David Sze

1. Snapshot and testable edge thesis

The evidence supports a decision vocabulary, not audited individual alpha. Facebook is the only defensibly dominant disclosed value marker, and even it lacks complete fund cash flows. Economic ranking [researcher inference · strong]

2. Core philosophy

Sze's earliest surviving memo asks whether professional identity, trusted connections, user quality, and useful workflows can support a network before revenue is mature. LinkedIn reflection, Business and Path Forward In later direct speech he says venture should listen bottoms-up to entrepreneurs, search beneath obvious objections for the positive mechanism, and tolerate informed disagreement without mistaking unanimity for truth. TechCrunch transcript, sourcing and valuation turns Forbes, Facebook/Baidu paragraphs Semil Shah transcript, Parts II–IV [investor-stated · strong]

The synthesis is conditional: discover through people, underwrite through product behavior, name the exact graph and bootstrap constraint, separate outlier potential from entry price, and use the partnership/board as a support and error-correction system. [researcher inference · strong] A network label, famous founder, contested vote, or later valuation is never sufficient by itself.

3. Sourcing

Sze said Greylock spent roughly 80–90% of its time listening to entrepreneurs and 10–20% on trend work because consensus themes are often crowded. TechCrunch transcript, closing sourcing turn [investor-stated · strong] The exact percentages are one interview's claim, not audited time allocation, but the procedure is clear: begin with anomalous founder/product evidence. [researcher inference · strong]

Relationships repeatedly opened doors: Hoffman at LinkedIn, known founders at Nextdoor, Citron at Discord, Williams and Elman at Medium, and years of observation around Roblox. LinkedIn reflection, People Nextdoor essay, founder history Discord essay, Citron paragraphs Medium essay, founder paragraph Roblox essay, science-fair paragraph [documented behavior · strong]

Repeatability is bounded. Greylock's West Coast platform, brand, partners, and founder alumni are institutional access advantages. SFGate, West Coast transition Fanbase shows familiarity did not force investment. A modern IC should record who introduced, sourced, advocated, diligenced, approved, and governed; a relationship is a diligence channel, not a thesis. [researcher inference · strong]

4. Picking

Founder and people evidence

Sze's deal writings emphasize domain history, product sense, persistence, intellectual honesty, and the ability to learn through pivots. LinkedIn reflection, People Roblox memo, Team [investor-stated · strong] The Duboe essay extends the test to investors: zero-to-one and scaled operating proof, quantitative rigor, people judgment, and teaching ability. Greylock, Mike Duboe criteria [investor-stated · strong]

Product, market, and diligence

The recurring product test is: who are the nodes, what relationship connects them, what repeated job creates native value, what density unlocks it, and why is a general network insufficient? LinkedIn, Nextdoor, Discord, and Roblox each supply a distinct topology. Writings, graph synthesis [researcher inference · strong]

Roblox supplies the clearest surviving diligence artifact: user interviews, expert calls, management meetings, a financial model, a base case, and an outlier case. Roblox memo, Summary The published summary omits the complete vote, security, actual funded amount, ownership, full downside, and safety work, so it is a model to interrogate, not copy blindly. [researcher inference · strong]

Red flags and decision rights

Kill criteria should include weak retained utility, artificial growth, unsafe or low-trust interaction, a graph that never reaches density, non-native monetization, creator/supply concentration, adverse regulation, and a price that forecloses later financing. [researcher inference · strong] These criteria synthesize filings and failures rather than claim to quote Sze. Roblox S-1, Risk Factors Nextdoor prospectus, Risk Factors Digg case

Sze describes a partnership in which one strong advocate plus relevant domain support can coexist with objections. No vote table or full outcome sample verifies predictive efficacy. Semil Shah transcript, Parts II–IV [investor-stated · moderate]

5. Deal and portfolio mechanics

Sze described checks from about $200,000 to more than $10 million and a two-stage price rule: first test whether the company can become exceptional, then test whether entry price preserves return and future financing flexibility. TechCrunch transcript, check-size and valuation turns [investor-stated · strong] Fund XIV's strategy paired early-stage concentration with follow-on capacity and selective growth rounds. Fund essay, stage/reserve paragraphs [investor-stated · strong]

Actual ownership discipline is only partly visible. LinkedIn, Facebook, Pandora, and Nextdoor filings show affiliated-vehicle snapshots, not entry-to-exit dilution or realized cash flows. LinkedIn S-1/A, printed pp. 115–117 Facebook S-1/A, printed pp. 144–145 Pandora S-1/A, printed pp. 112–113 Nextdoor 424B3, Principal Securityholders [contemporaneous record · strong]

The defensible practice is to model weak/base/outlier cases, target ownership, dilution, reserves, and next-round flexibility separately. Exit evidence is too thin for a general Sze sell discipline; an IPO is described as a company-building milestone, not a complete fund-return event. Forbes, Facebook IPO paragraphs [researcher inference · moderate]

6. Vehicle and incentive fit

Treehouse and Lumi Story AI are separately named personal-angel relationships. Their entity, check, security, ownership, follow-on, and proceeds remain unknown and must not be folded into Greylock economics. Treehouse founder history Latham, Lumi financing [researcher inference · strong]

7. Team attribution and work with founders

Sze's founder-service philosophy combines perspective on difficult standards with humility that the founder owns the journey. TechCrunch, Fund XIV founder-support paragraphs [investor-stated · moderate] Recruiting and other services belong to named operators such as Dan Portillo, not automatically to the visible partner. Portillo essay, role/output paragraphs [documented behavior · strong]

8. Views on the craft

• Power law: The job is to seek an outlier mechanism, but public Sze evidence does not quantify a hit-rate or loss-ratio model. Roblox's base/outlier cases are the clearest artifact. Roblox memo [researcher inference · moderate]
• Fund size: Early-stage focus and follow-on capacity can coexist; announcement size is not performance. Fund essay [investor-stated · strong]
• Access versus selection: Founder relationships improve access; Fanbase proves selection still matters. Nextdoor transcript [documented behavior · strong]
• Governance: Board work should be permissioned and stage-specific; detailed intervention outcomes remain sparse. Professional CEO essay [researcher inference · moderate]
• Exits: Listing is an operating milestone; no public general sell discipline exists. Forbes IPO interview [investor-stated · moderate]
• Great investor: Listen, find the positive mechanism below consensus objections, preserve independent advocacy, and remain corrigible after misses. Talks, recurring frameworks [researcher inference · strong]

9. Signature bets and decision process

Use the no-hindsight labs rather than outcome-first summaries:

1. LinkedIn 2004: invest before revenue; test professional identity, utility, founder, scale, and monetization.
2. Facebook 2006: separate user utility, price, observer rights, and co-investor/founder credit.
3. Pandora 2009: underwrite royalty survival and board role before seeing the IPO.
4. Nextdoor / Fanbase: explain exactly what mechanism changed.
5. Discord 2016: test organic behavior while preserving Elman/Sze attribution ambiguity.
6. Roblox 2018: rebuild base/outlier cases, creator economics, safety, and terms without treating the published summary as complete.
7. Digg, Path, Twitter/Medium, and Airbnb: keep failures, passes, and dissent inside the curriculum.

Each lab asks invest/pass, kill criteria, ownership/terms, and next-check evidence before revealing the actual decision and outcome. [researcher inference · strong]

10. Mental models and maxims

The reconciled field guide contains 13 decision rules:

1. Listen bottoms-up before naming the market.
2. Define the graph, native job, and bootstrap topology.
3. Sequence utility and habit before monetization.
4. Build the positive case one level below the objection.
5. Underwrite people through domain learning and intellectual honesty.
6. Use relationships as diligence channels, not theses.
7. Preserve informed dissent with named ownership.
8. Separate outlier potential from price and financing optionality.
9. Size from weak, base, and outlier cases with reserves.
10. Follow cost collapse to the new bottleneck.
11. Make founder service bounded, permissioned, and measurable.
12. Treat CEO succession as a role-system decision.
13. Keep an anti-portfolio and reopen only on changed evidence.

Creator-economy underwriting is merged into the graph/topology model because Roblox is the dominant direct case; pivot logic is merged into anti-portfolio/reopen because only Nextdoor fully documents pass-then-reopen; institutional specialization is handled in founder service and vehicle design rather than inflated into an investment rule. Mental-model inventory boundary [researcher inference · strong]

11. Evolution and contradictions

Founder conviction coexists with possible professional-CEO succession; price can become irrelevant for a great company yet remains unknowable ex ante; controversy can expose non-consensus insight but can also reflect a bad deal. These are productive tensions, not contradictions to average away. [researcher inference · strong]

12. Failures and limits

The bounded record includes Digg, Path, Vessel, Treehouse, the Twitter pass, Airbnb dissent, Fanbase pass, and unnamed early enterprise losses. Failure table It does not disclose enough board actions, write-offs, reserves, or postmortems to prove a consistent failure-response loop. [researcher inference · strong]

Selection bias is severe: public firm pages emphasize recognizable companies and acquisitions, even when terms are weak or undisclosed. The 33 relationships are not a win-rate denominator, and an acquisition is not automatically a profitable outcome. Portfolio pattern tests [researcher inference · strong]

Era dependence is also material. Cheap web distribution, early social graphs, mobile adoption, and less mature platform/privacy regulation shaped the visible wins. Current underwriting must add safety, privacy, platform dependency, creator labor, AI-content provenance, and exit/liquidity gates instead of copying the historical playbook. [researcher inference · strong]

13. Historical operating environment

• 2000–2003: technology-market contraction and weak West Coast incumbency formed the entry context; no vintage Greylock base rate is public. FRED SFGate [researcher inference · moderate]
• 2004–2008: LinkedIn had no revenue at the 2004 decision; Facebook reported seven million users in 2006. Later social-media adoption data cannot be leaked backward. LinkedIn Meta [researcher inference · strong]
• 2009–2013: Pandora faced royalty-policy risk; larger Greylock funds increased follow-on capacity. CRB Fund XIV [researcher inference · strong]
• 2014–2018: mobile, gaming communication, user-generated content, and creator marketplaces expanded possible graph types. Discord Roblox [researcher inference · strong]
• 2021 onward: public filings make moderation, privacy, retention, governance, and platform risks more visible; Inflection is a joint Hoffman/Sze AI-incubation relationship whose security and economics remain undisclosed. Nextdoor prospectus Greylock portfolio, Inflection [researcher inference · moderate]

These are information-set controls, not numerical base rates. [researcher inference · strong]

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

Current grounded controls include Nextdoor's 2025 filing for public-company retention/moderation risk, Roblox's issuer disclosure for creator/safety/platform risk, and Lumi Story AI's 2024 angel financing for strict personal-versus-firm attribution. Nextdoor 2025 10-K, Overview and Risk Factors Roblox S-1 Lumi financing All modern transfers are [researcher inference · moderate] unless otherwise labeled; they are not claims about Sze's current practice.

