled contrary safety evidence. Source-note contradiction and version ledger [researcher inference · provisional]

Access, authorship, and conflicts. The partial JAMA abstract page is dated 2026-08-17 and jointly authored by Ezekiel Emanuel, Abe Baker-Butler, Neal Khosla, and Vinod Khosla. A Perspective is not a clinical trial, and the missing body and disclosure section are not reconstructed. Source note

Cross-corpus evolution

The stable architecture is more important than any sector prediction: set a large endpoint, decompose it into independent risks, buy information with the cheapest discriminating experiment, reserve follow-on capital for evidence, and help the founder recruit or decide. What changes over time is the constraint set. Early writing emphasizes technical architecture and market creation; the 2010s add portfolio math and organizational assistance; current work increasingly admits regulation, trust, energy, identity, geopolitics, and distribution as coequal adoption risks. Model registry and evidence boundaries [researcher inference · strong]

That evolution is real but not linear. The corpus repeatedly returns to the same power-law justification, often without the transaction-level denominator needed to test it. The best reading is therefore a decision system with scoreable predictions—not a verified track record embedded in prose. Investment-economics boundary [researcher inference · strong]

Contradictions and counterevidence

• Failure tolerance versus capital discipline. Intelligent failure celebrates high failure rates; the risk matrix requires bounded, information-producing experiments. A loss is consistent with the doctrine only if the hypothesis, budget, milestone, and stopping rule were recorded ex ante. [researcher inference · strong]
• Founder autonomy versus board responsibility. The board essay leaves operations to founders, while the OpenAI excerpt criticizes a board exercising mission-based control. Neither source supplies a complete account of formal board authority, governing law, corporate or financing documents, information rights, or crisis governance. [researcher inference · strong]
• Unsubsidized economics versus enabling policy. The Big Green Opportunity and the anti-green excerpt demand main-tech competitiveness, yet fusion, geothermal, and infrastructure deployment can depend on public research, permitting, offtake, credits, and financing. The relevant distinction is catalytic support versus permanent economic dependence. [researcher inference · strong]
• Open access versus strategic restriction. AI: Scary for the Right Reasons favors attention to misuse while the AI-war essay and TikTok opinion endorse restrictions; AI for India seeks universal access. The corpus does not supply a rule for when diffusion or control produces the safer outcome. [researcher inference · moderate]
• Abundance versus distribution. The 2014 technology-revolution essay predicts abundance and labor devaluation simultaneously. The AI roadmap assigns the gap to transition policy, but offers no fiscal, administrative, or political implementation test. [researcher inference · strong]
• Autonomy forecasts versus clinical evidence. The 2016 doctor paper and 2026 JAMA opening move toward autonomous care. Benchmarks and capability ladders are not substitutes for prospective patient outcomes, liability design, distribution-shift testing, and escalation performance. [researcher inference · strong]
• Personal voice versus institutional results. Many essays are Vinod’s; many examples, tools, and outcomes are Khosla Ventures, partner, adviser, or portfolio-company claims. Authorship, board service, voting control, entity ownership, personal capital, sourcing, and fund return are different facts and should never be collapsed. Attribution controls [researcher inference · strong]
• Forecast revision versus moving goalposts. The dated climate and doctor ladders are valuable because they can be scored. Later versions should be compared against the original definitions; relabeling a capability stage after the date passes would convert a falsifiable forecast into narrative. Spoken forecast scorecard [researcher inference · strong]

Decision playbook

1. Define the non-incremental endpoint. Use the black-swan tests: material scale, durable advantage, and economics that survive without a protected niche. [researcher inference · strong]
2. Name the five ways the thesis dies. Convert technical, market, team, regulatory, financing, and adoption uncertainty into a risk matrix, including dependencies rather than treating scores as independent. [researcher inference · strong]
3. Buy information before scale. For every risk, specify the cheapest experiment, success threshold, time, capital, and what decision follows. Call a failure “intelligent” only when it changes the posterior or stops waste. Intelligent Failure, opening and “90-percent” paragraphs [researcher inference · strong]
4. Stage capital and preserve abandonment. Reserve follow-on funding for evidence, not enthusiasm; model dilution, ownership, project finance, and the option value of waiting. Power-law outcomes do not excuse an undefined entry price. Vehicle and reserve evidence boundary [researcher inference · strong]
5. Engineer the team against the risk map. Apply gene-pool engineering: recruit accountable expertise for both fatal risks and transformative upside, then verify actual contribution through adversarial references. [researcher inference · strong]
6. Make assistance observable. Under the board philosophy, log who asked, who advised, alternatives considered, founder choice, and measured outcome. Preserve founder agency while documenting legal duties and conflicts. [researcher inference · strong]
7. Underwrite the deployment system. Health, climate, transport, and AI require regulation, reimbursement or offtake, infrastructure, trust, and distribution. A technical milestone is one node in the prerequisite tree, not evidence of commercial completion. Societal Infrastructure, technology “necessary, though not sufficient” [researcher inference · strong]
8. Keep two audit ledgers. One scores dated forecasts against unchanged definitions; the other separates self-report, company claim, third-party evidence, filing, and audited economics. This prevents retrospective coherence and attribution drift. Source-map evidence controls [researcher inference · strong]