15. Comparison with completed peers

Underlying controls: Breyer's prepared-mind and team evidence is in his Stanford investing note and HBS note; Wilson's public-thesis and vehicle evidence is in his USV formation note and reserves note; Gurley's market-mechanics and risk evidence is in his market-structure note and LTV note. These are comparison evidence, not Sze evidence. [researcher inference · strong]

This is [researcher inference · moderate], not a return or greatness ranking. The peer dossiers are complete, but their underlying evidence and vehicles remain dimension-specific; no peer's economics are imported into Sze's record.

16. The strongest case against greatness

The skeptical thesis is that a prestigious partnership obtained privileged access during a historically favorable social/mobile regime, one disclosed outlier—Facebook—dominated visible value, and later storytelling selected LinkedIn, Facebook, Nextdoor, Discord, and Roblox while suppressing ordinary deals and losses. The denominator is incomplete, economics are not audited, and several major artifacts are retrospective or promotional. Economic contribution Evidence boundary [researcher inference · strong]

Team misattribution reinforces the case: Bhusri helped build the office; Hoffman founded LinkedIn and sponsored Airbnb; Elman has reported Discord/musical.ly lead credit; Rosenberg coauthored Roblox; specialists supplied services; founders built companies; Greylock vehicles held securities. Palo Alto Networks belongs to Asheem Chandna, not Sze. Responsibility matrix PANW proxy, pp. 50–51 [researcher inference · strong]

The favorable thesis retains three discriminating facts: an ex-ante LinkedIn memo, a documented Fanbase-to-Nextdoor decision reversal, and a record that preserves Digg/Twitter/Airbnb error. These support a genuine process, but not its economic efficacy. [researcher inference · strong] Complete memos/votes, a career-wide attributed denominator, partner labor records, vehicle cash flows, and results excluding Facebook would decide between the views.

17. Analyst study guide, glossary, and go deeper

Recommended path

1. Read the biography through the edge falsifiers.
2. Read the 12-writing record, beginning with LinkedIn and the fund strategy.
3. Complete the LinkedIn, Roblox, Digg, Twitter/Medium, and Airbnb no-hindsight exercises.
4. Use the mental-model field guide to write an IC memo, then test it against the 33/42 ledger.

Glossary

Reusable checklists

Sourcing

• What founder/product anomaly preceded the category label?
• Who introduced, sourced, and owns the relationship?
• What access advantage is institutional, and could an outside analyst verify the thesis?

Diligence and picking

• Name nodes, relationship, native job, density threshold, and bootstrap route.
• Freeze known, unknown, believed, disputed, base rates, regulation, and alternatives.
• Write the positive mechanism, kill criteria, abuse/trust surface, and people evidence.

IC and portfolio

• Name advocate, dissent owner, decision rights, and attribution.
• Separate weak/base/outlier company cases from price, ownership, dilution, reserves, and fund contribution.
• State next-check milestones and what evidence reopens a pass.

Governance and exit

• Obtain founder permission; name intervention owner and outcome metric.
• Separate board director, observer, specialist, and investor roles.
• Treat listing, marked value, realized sale, fund distribution, carry, and personal wealth as different events.

Go deeper

1. LinkedIn reflection — best ex-ante memo evidence plus explicit forecast correction.
2. Semil Shah transcript — best biography, early failure, Pandora, and IC-process evidence.
3. TechCrunch 2010 transcript — best contemporaneous sourcing, price, stage, and Digg control.
4. Nextdoor transcript — best pass/pivot, bootstrap, governance, and moderation sequence.
5. Roblox memo — best public diligence/sizing artifact, with joint-authorship and publication-selection caveats.
6. Issuer and vehicle evidence — necessary control on ownership, attribution, risk, and denominator.
7. Source map — complete public retrieval boundary and unresolved archive routes.

Pareto 80/20 — what drove the record

The public record does not disclose Greylock's deal cash flows, fund-by-fund attribution, dilution, distributions, or Sze's carry. The “80/20” lens therefore prioritizes research; it is not a literal return split. Economic ranks below are directional and separate company value or a security snapshot from realized fund return and personal wealth. [researcher inference · strong]

View 1 — economic contribution

Record without Facebook. LinkedIn, Pandora, and Roblox remain plausible material contributors, but their disclosed measures are incommensurable: a filing holding plus acquisition outcome, a preferred-share purchase/ownership snapshot, and an unfunded maximum recommendation plus listing. Removing Facebook therefore does not produce a defensible return ranking or establish that the remaining portfolio was economically successful or weak. [researcher inference · strong]

View 2 — historical importance

View 3 — decision-learning value

Evidence boundary and searched denominator

Scope and inclusion rules

• Entities and vehicles: Greylock Partners and named Greylock XI, XII/XII-A, XIII/XIII-A, XIV and affiliated principal, management, GP, or discovery entities when primary filings expose them. A named personal angel relationship enters only when a strong source expressly identifies Sze, and is kept separate from Greylock capital; Northern Lights funds' underlying holdings, nonprofit endowment investments, and companies merely mentioned by Sze remain excluded. [researcher inference · strong]
• Years: 2000, when Sze joined Greylock, through 2026-08-23. Earlier operating employers are not investments.
• Routes searched: Greylock's current Sze page, current consumer/full portfolio, migrated author archive, Yale's 2015 institutional profile, Forbes/Midas records, SEC issuer and insider filings, financing releases, Sze writings/talks, acquisition reports, adverse searches, and third-party discovery profiles. Greylock profile, portfolio sections Yale Endowment 2015, printed p. 9 SEC full-text route
• Deduplication: ToyTalk renamed PullString and is one relationship; SGN/Jam City is one underlying company; musical.ly's later TikTok/ByteDance outcome is not a second Sze deal; financing rounds and follow-ons do not create new companies; Oodle/Qdole is an OCR/name variant. [researcher inference · strong]
• Included: one of the official, institutional, contemporaneous, or credible press sources must attribute Sze or Sze jointly with another Greylock investor. “Greylock invested” alone is insufficient for personal credit.
• Deep reconstruction threshold: enough public evidence to freeze at least a decision-date information set, actor/vehicle boundary, decision or dissent, and outcome. A named acquired company with no thesis or terms remains in the ledger but is unresolved.

Count reconciliation

Coverage fraction: not computable. The numerator can be named—33 publicly discovered, defensibly attributed relationships and eight sufficiently reconstructed cases—but the career-wide denominator is missing. Greylock's lists are selected and current, the preserved 2007 bio is a snapshot rather than a complete lifetime ledger, Sze has described unnamed early enterprise losses, private funds do not publish every investment/pass/write-off, and public profiles favor exits and recognizable winners. Reporting 8/33 would measure depth within this selected list, not coverage of Sze's actual portfolio; it is therefore not a defensible career coverage fraction. [researcher inference · strong] Semil Shah transcript, Parts II–III Greylock profile, selected portfolio Preserved 2007 Greylock-bio quotation

Historical-profile provenance. SourceWatch preserved a page it says quoted Greylock's then-current biography and was accessed on 2007-07-31; the original Greylock page was not recoverable through the public archive route used here. The preserved list expressly names Glimpse, Wink, Troux and Claria. Troux's 2003 Greylock-led Series B and Sze's board role have independent secondary corroboration, and a separate role database dates Sze's Glimpse directorship, but the four relationships remain [contemporaneous record · provisional] for terms, thesis, and economics rather than being silently omitted. Preserved 2007 Greylock-bio quotation, full text and dated reference Troux financing record, investor and round MarketScreener, former-position dates

Exclusions and selection bias

The surviving list over-samples public listings, acquisitions, active boards, and deals important enough to narrate. Failed enterprises, small checks, passes, write-offs, and companies erased from current firm pages are structurally less observable. Every portfolio-level conclusion below is conditional on that bias. [researcher inference · strong]

Historical market, regulatory, and base-rate context

2000–2003 — joining after the internet peak

Sze entered Greylock as the Nasdaq boom reversed. The official FRED series provides the daily market record, while Sze later described a poor initial attempt at enterprise investing and a California office still earning local credibility. FRED Nasdaq Composite, daily historical series Semil Shah transcript, Parts II–III [researcher inference · strong] A diligent contemporary could observe financing contraction and failed internet models; the public record found here does not provide vintage-specific US venture loss rates or Greylock XI cash flows, so no numerical base rate is asserted.

2004–2008 — social graphs before mature monetization

LinkedIn approached one million registered users without revenue when Greylock invested, and Facebook reported more than seven million users when the 2006 round was announced. LinkedIn reflection, “The Context of the Investment” Meta financing release, full announcement Pew's later 2005–2015 survey series helps locate adoption but is retrospective relative to both decisions; it must not be treated as information available in 2004. Pew, adoption chart and methodology [researcher inference · strong] The investable question was whether engagement and graph utility could precede monetization without exhausting capital—not whether social networking would later be popular.

2009–2013 — public exits, regulation, and larger reserves

Pandora's investment followed a prolonged royalty-rate dispute; the Copyright Royalty Board's Web II docket is the primary contemporary route, and Pandora's S-1 makes music-license cost and regulation explicit risks. Copyright Royalty Board, Web II docket Pandora S-1/A, Risk Factors [contemporaneous record · strong] By 2013, Greylock announced a $1 billion Fund XIV and said 84% of the prior fund's investments had been seed or Series A, using the larger pool partly to support follow-ons. VentureBeat, fund announcement and Sze remarks The statement is fundraising-era strategy evidence, not a base-rate or performance audit.