Five most useful works

1. Sun Microsystems Business Plan. The highest-value artifact because it is contemporaneous with company formation and exposes actual assumptions about product architecture, ecosystem, distribution, competition, organization, and capital—not a later success story. Use when evaluating whether a deep-tech founding plan joins invention to a credible operating system. [researcher inference · strong]
2. Black Swans Thesis of Energy Transformation. The clearest statement of Khosla’s selection logic: extreme technical upside, high expected failure, scale economics, and a portfolio broad enough to make outliers possible. Use when designing a power-law search strategy, but treat it as a hypothesis pending denominator and return evidence. [researcher inference · strong]
3. The Risk Matrix. The most transferable operating tool because it converts rhetoric about audacity into named failure modes, evidence milestones, sequencing, and capital requirements. Use when converting a visionary pitch into an experiment and reserve-capital plan. [researcher inference · strong]
4. Venture Assistance: What Boards Should and Should Not Do. The best primary statement of the claimed investor–founder contract. Use when setting board expectations or testing actual conduct against the promised balance of assistance and founder agency. [researcher inference · strong]
5. AI: Dystopia or Utopia?. The most complete current cross-sector thesis, integrating abundance, labor, security, concentration, and policy. Use when building a modern Khosla forecast ledger, not as proof that the predicted transitions will occur. [researcher inference · moderate]

Open gaps

• Investment economics: no complete transaction ledger, audited deal-level cash flows, ownership-through-time tables, reserve history, write-off denominator, or attribution of sourcing and board work across Vinod, partners, entities, trusts, and funds. Retrieval ledger [researcher inference · strong]
• Fund evidence: no authored LP letters, personal newsletter, book or memoir, or accessible internal investment-committee memoranda surfaced. Public fund snapshots and company filings can triangulate vehicles, but do not reveal the full decision record. Exhausted public barriers [researcher inference · strong]
• Missing and partial texts: The Innovator’s Ecosystem needs a recoverable artifact. Full publisher bodies are still needed for the market, crypto, FT climate, OpenAI-board, computer-interface, SB 1047, valuation, and JAMA works; internal Parts II–III of How to Hire also remain unavailable. Written-work access ledger [researcher inference · strong]
• Advice denominator: the archive needs a complete ledger of assistance offered, accepted, rejected, and later evaluated—including adverse cases—to test whether the board philosophy is partner-independent and causally useful. Attribution controls [researcher inference · strong]
• Forecast audit: the 2016/2019 doctor ladders, 2020 climate targets, 2024 plausible tomorrows, and 2025–26 AI clocks should be scored against contemporaneous definitions and external outcome data, including failed or delayed predictions. Spoken forecast scorecard [researcher inference · strong]
• Governance and legal counterevidence: the writing corpus should be read beside primary records on KiOR, View, Nextdoor, SPAC incentives, and Martins Beach so that philosophy is not mistaken for observed conduct or legal resolution. Failure and legal records [researcher inference · strong]
• Authorship and version control: institutional tools need creator attribution; joint works need contribution boundaries; revised decks and migrated pages need version diffs. A current official mirror can preserve access without becoming an independent source or changing the original clock. Source-map date/version controls [researcher inference · strong]

Coverage and evidence base

This synthesis covers 193 distinct spoken-work identities: 128 ★★★ and 65 ★★, each represented by one deep note. Work-level access is 149 full / 42 partial / 2 unavailable. The current broader research map contains 413 annotated routes—17 biography/archive, 68 writings/decks, 228 talks/interviews/testimony, 73 firm/deal/vehicle/ownership, and 27 failure/legal/context—with route-level priority 288 ★★★ / 118 ★★ / 7 ★ and access 296 full / 114 partial / 3 unavailable. The spoken denominator collapses 228 talk routes into 193 works because 35 landing, transcript, caption, video, or multipart paths are access routes to an existing appearance. Contemporaneous record · strong for the mechanical inventory; researcher inference · strong for work-family boundaries. Annotated source map, inventory and access

The identity rule counts a substantive talk, interview, hearing appearance, podcast, or sufficiently detailed contemporaneous host report once. Mirrors, clips, replays, landing/transcript/video pairs, and multipart installments do not create new works; the five surviving Stanford clip pages are fragments of one 2002 appearance. A work inherits the best legitimate mapped asset, so a partial landing paired with a complete captioned video can be full. An untranscribed player is not full merely because it plays. The frozen public-search boundary includes all 368 official KV /posts/ pages, 108 linked YouTube identities, official/government archives, and external venue/title/date variants through 2026-08-22; it does not imply that private, deleted, paywalled, or unindexed appearances do not exist. Researcher inference · strong. Coverage controls and exclusions

Evidence quality is two-dimensional. “Investor-stated” records Khosla's stated belief or memory, not whether it was correct; “documented behavior” requires a record of action; “contemporaneous record” freezes what was knowable at the time; and “researcher inference” marks synthesis. Strength then reflects source proximity and completeness. Exact quotations are restricted to authenticated transcripts, uploaded subtitles, or audio-checked passages; automatic captions alone are navigation evidence, not quote-grade. Every ledger row links the full note containing the claim-local timestamp, page, speaker block, paragraph, or explicit no-body boundary. Repetition across appearances establishes consistency, never truth or economic performance.

Attribution remains narrow throughout. “We,” a KV-branded stage, firm portfolio membership, or a fund-controlled security does not establish Vinod-only sourcing, diligence, approval, board work, capital, ownership, carry, or proceeds. Partner, trust, personal-capital, fund, sponsor, and incubator actions remain separate unless a filing or named participant joins them. Host introductions and edited summaries are not silently converted into Khosla testimony. Retrospective memories—especially Sun, KPCB, Juniper, Cerent, cleantech, Impossible Foods, Square, and OpenAI—are useful witness accounts but require contemporaneous records and counterparties. Chronology and attribution scaffold · written-corpus synthesis

Dated appearance and access ledger

This ledger is also the source-by-source guide: its final column states the primary analytical use and preserves the note's two-axis evidence label plus its most specific public locator. Dates retain recording, event, publication, upload, and migration clocks when they differ. “Full,” “partial,” and “unavailable” are work-level states under the rule above, not judgments about whether every private underlying artifact exists.