2014–2018 — mobile, gaming communication, and creator platforms

Discord's 2016 announcement placed PC gaming inside a reported $90 billion global games market and showed three million early users; these are company/firm-promotional figures, not independent market validation. Discord essay, market and adoption paragraphs Roblox's 2018 memo treated efficient user-generated content, word of mouth, social retention, and aging payer cohorts as structural advantages. Roblox memo, first Summary [investor-stated · moderate] The 2020 S-1 later disclosed child safety, privacy, moderation, platform dependence, developer relations, and regulatory risk; those later disclosures cannot be smuggled into what the public memo says the 2018 IC knew. Roblox S-1, Risk Factors and Business [researcher inference · strong]

2021 onward — public-market scrutiny and AI incubation

Nextdoor's SPAC listing exposed a neighborhood network to public-company retention, monetization, moderation, privacy, and governance scrutiny. Nextdoor prospectus, Business and Risk Factors Inflection was incubated at Greylock and is listed as a joint Hoffman/Sze seed relationship, but the firm page does not disclose security, check, allocation, governance, or economics. Greylock portfolio, Inflection entry [contemporaneous record · moderate] Because the eras differ in capital supply, distribution platforms, regulation, and exit markets, one network winner cannot serve as an undated base rate for another. [researcher inference · strong]

Vehicle economics and incentives

A current LP-backed venture fund can be compared only on supported dimensions: pooled capital, finite-fund incentives, ownership/reserve choices, partnership governance, and illiquidity. Fee/carry levels, check cadence, portfolio construction, and liquidity rights are missing for the relevant Greylock vehicles, so a numerical modern-fund comparison would be false precision. [researcher inference · strong]

People, capital, and influence map

Institutional map

• Greylock partnership
  • David Sze: attributed lead or board relationship on LinkedIn, Facebook, Pandora, Nextdoor and other named companies; one voice inside a partnership and shared manager/GP control in filings, not sole fund ownership.
  • Aneel Bhusri: recruited Sze and co-built the West Coast practice; Workday belongs to Bhusri's firm record, not Sze's.
  • Reid Hoffman: LinkedIn founder, later Greylock partner, Sze recruit, Airbnb sponsor, fund-strategy coauthor, and co-investor on Inflection.
  • Josh Elman: coauthor/board partner on Medium; contemporaneous reporting credits him with leading Discord and musical.ly, limiting Sze-only credit.
  • Seth Rosenberg: joint author of the Roblox memo and co-credited investor.
  • Asheem Chandna: documented Palo Alto Networks investor/director; establishes the exclusion boundary.
  • Specialists and operating advisers: recruiting and portfolio-support work is institutional labor, not automatically Sze labor. Dan Portillo essay, role/output paragraphs
• Capital chain
  • LPs and principals → named Greylock limited partnerships/affiliates → company securities → possible later sale/distribution → fees/carry/partner allocations, the last two layers not publicly disclosed.
  • An issuer footnote saying Sze shares voting/dispositive power describes control for reporting; it does not trace cash through that chain. [researcher inference · strong]
  • Treehouse and Lumi Story instead follow a separate personal-capital branch: Sze or an undisclosed personal entity → company security → unknown outcome; the public record does not disclose the intermediate terms. [researcher inference · strong]

Material-deal responsibility matrix

Unknown cells are evidence, not omissions: public storytelling reveals the visible partner more often than the associate, counsel, finance team, co-investor, or specialist who performed work. [researcher inference · strong]

Comprehensive deal ledger

This ledger covers all 33 defensibly attributed public relationships in the bounded set: 31 Greylock-attributed relationships and two separately sourced personal angel relationships. “Reported acquisition value” is company-level consideration, never a Greylock or personal return. An “unknown” field means the searched public record did not establish it.

Ex-ante forensic case studies and decision labs

LinkedIn 2004 — non-obvious network before revenue

Decision record. Greylock led LinkedIn's $10 million Series B and Sze joined the board; his later essay reproduces selected passages from the 2004 memo, not the complete memo or partnership minutes. [investor-stated · strong] LinkedIn release, announcement LinkedIn reflection, opening and memo sections

Alternatives, risk, and governance. The practical alternative was pass/defer until revenue or denser engagement, at the cost of access and ownership. Sze's selected memo framed product utility, founder quality, monetization sequencing, and non-obviousness; it does not disclose security, liquidation preference, vote, negotiation, reserves, kill criteria, or explicit next-check milestones. He joined the board in September 2004. LinkedIn S-1/A, Management—David Sze, printed p. 89

No-hindsight reader exercise.

• Would you invest then? Decide using one million registrations, no revenue, a professional-use case, and a contested TAM—not the IPO or Microsoft outcome.
• What would kill the deal? Propose a density/retention failure, inability to create repeat professional utility, founder/reference failure, or a security/price that eliminates upside.
• What ownership/terms fit the risk? State a target and protections without assuming the unpublished Greylock terms; explain the dilution budget.
• What evidence unlocks the next check? Define retained cohort use, invitations/connections per user, recruiting willingness to pay, system reliability, and a dated monetization test.

Actual decision and outcome—reveal after the exercise. Greylock invested; Sze became a director; LinkedIn went public in 2011 and Microsoft later acquired it. Sze also admitted two errors: he underestimated future scale work and Reid Hoffman's capacity to build a lasting company. LinkedIn reflection, “Some Last Reflections” [investor-stated · strong] The learning is to preserve the original errors alongside the win; neither acquisition value nor a filing holder row discloses Greylock's realized fund return or Sze's personal proceeds. [researcher inference · strong]

Facebook 2006 — observer seat, not lone board control

Decision record. Facebook announced a $25 million financing led by Greylock and Meritech when it reported more than seven million users; Sze's statement emphasized daily utility, demographic reach, and value to users and advertisers. Meta release, full announcement [investor-stated · strong]

Alternatives, terms, and control. The alternative was pass or smaller participation until broader-market proof. Press reports Greylock's portion at $12.5m, but the issuer release does not allocate the round. Sze held observer rights, not a formal director seat; Zuckerberg, formal directors, Accel/Breyer, Meritech/Madera, and employees carried distinct roles. Facebook S-1/A, footnote 27 and voting agreements [researcher inference · strong]

No-hindsight reader exercise.

• Would you invest then? Use reported engagement, demographic concentration, founder evidence, and the contested price.
• What would kill the deal? Define engagement decay after campus expansion, weak advertiser demand, unsafe/privacy-eroding behavior, founder-control conflict, or ownership below the fund's return threshold.
• What ownership/terms fit the risk? Choose a position consistent with a large round and observer—not director—governance.
• What evidence unlocks the next check? Expansion-cohort retention, daily use, revenue experiments without user harm, infrastructure performance, and governance information rights.

Actual decision and outcome—reveal after the exercise. Greylock invested and held observer rights. At the IPO, Greylock XII entities held 36,656,372 pre-offering shares and offered 6,998,763; the Boston Globe reported a $12.5m Greylock investment, roughly $1.5bn stake value, and about $330m IPO sale, all as journalistic figures rather than audited fund DPI. Facebook S-1/A, Principal and Selling Stockholders, printed pp. 144–145 Boston Globe, opening and IPO-sale paragraphs; access partial/paywalled [contemporaneous record · strong] Sze's personal economics remain unknown.

Pandora 2009 — regulatory risk before the IPO

Decision record. Pandora's filing says Greylock-affiliated entities first invested in its Series F between July and October 2009, buying 20,012,911 shares at $0.7745 each; Sze had joined the board in May. Pandora S-1/A, Management and Certain Relationships, printed pp. 94 and 112–113 [contemporaneous record · strong]

Alternatives, terms, and governance. Greylock could pass, stage capital, or demand protections after a regulatory milestone. The filing supplies issue price, share count, conversion/liquidation mechanics, and Sze's board role, but not diligence, ownership target, partner vote, reserves, or stop criteria. The Copyright Royalty Board docket, not later acquisition value, is the right contemporary regulatory record. Copyright Royalty Board, Web II docket

No-hindsight reader exercise.

• Would you invest then? Underwrite user value and distribution only after modeling royalty sensitivity.
• What would kill the deal? A royalty/revenue ratio with no credible path to contribution margin, weak retained listening, inability to fund the runway, or adverse regulatory reopening.
• What ownership/terms fit the risk? Decide whether preferred protections and staged funding compensate for policy risk.
• What evidence unlocks the next check? Royalty certainty, listening retention, ad load/yield, mobile adoption, gross margin, cash burn, and licensing concentration.

Actual decision and outcome—reveal after the exercise. Greylock XII entities invested about $15.5m by multiplying disclosed shares and issue price. Pandora went public; its 2012 proxy showed 13.03% Greylock-affiliated beneficial ownership. Pandora 2012 proxy, Security Ownership, printed p. 17 [researcher inference · strong] Neither snapshot supplies realized proceeds, net fund return, or Sze's personal economics.

Nextdoor 2011–2013 — pass the first concept, back the pivot

Decision record. Sze says he knew Nirav Tolia, Sarah Leary, and Prakash Janakiraman, passed on Fanbase because distributed content supply and advertising looked hard, and invested after they pivoted to a private neighborhood network. Nextdoor transcript, founder/Fanbase and pivot turns [investor-stated · strong]

Alternatives, risk, governance, and follow-ons. The team could have been backed at Fanbase, rejected again, or funded only after density proof. Sze's account shows the pass/pivot logic and repeated follow-ons; it does not disclose security, ownership target, IC dissent, reserves, or the exact board intervention that led to Sarah Friar's 2018 appointment. Product criticism around racial profiling and censorship is a company/governance risk signal, not proof of a personal Sze act. Axios, 2020 criticism and CEO response [researcher inference · strong]

No-hindsight reader exercise.

• Would you invest then? Separate confidence in the founders from evidence that the neighborhood graph solves a frequent job.
• What would kill the deal? Density that does not retain, unsafe/discriminatory use, unverifiable identity, no native monetization, or capital needs that overwhelm ownership.
• What ownership/terms fit the risk? Choose staged capital and information/governance rights suited to city-by-city activation.
• What evidence unlocks the next check? Retained neighborhood cohorts, posts/replies per verified household, time-to-density, safety incidents/resolution, local-business demand, and burn per activated neighborhood.

Actual decision and outcome—reveal after the exercise. Greylock participated in the 2012 financing, led the reported 2013 round, Sze joined the board, and the company listed via SPAC in 2021. A 2022 filing associated 21,196,977 shares/6.92% with Greylock entities through shared control, not direct Sze ownership. Nextdoor 424B3, Principal Securityholders and footnotes 5/10 [contemporaneous record · strong] Public listing and a security snapshot do not establish venture return.

Discord 2016 — a team pivot with disputed internal credit

Decision record. Greylock led the first financing after Hammer & Chisel shifted to Discord; Sze announced that he joined the board. The current portfolio co-credits Sze and Josh Elman, while a contemporaneous Forbes profile says Elman led the investment. The defensible attribution is shared/uncertain, not Sze-only. Discord essay, financing paragraph Forbes Midas Brink, Josh Elman section [researcher inference · strong]

Alternatives, milestones, and governance. Alternatives were stop after the failed game, fund the game rather than communication tool, wait for retention proof, or back the pivot. The record supports founder history, product mechanism, early adoption, Greylock lead, and Sze board role; it does not reveal terms, exact investment amount, ownership, recruitment, follow-ons, or exit decisions.

No-hindsight reader exercise.

• Would you invest then? Decide from a founder-earned pivot and early word of mouth, not present-day Discord scale.
• What would kill the deal? Poor retained groups, usage only during a transient game, unmanageable abuse, high voice infrastructure cost, or platform exclusion.
• What ownership/terms fit the risk? Set a staged plan that recognizes pivot and moderation risk.
• What evidence unlocks the next check? Retained servers/cohorts, messages/voice hours, organic invitation conversion, cross-game use, cost per active group, and safety response.

Actual decision and outcome—reveal after the exercise. Greylock invested and Sze joined the board; the company remained private/active in the reviewed firm record. Greylock consumer page, Discord entry The strongest lesson is about reopening a decision after a mechanism changes, not a supportable Sze return claim. [researcher inference · strong]

Roblox 2018 — creator economics and social retention

Decision record. A summarized July 2018 memorandum from Sze and Seth Rosenberg recommended investing up to $50 million after user interviews, expert calls, management meetings, and modeling. It was published in 2021, so completeness and retrospective selection cannot be verified. Greylock Roblox memo, Summary and Team [contemporaneous record · moderate]

Alternatives, terms, and governance. Alternatives included pass because the company was already mature, invest less than $50m, wait for safety/economic evidence, or negotiate for ownership/governance suited to platform risk. The memo says “up to” and later “recommend investing $50M”; it does not prove that amount was funded. The firm credits Sze and Rosenberg; Baszucki and the company team own execution.

No-hindsight reader exercise.

• Would you invest then? Underwrite creator economics, social retention, founder durability, and price without the 2021 listing.
• What would kill the deal? Falling retained cohorts, extreme creator/payer concentration, safety incidents without scalable response, platform fee shock, or no ownership-adjusted base return.
• What ownership/terms fit the risk? Specify ownership target, information rights, safety/governance covenants if legally/practically available, and reserve policy.
• What evidence unlocks the next check? Cohort retention by age, creator earnings/concentration, payer conversion, content cost, moderation cost/incidents, cross-platform acquisition, and regulatory milestones.

Actual decision and outcome—reveal after the exercise. Greylock invested and Roblox listed in 2021. Roblox's 2020 S-1 describes extensive safety/moderation systems and material privacy, platform, regulatory, and developer risks absent from the public memo summary. Roblox S-1, Business and Risk Factors [researcher inference · strong] Because Greylock is absent from the >5% holder table, the record cannot rank Roblox's economic contribution from the listing alone.

Digg 2005–2012 — a network that did not compound

Decision record. In 2006 Forbes reported that Sze understood Digg's user-driven news product and accepted a smaller-than-usual investment; by April 2010 Sze said Digg was doing well. Forbes, “Googlespawn,” Digg paragraphs TechCrunch 2010 transcript, Digg discussion [retrospective witness · moderate]

Warnings, capital response, and governance. The reviewed record does not establish Sze's response to warning signs, board interventions, follow-on decisions, or an exit/stop directive. That absence prevents a complete causal postmortem. The public outcome reporting says Digg raised about $45m, then sold technology, staff, and brand/assets in parts for disputed amounts. TechCrunch 2012, opening and financing paragraph Wired 2012, opening [contemporaneous record · strong]

No-hindsight reader exercise.

• Would you invest then? Decide from the curation mechanism and small-check option, not Reddit's or Digg's later trajectories.
• What would kill the deal? Contributor capture, manipulation, retention collapse, publisher conflict, failed monetization, or a redesign that breaks the core community.
• What ownership/terms fit the risk? Price the small check against governance and follow-on needs; do not assume a small check makes a weak mechanism safe.
• What evidence unlocks the next check? Retained reader/contributor cohorts, submission concentration, manipulation rate, direct traffic, advertiser demand, and successful product-change tests.

Actual decision and outcome—reveal after the exercise. Greylock/Sze invested; the later fragmented sale was reported at amounts far below aggregate capital raised, but exact Greylock ownership, proceeds, and write-off are unavailable. [researcher inference · strong] No sourced Sze lesson or subsequent process change was found; claiming one would be invention.

Path 2012–2018 — capitalized privacy network, undisclosed asset sale

Decision record. Path announced more than $30 million in 2012 financing; TechCrunch's contemporaneous list explicitly paired Greylock Partners with David Sze, while Redpoint led. TechCrunch 2012 financing, announcement and investor list [contemporaneous record · strong]

Warnings, capital response, and governance. A later proposed financing came amid reported US iPhone usage decline; the company ultimately raised a reported ~$77m. TechCrunch 2013, Series C and usage paragraphs The public record does not show whether Sze supported follow-ons, held a board seat, imposed milestones, or sought a stop. Kakao announced an asset acquisition in 2015 without disclosed terms; Path announced service closure in 2018. Kakao release, acquisition announcement TechCrunch 2018, shutdown announcement

No-hindsight reader exercise.

• Would you invest then? Test whether a deliberately smaller/private graph creates frequent utility or merely limits network reach.
• What would kill the deal? Registration without retained use, no monetization native to privacy, expensive international growth, privacy failure, or a round price requiring Instagram-like scale.
• What ownership/terms fit the risk? Tie capital to active-cohort and monetization milestones, with enough information rights to distinguish registrations from use.
• What evidence unlocks the next check? DAU/MAU, retained private groups, content creation/response, country cohorts, privacy incidents, revenue tests, and burn.

Actual decision and outcome—reveal after the exercise. Greylock/Sze joined the 2012 round; Path's principal assets sold on undisclosed terms and the service later shut. This is a full adverse outcome in the observed portfolio, but Greylock loss, recovery, and Sze's claimed lesson remain insufficient public record. The defensible process lesson is analyst-derived: require activity quality and financing-to-exit math, not registration or narrative comparables. [researcher inference · strong]

Twitter pass and Medium 2010–2014 — learning without retroactive certainty

Decision record. Sze said in 2010 that he had evaluated Twitter multiple times and that not investing was his mistake. In 2014, a joint Josh Elman/Sze Medium essay explicitly connected the new investment to Sze's desire to work with Ev Williams after that pass. TechCrunch, Twitter report Medium essay, founder paragraph [investor-stated · strong]

Alternatives, terms, and governance. The counterfactual Twitter investment cannot be valued without knowing round, allocation, follow-ons, dilution, and opportunity cost. Medium alternatives were pass, stage a smaller round, or invest on evidence that discovery/creator incentives had improved. The joint essay credits Elman's founder knowledge and both partners; later Medium communication says Sze stepped down after more than two years but does not state outcome economics. Medium board update, final side note

No-hindsight reader exercise.

• Would you invest then? For the pass, identify the missing historical record before pretending to decide. For Medium, underwrite creator supply, reader discovery, retention, and monetization independently of Williams's prior wins.
• What would kill the deal? Low repeat reading/writing, distribution capture, creator churn, no native monetization, or price that requires Twitter-scale reach.
• What ownership/terms fit the risk? Use staged milestones; do not pay a “miss-repair premium.”
• What evidence unlocks the next check? Repeat readers/writers, completion/saves, creator retention, discovery quality, willingness to pay, and cohort economics.

Actual decision and outcome—reveal after the exercise. Sze passed on Twitter and jointly backed Medium with Elman. Twitter's later success proves the pass costly only in an unconstrained hindsight sense; the forgone portfolio return is not computable. Medium's Greylock return remains unknown. [researcher inference · strong] No public evidence shows a formal process change beyond Sze's stated self-correction.

Airbnb 2010 — constructive dissent, not investment credit

Decision record. Reid Hoffman's retrospective account says he brought Airbnb to the Greylock partnership; Sze argued the investment looked like a mistake but gave Hoffman license to proceed and later kept thinking about how to help. Hoffman, Airbnb partner-meeting and Sze paragraphs [retrospective witness · strong]

Alternatives, governance, and attribution. Greylock could pass, delay, syndicate, or allow Hoffman's sponsored deal within partnership risk limits. The sourced action is organizational: Sze opposed yet did not block; Hoffman invested. Airbnb belongs to Hoffman's investment record, not Sze's wins. Any claimed Sze board work, follow-on authorization, or exit role is unknown.

No-hindsight reader exercise.

• Would you invest then? Write both the sponsor and skeptic cases without knowing the IPO.
• What would kill the deal? Trust/safety failure, illegal supply, weak repeat booking, poor marketplace liquidity, negative unit economics, or unacceptable price.
• What ownership/terms fit the risk? Decide how much loss budget one partner may own when the partnership disagrees.
• What evidence unlocks the next check? Repeat bookings, city liquidity, host retention, incident rate, contribution margin, regulatory permits, and cohort payback.

Actual decision and outcome—reveal after the exercise. Greylock invested under Hoffman's sponsorship and Airbnb later went public; Sze did not earn deal-picking credit for the investment. His learning value lies in constructive dissent and willingness to help after losing, while one famous disagreement cannot prove that mixed votes predict success. [researcher inference · strong]

Portfolio-level forensics

Observed construction in the 33-relationship public set

Pattern tests

1. Network thesis, conditional support. LinkedIn, Facebook, Pandora, Nextdoor, Discord and Roblox provide repeated visible cases of habit, identity, personalization, local density, communication, or creator loops. [documented behavior · strong] Digg and Path show that a network label does not ensure durable retention, governance, monetization, or value capture. [researcher inference · strong]
2. Founder relationships as access, not proof. Hoffman/LinkedIn, Nextdoor's founders, Baszucki/Roblox and Williams/Medium show relationship-rich sourcing. Fanbase demonstrates that Sze could decline familiar founders; Airbnb shows he could also misread an unfamiliar mechanism. [researcher inference · strong]
3. Stage flexibility. Visible stages run from seed/A to F, but later checks may reflect established evidence and require larger outcomes. Without valuation/ownership, stage cannot be tied to return. [researcher inference · strong]
4. Team production. The visible modern record increasingly includes Hoffman, Elman and Rosenberg. That may reflect specialist leverage or simply better surviving attribution; public evidence cannot isolate either mechanism. [researcher inference · moderate]
5. Exit-label bias. Current profile marks even adverse outcomes such as Digg and quickly shuttered Vessel as “acquired.” That taxonomy records corporate disposition, not investment quality. [researcher inference · strong]

No public source found reports fund-level cash flows, gross/net IRR, TVPI, DPI, loss ratio, or a methodology for allocating performance to Sze. The Masters of Scale transcript's narrator-level statement that Sze “personally returned” $2.5bn has unknown vehicle, period, cash-flow basis and calculation; it is reported/provisional, not audited personal performance, and is not used in the ranking. Masters of Scale transcript, PDF search “David Sze” [retrospective witness · provisional]

Failure, miss, and anti-portfolio postmortems

Counterfactual uncertainty is largest where the result is famous: Twitter and Airbnb do not reveal whether the available allocation, terms, reserves, and competing uses would have produced the imagined return. It is also large for Path, Digg and Vessel because company-level adverse outcomes do not reveal Greylock's exact loss or recovery. [researcher inference · strong]

Uncertainty and retrieval ledger

The evidence supports a visible pattern of consumer/network bets and several historically important outcomes. It does not support a complete win rate, audited return multiple, personal-return claim, or claim that the famous set represents all Sze decisions. That boundary is the central result of the deals phase, not a defect to conceal. [researcher inference · strong]

Coverage and evidence base

Exact denominator

The cross-phase source map is broader, but only 12 page-level sources are priority written works by Sze or a named coauthor. Those wrappers represent 14 textual units: 12 public-facing essays/articles plus selected excerpts from Sze’s distinct 2004 LinkedIn memo and a summary of the distinct 2018 Sze/Seth Rosenberg Roblox memo. All 12 pages were accessible, deeply analyzed, and assigned one unique notes/w-*.md file. [researcher inference · strong] Source inventory, Direct writings and firm-authored record

The 12 priority pages are: LinkedIn; Nextdoor investment; Medium investment; Greylock's next fund; $100 million marketplace commitment; professional-CEO reflection; Dan Portillo; Discord; Roblox 2018; Roblox 2021 plus memo; Mike Duboe; and Nextdoor 2021.

Partial, excluded, and inaccessible material named

• LinkedIn: the canonical 2013 personal Medium post is fully accessible, and its Greylock “Early Principles” page is a later migration mirror of the same work, not an independent source. The 2004 investment memorandum survives only as Sze-selected excerpts; the full memo, downside section, vote, price, and terms remain unavailable. LinkedIn essay, “Reid’s Publication” [researcher inference · strong]
• Roblox: the 2021 page is fully accessible, but Greylock explicitly calls the embedded July 2018 memo “summarized.” The unpublished full memo, model, adverse case, and vote record remain inaccessible. Roblox page, “Recognizing Potential” and “Roblox Investment Memo” [researcher inference · strong]
• Nextdoor 2021: only the short written article before Episode Transcript is processed here. The embedded interview transcript is accessible but belongs to TALKS and is excluded from WRITINGS so one page does not become two phase counts. Nextdoor page, article through “Episode Transcript” [researcher inference · strong]
• Not writings by Sze: the official profile, author/archive index, LinkedIn issuer release, Facebook issuer release, consumer portfolio, and full portfolio are metadata, corporate, issuer, or denominator records. Reid Hoffman's Airbnb Reflections is a priority teammate account about Sze, not Sze-authored prose. They inform attribution and counterevidence but do not receive w- notes or inflate the corpus.
• Spoken/metadata routes: “What it Takes to Be a Good Partner to Founders” is a one-quotation wrapper pointing to a spoken TechCrunch source; “David Sze on Investment Strategy” is one-minute video metadata. Both are TALKS/supporting routes, not writings.
• No hidden book corpus: the archive sweep found no Sze-authored book, public LP letter, shareholder letter, testimony, personal newsletter, or complete public IC-memo collection. This is a concentrated, firm-hosted corpus; absence of a public work is not evidence that private writing never existed. Greylock author/archive route, David Sze filter [researcher inference · moderate]

Evidence character

Most pieces are financing, listing, promotion, or fundraising communications. They are first-party evidence of what Sze and named coauthors chose to say, but weak independent evidence that a thesis caused an outcome. The strongest decision artifact is LinkedIn's selected 2004 memo language; the strongest process artifact is the summarized 2018 Roblox memo. Authorship controls are strict: Medium is Josh Elman + Sze; the fund and marketplace essays are Hoffman + Sze; Dan Portillo is Hoffman + Sze; Roblox's memo is Sze + Rosenberg; Airbnb is Hoffman alone; and Greylock—not Sze personally—is the investment vehicle. [researcher inference · strong] LinkedIn essay, memo excerpts Roblox page, memo header

Source-by-source analysis

1. Investing in LinkedIn in 2004: “That’ll Never Work” — 2004 memo excerpts, published 2013

Context, authorship, and access. Sze published this retrospective after LinkedIn's public listing, using selected text from the memorandum he says he circulated before Reid Hoffman's August 2004 Greylock presentation. The full memo and partnership record are unavailable; the 2014 Greylock “Early Principles” page is a duplicate migration. Canonical essay, “The Context of the Investment” and “Reid’s Publication” [investor-stated · strong]

Argument and reasoning. The argument moves in stages: LinkedIn's trusted professional graph could distribute virally; network scale could become difficult to catch; product/UI improvements could turn the database into repeated user value; only then should native recruiting, hiring, sales, and business-development monetization deepen. This is a mechanism, not merely a “network effects” label. Canonical essay, Business, Path Forward, Monetization, and Challenge bullets [investor-stated · strong]

Evidence, examples, and assumptions. Sze freezes several decision-date facts: LinkedIn was approaching 900,000–one million registered users, had no revenue, faced larger professional/social competitors, and operated in a category most people did not yet understand. His memo treats user quality and reasonable acquisition cost as partial de-risking and assumes professional intent will make monetization useful rather than intrusive. Canonical essay, Context objections and Scale bullet [investor-stated · strong] The essay does not publish cohort retention, acquisition cost, engagement frequency, price, ownership, or alternative deals, so the underwriting cannot be fully replayed. [researcher inference · strong]

Decision rules and operating implications. Build the identity/relationship asset, surface recurring utility, delay extraction that could “kill” an immature network, and underwrite people for competitive drive, analytical quality, intellectual honesty, and willingness to take risk. Canonical essay, People and Challenge [investor-stated · strong] Treat mixed partner reaction as compatible with a non-obvious winner—but never as sufficient evidence that a deal is good. Canonical essay, opening mixed-vote paragraphs [researcher inference · strong]

Biography, evolution, and contradiction. Sze's memo authorship supports sponsorship and diligence, not sole deal credit; LinkedIn's filing records his board service from September 2004 and shared Greylock vehicle control rather than personal ownership. LinkedIn S-1/A, Management p. 89 and Principal Stockholders pp. 115–117 He admits that his scale-risk conclusion was too optimistic and that he underestimated both the work ahead and Hoffman as a builder. This is the corpus's clearest self-correction. Canonical essay, “Some Last Reflections” [investor-stated · strong]

2. Our Investment in Nextdoor — early investment essay, migrated 2014

Context, authorship, and access. This is a full Sze-signed essay whose Medium import says 2014, while its deal context and selected biography point to the 2012 investment period. Because the original publication timestamp was not recovered, the essay is not assigned a false exact date. Essay, opening and author sign-off [researcher inference · strong]

Argument and reasoning. Broad social networks did not exhaust social demand; different relationship contexts can support distinct graphs. Sze maps Facebook to friends/family, LinkedIn to work, Twitter to sharing, Path to intimate ties, and Nextdoor to people connected by place. The investable premise is that neighborhood identity supports recurring coordination and local commerce that a general graph handles poorly. Essay, paragraphs beginning “I was an early believer” and “However” [investor-stated · strong]

Evidence, examples, and assumptions. The evidence is observed use—contractor referrals, proposed-stoplight alerts, lost pets, suspicious activity, and events in three Bay Area neighborhoods. Essay, paragraph beginning “Similarly” [investor-stated · strong] The essay offers no cohort table; recurring jobs, scalable density, and transfer beyond affluent launch neighborhoods remain assumptions. [researcher inference · moderate]

Decision rules and operating implications. Search for an unbuilt but socially native graph; articulate the jobs generated by that relationship; then observe behavior inside actual communities. Relationship history improves access but does not substitute for product conviction. Essay, founder-history and use-case paragraphs [researcher inference · moderate]

Biography, evolution, and contradiction. Sze knew Sarah Leary from Greylock and Prakash Janakiraman from Excite; that continuity explains sourcing access. The 2021 article later records that he passed on the same founders' Fanbase idea and invested after the pivot, refuting a “back the people regardless of idea” caricature. 2021 article, paragraphs naming Fanbase and the pivot [documented behavior · strong] The early essay omits safety and moderation; later racial-profiling criticism shows that high local relevance can amplify local harm as well as utility. Axios, opening and product-change discussion [researcher inference · moderate]

3. Our investment in Medium — 2014

Context, authorship, and access. The full post is hosted on Sze's personal Medium account but signed by Josh Elman and David Sze. It says Greylock led, both joined the board, and Elman brought prior work with Ev Williams. Body attribution controls over account byline. Essay, opening, founder paragraph, and sign-off [investor-stated · strong]

Argument and reasoning. Digitization made publishing abundant while discovery remained biased toward recency and click velocity. Medium would combine open creation, stories that move through collections, and feeds using spread and time spent to make durable work discoverable. The thesis is a network/discovery layer, not simply online publishing. Essay, paragraphs beginning “We believe,” “It’s also,” and “For readers” [investor-stated · strong]

Evidence, examples, and assumptions. Blogger and Twitter provide product lineage; Williams's repeated network-building and Elman's direct operating history provide people evidence. The piece assumes time spent is a better quality signal, creators will publish openly, and a feed can balance novelty with depth. It offers no retention, creator economics, moderation, monetization, valuation, or downside case. Essay, product and founder paragraphs [researcher inference · moderate]

Decision rules and operating implications. After a platform transition, look for the bottleneck created by abundance; underwrite both the mechanism and a founder's accumulated domain learning. Use anti-portfolio learning: Sze explicitly describes the deal as a chance to work with Williams after passing on Twitter. Essay, founder paragraph [investor-stated · strong]

Biography, evolution, and contradiction. The shared board appointment prevents sole-credit narration. Medium later announced that Sze left its board after more than two years while Elman remained; the public record here does not explain the handoff or investment outcome. Medium, board update, final side note [contemporaneous record · strong] The predicted “default” status therefore remains an untested aspiration in this corpus, not a demonstrated venture result. [researcher inference · strong]

4. Our next fund and our path forward — Greylock XIV strategy, 2013/2014 migration

Context, authorship, and access. Sze and Hoffman jointly wrote this full fundraising/strategy essay. Medium shows a 2014 import; Greylock XIV's Form D and the fund announcement are dated September 10, 2013. Essay, opening and sign-off Greylock XIV Form D, filing detail

Argument and reasoning. Greylock combines historical continuity with a Silicon Valley move, operator-investors, separate consumer/enterprise teams, recruiting/operating specialists, early-stage concentration, follow-on capacity, and selected growth investing. Networks/marketplaces and cloud/consumerized enterprise software are the stated opportunity clusters. Essay, paragraphs beginning “So,” “Of course,” “We also focus,” and “Moving forward” [investor-stated · strong]

Evidence, examples, and assumptions. The preceding fund reportedly made more than 140 investments, 120 at seed/Series A and ten at Series B. This is a claimed check/company count with no definitions, return denominator, or follow-on reconciliation. The essay assumes operator background and portfolio services improve outcomes and that Silicon Valley concentration improves opportunity access. Essay, team and stage paragraphs [investor-stated · moderate]

Decision rules and operating implications. Institutionalize stage, sector, and service specialization; reserve capital to keep supporting exceptional companies; use growth rounds only when the firm can contribute comparably. The operational corollary is to measure whether services are used and whether ownership/reserves concentrate enough to matter. Essay, service and stage paragraphs [researcher inference · moderate]

Biography, evolution, and contradiction. The named consumer team includes Sze, Hoffman, James Slavet, John Lilly, Josh Elman, and Simon Rothman; enterprise work is attributed separately. The essay is evidence of Sze as institution builder, not evidence he drove every named Greylock deal. Its 140-plus-investment breadth also creates a tension with outlier-return rhetoric: without ownership, reserves, and loss data, the corpus cannot tell whether broad access or concentrated exposure drove results. [researcher inference · strong]

5. Why Greylock is Committing $100M to Marketplaces — 2013/2014 migration

Context, authorship, and access. The full announcement is signed by Hoffman and Sze and tied to Simon Rothman's promotion. It announces a $100 million commitment but does not identify a separate legal vehicle, deployment period, or completed investments. Essay, opening, commitment paragraph, and sign-off [investor-stated · strong]

Argument and reasoning. The case is institutional rather than a deep marketplace memo: pair thematic capital with a partner who scaled eBay Motors, then add office hours, advice, an industry network, content, relationships, and a conference to attract and help marketplace founders. Essay, Rothman and support paragraphs [investor-stated · strong]

Evidence, examples, and assumptions. Rothman's reported eBay record and advising relationships with Lyft, Wanelo, and Poshmark are the principal evidence. The post assumes operator scale experience transfers to investing and that ecosystem services improve sourcing and execution; neither causal link is tested. Essay, paragraphs beginning “First” and “One of the things” [researcher inference · moderate]

Decision rules and operating implications. A thematic program should combine capital, domain leadership, distribution, and reusable founder support. Analysts should treat announced dollars as strategy only after reconciling actual deployment, ownership, reserves, and outcome. Essay, commitment and support paragraphs [researcher inference · strong]

Biography, evolution, and contradiction. The article gives Rothman—rather than Sze—marketplace expertise and future deal leadership. It shows Sze helping design and communicate Greylock's platform. Despite its title, it barely analyzes liquidity, take rate, trust, disintermediation, or marketplace failure; the deep “why” remains missing. [researcher inference · strong]

6. If, Why, and How Founders Should Hire a “Professional” CEO — 2014 migration

Context, authorship, and access. Sze signs this full, short response to a longer Reid Hoffman piece. It is a governance reflection, not a general empirical study. Essay, opening and sign-off

Argument and reasoning. No founder/CEO model works universally. A board contemplating succession should pair forethought with founder self-assessment, define the founder's continuing role, and obtain genuine commitment so the incoming CEO can lead. LinkedIn's Hoffman/Jeff Weiner configuration is Sze's positive case. Essay, paragraphs beginning “Being one” and “Replacing a founder” [investor-stated · strong]

Evidence, examples, and assumptions. Sze cites nine Greylock-backed companies that had gone public in the prior two years: five non-founder CEOs and four founder-CEOs. This success-selected set demonstrates variety, not superiority or causation; failed and private transitions are missing. Essay, comparison paragraph and company footnote [researcher inference · strong]

Decision rules and operating implications. Diagnose stage needs, founder desire/fit, role clarity, and trust before recruiting; define post-transition decision rights before the candidate signs. Use a replacement only when the new configuration can function, not because “professional CEO” is fashionable. Essay, governance paragraph [researcher inference · strong]

Biography, evolution, and contradiction. Sze calls himself “one of the players” in Hoffman's LinkedIn decision, a properly bounded claim. The 2004 LinkedIn memo stressed founder quality; this later piece shows founder conviction coexisting with willingness to redesign leadership at scale. That is conditional governance, not a contradiction. LinkedIn essay, People and final reflections [researcher inference · strong]

7. Please Help Us Congratulate Dan Portillo! — 2013/2014 migration

Context, authorship, and access. Hoffman and Sze jointly sign this full promotion announcement. It is first-party evidence of Greylock's talent platform and internal credit allocation, with promotional bias. Essay, opening and sign-off

Argument and reasoning. Recruiting is treated as an institutional investment capability. Portillo's work spans candidate introduction, university programs and events, portfolio-company systems, and founder education; Jeff Markowitz owns executive talent and Julie Deroche is credited for university programming. Essay, paragraphs beginning “Dan first joined,” “Dan has done,” and “But Dan” [investor-stated · strong]

Evidence, examples, and assumptions. The firm reports an offer every other day, nearly 200 hires in two years, specific Edmodo and Sumo Logic help, and dozens of hires from events. Definitions, acceptance, retention, quality, and a company denominator are absent. Essay, output and program paragraphs [investor-stated · moderate]

Decision rules and operating implications. Find a recurring portfolio bottleneck; hire specialist operators; supply direct help; measure activity; and teach founders to build the capability themselves. The strongest stated service test is to leave every entrepreneur better armed, whether or not the company is in the portfolio. Essay, closing paragraph [investor-stated · strong]

Biography, evolution, and contradiction. This is evidence of Sze/Hoffman as firm builders, while all execution credit belongs to Portillo, Markowitz, Deroche, and teams. The 2020 Mike Duboe essay later applies a similar operator-to-investor logic. Activity counts are not return evidence, and the corpus contains no failed service program. [researcher inference · strong]

8. Our Investment in Discord — 2016

Context, authorship, and access. Sze's full financing essay dates to January 27, 2016, after Hammer & Chisel pivoted from Fates Forever to Discord. Discord essay, opening through financing paragraph

Argument and reasoning. Gaming had become a social activity, but voice/text tools were old, complex, resource-heavy, and bounded to the live session. Discord reduced joining friction and provided a persistent hub before, during, and after play. Discord essay, paragraphs beginning “More importantly,” “But the tools,” and “They created” [investor-stated · strong]

Evidence, examples, and assumptions. The post supplies category-scale claims, a May 2015 launch, reported word-of-mouth adoption past three million users, and the team’s OpenFeint history. Discord essay, launch and financing paragraphs [investor-stated · strong] Organic adoption as product pull and communication as a cross-game layer are assumptions; retention, moderation, safety, unit economics, valuation, and downside are absent. [researcher inference · moderate]

Decision rules and operating implications. Follow a proven team across a pivot when the new problem arises from first-hand pain, the product removes concrete friction, and users distribute it organically. Verify that the product owns the workflow around the transaction—not just the visible moment. Discord essay, founder and product paragraphs [researcher inference · strong]

Biography, evolution, and contradiction. Sze says he knew Jason Citron from 2012, Greylock led the first post-pivot round, and he joined the board. A contemporaneous Forbes profile says Josh Elman led Greylock's Discord investment; sourcing and internal lead therefore remain shared or unresolved, regardless of Sze's byline and board seat. Forbes Midas Brink List, Josh Elman section [researcher inference · moderate]

9. Our Investment in Roblox — 2018

Context, authorship, and access. Sze's full September 2018 essay is the contemporaneous public rationale. It predates Greylock's 2021 publication of the joint memo summary. Roblox 2018 essay, full post

Argument and reasoning. Roblox is “more YouTube than Netflix”: distributed creators, a cross-platform engine, discovery, multiplayer interaction, and a reinforcing creator/user loop can escape the cost and hit concentration of a studio publisher. Roblox 2018 essay, platform-model through network paragraphs [investor-stated · strong]

Evidence, examples, and assumptions. Greylock reports 70 million monthly users, 40 million titles, four million creators, and several billion-play games; Sze adds longitudinal observation of children using early Roblox at a school science fair and Baszucki’s persistence. Roblox 2018 essay, metrics and founder paragraphs [investor-stated · strong] Creator abundance as useful variety, social play as retention, and sustainable payout economics are assumptions; cohort, creator-concentration, cost, safety, and valuation data are absent. [researcher inference · moderate]

Decision rules and operating implications. Underwrite both marketplace sides and the loop between them; compare structural content economics with hit-driven incumbents; verify that creators earn enough and users discover enough variety for compounding to occur. Roblox 2018 essay, creator, player, and network paragraphs [researcher inference · strong]

Biography, evolution, and contradiction. Sze links Roblox to his network-effect career, but the later memo header names Seth Rosenberg as coauthor and Greylock's current portfolio names both partners. Roblox 2021 page, memo header Greylock profile, portfolio companies [contemporaneous record · strong] The post ignores child safety and moderation; Roblox's later S-1 treats user safety, privacy, regulation, and developer relations as material risks. Roblox S-1, Risk Factors [researcher inference · strong]

10. Congratulations Roblox! plus summarized July 2018 memo — published 2021

Context, authorship, and access. The page has two units: Sze's listing essay and a separately headed July 2018 memo summary from Sze and Seth Rosenberg. The embedded audio is excluded. The page is full; the original memo is partial because Greylock calls the version “summarized.” Roblox 2021 page, “Recognizing Potential,” memo header, and Summary [researcher inference · strong]

Argument and reasoning. The 2021 essay reframes Roblox as human co-experience and community. The memo gives the deeper chain: indie creation lowers hit risk and content spend; global creators support localization and variety; word of mouth and influencers lower acquisition; social interaction deepens use; paying users retain and spend; demand attracts more developers. Roblox 2021 page, opening essay and memo Summary [investor-stated · strong]

Evidence, examples, and assumptions. The authors cite user interviews, expert calls, management meetings, a model, emergent social norms, repeat paying behavior, and Baszucki's prior operator history. The published page omits sample sizes, underlying cohorts, model outputs, adverse case, price, ownership, and vote. Roblox 2021 page, memo Summary and Team [investor-stated · strong]

Decision rules and operating implications. Separate base execution from outlier category creation; use user language to detect identity-level attachment; model whether distributed supply improves variety and cost; and size only after testing the downside. The recommendation was up to $50 million alongside another lead—not proof Greylock invested exactly that amount. Roblox 2021 page, opening and closing memo Summary [researcher inference · strong]

Biography, evolution, and contradiction. The memo is joint Sze/Rosenberg work, preserving team credit. The 2021 page appears after a successful listing and can reflect retrospective curation. It also contains an unresolved founding-date inconsistency: the essay says 2004, the memo Team section says 2005. Roblox 2021 page, opening and Team [researcher inference · strong]

11. Mike Duboe Is Greylock's New General Partner — 2020

Context, authorship, and access. This is a full Sze-authored promotion essay, written while the page labels him advisory partner. It is primary evidence of stated partner-selection criteria, not independent evidence of Duboe's efficacy. Mike Duboe essay, byline and full post

Argument and reasoning. Greylock should help from whiteboard through IPO; a consumer investor should understand both zero-to-one product/community growth and at-scale acquisition. Duboe is selected for operating depth, technical empathy, experimentation, data discipline, talent development, and founder usefulness. Mike Duboe essay, partnership through Stitch Fix paragraphs [investor-stated · strong]

Evidence, examples, and assumptions. Tilt and Stitch Fix supply the operating cases; Duboe's portfolio support and stealth marketplace investments supply early investing evidence. The essay assumes functional operating excellence transfers to picking and board work. It provides no references, failed experiments, deal outcomes, promotion vote, or economics. Mike Duboe essay, Tilt, Stitch Fix, and Greylock paragraphs [researcher inference · moderate]

Decision rules and operating implications. Underwrite investors through repeated functional evidence and stage range; value paired traits—competitive/collaborative, rigorous/trustworthy, data-driven/people-driven—and the ability to teach. Mike Duboe essay, final criteria paragraphs [investor-stated · strong]

Biography, evolution, and contradiction. Sze's earlier writings focus on individual consumer decisions; this piece foregrounds specialist succession and institution building. That may show evolution toward organizational leverage, though genre alone cannot establish a change in belief. All named commerce/marketplace theses belong to Duboe. [researcher inference · moderate]

12. Congrats, Nextdoor! — written article only, 2021

Context, authorship, and access. The page displays Sze as the bylined author of the short third-person listing article; editorial composition remains uncertain. Its embedded “Nextdoor Reflections” transcript is reserved for TALKS and is not used here. Nextdoor page, article before “Episode Transcript” [researcher inference · strong]

Argument and reasoning. Unlike “move fast” networks, a hyperlocal graph may grow more slowly because connections require geographic density, yet become more relevant once formed. Pandemic-era neighborhood information, mutual aid, and small-business support are offered as evidence of recurring utility. Nextdoor article, paragraphs 2–5 before transcript [investor-stated · moderate]

Evidence, examples, and assumptions. The article names Sze’s prior relationships with Tolia, Leary, and Janakiraman; his Fanbase pass; investment after the pivot; multiple follow-ons; board service; and Sarah Friar’s later leadership. Nextdoor article, founder, financing, and pandemic paragraphs [contemporaneous record · moderate] Slower formation as stronger durability and local commerce as sufficient support are assumptions. [researcher inference · moderate]

Decision rules and operating implications. Revisit a team after a substantive pivot; distinguish quality/density from headline growth; and treat local utility as a product mechanic. But set explicit density, safety, and monetization milestones so patience does not become immunity from falsification. Nextdoor article, “taken its time” paragraphs [researcher inference · strong]

Biography, evolution, and contradiction. The page says Greylock first invested in March 2012 and led a $21.6 million round in October 2012; contemporaneous issuer and TechCrunch records instead identify a July 2012 $18.6 million round and a February 2013 $21.6 million Series B. Nextdoor issuer release, full announcement TechCrunch, financing and board paragraphs [researcher inference · strong] The discrepancy stays unresolved. Listing-day “built to last” language is promotion, not return evidence. Nextdoor's 2025 10-K reports 21.0 million platform weekly active users, down 5% year over year, while ARPU rose 13%; that mixed current record cautions against equating durability with unbroken user growth. Nextdoor 2025 10-K, Overview and key metrics [contemporaneous record · strong]

Cross-source synthesis

1. The durable core: invest in relationship-specific networks with native utility

Across LinkedIn, Nextdoor, Discord, Roblox, and Medium, Sze's recurring object is not “consumer” in general but a graph organized around a specific context: professional identity, geography, gaming friendship, creator/player interaction, or publishing/discovery. Each case proposes a native repeated job and a reinforcing loop. LinkedIn essay, Business and Path Forward Nextdoor investment essay, relationship-context paragraphs Discord essay, product paragraphs Roblox 2018 essay, network paragraph Medium essay, discovery paragraphs [researcher inference · strong]

The causal claim is conditional: context creates a reason to join; useful actions create repeat behavior; repeated behavior increases graph value; and graph value can enable native monetization. Repetition across five writings establishes stability in Sze's stated lens, not that the lens alone caused returns. [researcher inference · strong]

2. Utility and trust precede extraction—and sometimes speed

LinkedIn's memo delays monetization until user value and habit improve; Nextdoor accepts slower growth while local density forms; Discord removes communication friction around the full session; Roblox pays creators after content succeeds. LinkedIn essay, Path Forward and Challenge Nextdoor 2021 article, “taken its time” paragraphs Discord essay, tool-friction paragraphs Roblox memo summary, structural advantages [researcher inference · strong]

The operational implication is to distinguish high-quality compounding from merely slow growth. Sze's corpus does not publish kill criteria, so an analyst must precommit to density, retention, safety, CAC, and monetization milestones; otherwise “patient network building” can rationalize weak product-market fit indefinitely. [researcher inference · strong]

3. Relational sourcing is an access edge, not a substitute for product judgment

The writings repeatedly connect deals to prior relationships: Hoffman/LinkedIn, Leary and Janakiraman/Nextdoor, Citron/Discord, Williams and Elman/Medium, and years of school-science-fair exposure to Roblox. LinkedIn essay, People and reflections Nextdoor investment essay, founder-history paragraphs Discord essay, Citron paragraph Medium essay, founder paragraph Roblox 2018 essay, science-fair paragraph [documented behavior · strong]

The Fanbase pass and later Nextdoor investment provide the necessary boundary: knowing a strong team generated a second look, not automatic capital. Nextdoor 2021 article, founders and pivot [documented behavior · strong] A modern analyst should record what privileged access revealed and then demand product-specific evidence that an unaffiliated investor could audit. [researcher inference · strong]

4. “Non-obvious” means a mechanism survives objections—not contrarian theater

LinkedIn records larger competitors, no revenue, and an unfamiliar category; Discord records old tools despite large gaming behavior; Roblox contrasts platform economics with hit publishers; Nextdoor accepts slower bootstrapping for denser local ties. LinkedIn essay, Context objections Discord essay, category/tool gap Roblox 2018 essay, model comparison Nextdoor 2021 article, growth paragraphs [investor-stated · strong]

Sze's mixed-vote claim is useful only as process evidence. The corpus supplies no complete vote/outcome denominator, and Hoffman records Sze opposing Airbnb before later revising his view. LinkedIn essay, opening mixed-vote claim Hoffman, Airbnb partner-meeting and later Sze paragraphs [researcher inference · strong] The transferable rule is to preserve dissent and name the disputed mechanism, not to prefer controversy for its own sake. [researcher inference · strong]

5. People underwriting evolves from founder pattern recognition to institutional design

The early deal pieces emphasize founders' domain history, persistence, product sense, and intellectual honesty. The fund, Portillo, marketplace, and Duboe pieces broaden the unit of advantage to teams: operator-investors, talent specialists, domain partners, portfolio education, and succession. LinkedIn essay, People bullet Roblox 2021 page, Team Fund essay, team/service paragraphs Dan Portillo essay, service paragraphs Mike Duboe essay, operating criteria [researcher inference · strong]

This is compatible with a partnership-level edge, but it complicates individual greatness. Medium is Elman + Sze; Roblox memo is Rosenberg + Sze; fund/marketplace/Dan pieces are Hoffman + Sze; Discord lead credit is contested; service work belongs to named specialists. [researcher inference · strong]

6. Governance is conditional stewardship, not founder control

The professional-CEO essay treats founder succession as a role-design problem, while Sze's LinkedIn reflection tempers confidence with an explicit admission that he underestimated both the remaining work and Hoffman's company-building capacity. Professional-CEO essay, governance paragraph LinkedIn essay, concluding self-correction [investor-stated · strong]

The tension is real: high-impact boards can become overreach, while humility can become passivity. The corpus offers positive examples but no detailed failed CEO transition or board intervention, so the lesson remains a decision framework rather than a demonstrated governance record. [researcher inference · moderate]

7. Rhetoric versus documented behavior

• Sze says utility should precede monetization at LinkedIn; the contemporaneous filing verifies board tenure and vehicle holdings but cannot prove that this sequencing caused the result. LinkedIn essay, Path Forward LinkedIn S-1/A, Management and Principal Stockholders [researcher inference · moderate]
• He says relationships do not eliminate product judgment; the Fanbase pass/Nextdoor pivot is documented in a later first-party article, but only retrospective board records could verify the complete decision. Nextdoor 2021 article, founders and pivot [retrospective witness · moderate]
• He frames Roblox as distributed creator economics; the 2020 S-1 verifies the platform model and independently documents safety, regulation, and developer risks absent from the celebratory essays. Roblox S-1, Business and Risk Factors [researcher inference · strong]
• Greylock promises portfolio help; the Portillo post supplies firm-reported activity but no counterfactual company outcomes. Dan Portillo essay, output paragraphs [researcher inference · strong]

8. What the corpus cannot establish

The writings do not disclose a complete portfolio, check/ownership ledger, reserves, realized returns, write-offs, fund economics, IC votes, founder references, board interventions, or counterfactual passes. They are therefore strong for stated mechanisms and selected decision history, moderate for behavioral consistency, and insufficient for proving individual causal alpha or fund economics. Greylock profile, selected current/prior portfolio Greylock full portfolio, filters [researcher inference · strong]

Intellectual chronology, evolution, and contradictions

2004 decision / 2013 publication — professional identity before the category had a settled name

Sze's LinkedIn memo treats professional context, trusted connections, user quality, and native monetization as the mechanism. His 2013 reflection admits that “scale risk” was less solved than he believed and that he underestimated the founder. LinkedIn essay, memo excerpts and “Some Last Reflections” [investor-stated · strong] This establishes the earliest documented core and an explicit correction: a promising network threshold is not completion. [researcher inference · strong]

2012–2013 — social graphs fragment by context; the firm scales with them

The Nextdoor essay extends the network lens from professional/friend graphs to private geographic ties and grounds it in concrete jobs. Nextdoor investment essay, context and use cases [investor-stated · strong] The Greylock XIV essay simultaneously expands the unit of investing from Sze's judgment to operator-heavy sector teams, services, early-stage volume, and follow-ons. Fund essay, team and stage paragraphs [investor-stated · strong] Chronology supports coexistence, not proof that one caused the other. [researcher inference · moderate]

2013–2014 — anti-portfolio learning, governance contingency, and platform services

Medium explicitly converts the Twitter pass into a second chance with Ev Williams, while the CEO essay rejects one universal leadership model. Medium essay, founder paragraph Professional-CEO essay, comparison and governance paragraphs [investor-stated · strong] The marketplace and Portillo essays add domain talent and portfolio recruiting as institutional products. Marketplace essay, support paragraph Dan Portillo essay, service paragraphs [investor-stated · strong]

2016 — follow teams through pivots when a new behavioral layer appears

Discord joins the existing network lens to a pivot-learning pattern: Sze knew Citron during Fates Forever, then backed the communication product after word-of-mouth adoption. Discord essay, Citron, launch, and financing paragraphs [investor-stated · strong] The unresolved Josh Elman lead-credit report prevents a Sze-only reading. Forbes Midas Brink List, Josh Elman section [researcher inference · moderate]

2018 — from user graphs to creator ecosystems and explicit outlier cases

Roblox expands the model: the graph includes creators as supply, users as players/social participants, and a payout/discovery loop. Roblox 2018 essay, creator/player/network paragraphs [investor-stated · strong] The joint memo adds user interviews, industry calls, management meetings, a model, a base-return case, and outlier category upside. Roblox 2021 page, memo Summary [investor-stated · strong] It also reveals that the most decision-useful artifact is team-authored and only partly public. [researcher inference · strong]

2020 — succession and specialist leverage

The Duboe piece shifts the foreground from a specific investment to selecting the next investor: functional operating proof, zero-to-one and scaled growth, quantitative rigor, people judgment, and founder service. Mike Duboe essay, Tilt through criteria paragraphs [investor-stated · strong] This could represent intellectual evolution toward institutional leverage; because genre changed, it cannot prove Sze abandoned direct picking. [researcher inference · moderate]

2021 — retrospective durability meets omitted downside

The Roblox and Nextdoor listing pieces present community, adaptability, and durability after positive milestones. Roblox 2021 essay, opening and “A Whole New World” Nextdoor article, opening through pandemic paragraphs [investor-stated · strong] They also sharpen the corpus's bias: safety, moderation, regulation, valuation, and post-listing performance are much clearer in issuer filings and adverse reporting than in firm celebration. Roblox S-1, Risk Factors Nextdoor 2025 10-K, Overview and Risk Factors [researcher inference · strong]

Stable ideas, genuine changes, and unresolved contradictions

Decision playbook and modern translation

Historical rule 1 — define the graph and its native job

Rule. Name who is connected, why the connection matters, which action repeats, and why a general network cannot serve it as well. LinkedIn's professional identity, Nextdoor's geography, Discord's gaming friendship, and Roblox's creator/player loop are the grounded cases. LinkedIn essay, Business Nextdoor investment essay, context paragraphs Discord essay, product paragraphs Roblox 2018 essay, network paragraph [researcher inference · strong]

Diagnostic questions. What is the graph's atomic relationship? What recurring job brings a user back without paid prompting? Is value local, global, synchronous, asynchronous, professional, intimate, or creator-mediated? What minimum density unlocks value? [researcher inference · strong]

Boundary and failure mode. A distinctive graph can still be too sparse, unsafe, costly, or hard to monetize. “Network effect” without cohort, density, acquisition, trust, and disintermediation evidence is a label, not underwriting. [researcher inference · strong]

Historical rule 2 — sequence utility, habit, and monetization

Rule. Protect an immature network from extraction that damages its reason to exist; first improve value, repeated use, and product clarity. LinkedIn essay, Path Forward and Challenge [investor-stated · strong]

Diagnostic questions. Which monetization surface is native to user intent? What behavior proves habit? What user harm would an advertisement, fee, or lead product create? Which cohort shows willingness to pay without degrading supply? [researcher inference · strong]

Boundary and failure mode. Delayed monetization consumes runway and can hide weak willingness to pay. Set a dated experiment and a harm guardrail, not an indefinite promise. [researcher inference · strong]

Historical rule 3 — use privileged relationships as diligence channels, not thesis substitutes

Rule. Prior work with founders can reveal integrity, speed, and domain judgment; a changed product still needs independent underwriting. Fanbase/Nextdoor is the explicit pass-then-invest case. Nextdoor investment essay, founder history Nextdoor 2021 article, Fanbase and pivot [documented behavior · strong]

Diagnostic questions. What did the relationship reveal that references could verify? What evidence changed after the prior pass? Are friendship and access biasing downside assessment? Who else on the team has decision-relevant history? [researcher inference · strong]

Boundary and failure mode. Familiarity can produce overconfidence, insider narrative, and weak price discipline. Require a written product-specific case and a dissent owner. [researcher inference · strong]

Historical rule 4 — follow a capable team through a pivot only when the new mechanism is observable

Rule. Discord's new product came from the team's own frustration, removed onboarding/performance friction, and showed word-of-mouth adoption. Discord essay, founder, product, and launch paragraphs [investor-stated · strong]

Diagnostic questions. Did the team earn a proprietary insight from the failed/old product? Is adoption pull or promotion? Which retained cohort proves the pivot? Which assets transfer and which sunk costs should be ignored? [researcher inference · strong]

Boundary and failure mode. Founder quality plus a compelling pivot story can conceal repeated wandering. Predefine retention, engagement, and burn milestones. [researcher inference · strong]

Historical rule 5 — underwrite creator platforms as two-sided economic systems

Rule. Compare content cost, variety, localization, creator incentives, discovery, social retention, and monetization—not just aggregate users. Roblox 2018 essay, creator/player/network paragraphs Roblox memo, structural advantages and monetization [researcher inference · strong]

Diagnostic questions. What share of creators earn meaningful payouts? How concentrated are discovery and bookings? Does social use improve retention? Who bears moderation, safety, hosting, and payment costs? Can creators multi-home? [researcher inference · strong]

Boundary and failure mode. User-generated supply can lower studio risk while increasing safety, labor, platform-governance, and regulatory exposure. Roblox S-1, Risk Factors [researcher inference · strong]

Historical rule 6 — preserve disagreement and record the disputed mechanism

Rule. Mixed votes can surface non-obvious outcomes; the memo should state what would make the optimist and skeptic wrong. LinkedIn supplies the claim, while Airbnb supplies a documented Sze dissent and later update. LinkedIn essay, opening Hoffman, Airbnb partner meeting and Sze update [researcher inference · strong]

Diagnostic questions. Is disagreement about facts, mechanism, price, team, or tail size? Which evidence resolves it? Does the skeptic retain license to help after losing the vote? [researcher inference · strong]

Boundary and failure mode. Celebrating division after a winner creates survivor bias. Audit all mixed-vote deals and all unanimous deals before claiming predictive value. [researcher inference · strong]

Historical rule 7 — make CEO succession a role-system decision

Rule. Decide founder role, incoming CEO authority, trust, and commitment together; do not start with ideology. Professional-CEO essay, governance paragraph [investor-stated · strong]

Diagnostic questions. Which company-stage tasks exceed the current CEO's desire or capability? Can gaps be hired around? What does the founder retain? Who resolves conflict? What evidence would trigger reversal? [researcher inference · strong]

Boundary and failure mode. Success-only examples obscure forced removals, ambiguous authority, and culture damage. Run references on the configuration, not only the candidate. [researcher inference · strong]

Historical rule 8 — evaluate a firm platform by counterfactual founder outcomes

Rule. Specialist recruiting, domain operators, and portfolio education matter only if they alter time-to-hire, quality, retention, decisions, or company survival. Dan Portillo essay, service and output paragraphs Mike Duboe essay, partnership and criteria paragraphs [researcher inference · strong]

Diagnostic questions. Which company used the service? What changed versus its baseline? Who did the work? What did it cost? Does access depend on check size or partner? [researcher inference · strong]

Boundary and failure mode. Offers, events, and office hours are activities, not outcomes. Platform branding can also misattribute specialist work to investing partners. [researcher inference · strong]

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

1. Translate “network quality” into cohort-level density, not a narrative exemption from growth. The durable part of Sze's Nextdoor lesson is that growth method should fit graph topology. The changed assumption is that scaled networks now face mature trust/safety, privacy, and monetization constraints. Nextdoor's 2025 record—21.0 million platform weekly active users, down 5% year over year, alongside 13% ARPU growth—shows why engagement and monetization must be read together. Nextdoor 2025 10-K, Overview and key metrics [researcher inference · strong] Observable applicability signals: neighborhood activation time, retained density, verified-user participation, safety incidents, local-business repeat spend, and cohort ARPU. Likely misuse: calling low growth “quality” without evidence.

2. Translate creator-platform upside together with a safety/regulatory cost model. Roblox's creator/user loop clearly transferred: its 2026 proxy says Q4 2025 daily active users exceeded 144 million and attributes engagement to content breadth and social features. Roblox 2026 proxy, Founder and CEO letter [contemporaneous record · strong] The changed assumption is that children's platforms face a clearer regulatory burden: the FTC published amended COPPA rules in April 2025, and the European Commission published DSA minors-protection guidelines in July 2025. FTC, COPPA Final Rule Amendments European Commission, protection-of-minors guidelines [contemporaneous record · strong] Observable signals: creator payout concentration, moderation cost per engaged hour, age-assurance friction, report-response time, and retention after safety changes. Likely misuse: treating distributed content as “cheap” while externalizing trust and compliance.

3. Use Sze's Medium thesis to underwrite discovery in an AI-abundant content market, but label the extension. The historical mechanism—content supply rises faster than high-quality discovery—transfers. The AI-specific application is an analyst-created extension, not Sze's stated view: require provenance, repeat reader value, creator incentives, and defensible feedback signals rather than raw generation volume. Medium essay, discovery paragraphs [researcher inference · provisional] Observable signals: saves/completions, repeat topical use, source clicks, creator retention, and spam/duplication rate. Likely misuse: calling any AI feed a network before contribution and retention loops exist.

4. Treat partner/operator specialization as an investable operating system only after measuring transfer. The historical writings support Greylock's belief in operators, recruiting specialists, and domain partners; they do not show causal fund returns. Fund essay, team/service paragraphs Dan Portillo essay, output paragraphs Mike Duboe essay, operating criteria [researcher inference · strong] Observable signals: service adoption, time-to-key-hire, retention, founder NPS with response bias shown, follow-on ownership, and decisions changed. Likely misuse: counting headcount and events as value-add.

5. Keep fund announcements separate from vehicle reality. “$100 million to marketplaces” and “140+ investments” describe strategy and activity; a 2026 allocator needs the actual vehicle, check pace, ownership, reserves, fees/carry, loss ratio, and DPI/TVPI before comparing. Marketplace essay, commitment paragraph Fund essay, stage paragraph [researcher inference · strong] Observable applicability signal: deployed capital and maintained ownership by cohort. Likely misuse: treating announced allocation as performance or personal Sze capital.

6. Preserve an explicit “what changed?” lane for pivots and prior passes. The Discord and Nextdoor cases support reopening a decision when product mechanism and evidence change, while Medium shows anti-portfolio learning. Discord essay, pivot and adoption Nextdoor 2021 article, Fanbase and pivot Medium essay, Twitter-pass sentence [researcher inference · strong] Observable signals: new retained cohort, new user job, new distribution, or a team capability earned from the failed version. Likely misuse: funding serial pivots because the founders are familiar.

7. Operationalize board humility with written decision rights. The transferable lesson combines contingent CEO succession with Sze's own acknowledgment that he underestimated a founder's company-building capacity. Professional-CEO essay, governance paragraph LinkedIn essay, concluding self-correction [researcher inference · moderate] Observable signals: agreed role charter, information cadence, conflict path, founder/CEO trust checks, and board-action follow-through. Likely misuse: using “founder-friendly” to avoid hard governance or “professionalization” to seize control.

Best writings and unresolved record

Ranked best five

1. Investing in LinkedIn in 2004. Best because it preserves selected ex-ante memo language, objections, sequencing, people criteria, and two explicit errors. Its limitation—the missing full memo—is visible. [researcher inference · strong]
2. Congratulations Roblox! plus the summarized 2018 memo. Best process artifact: named joint authorship, proposed sizing, diligence routes, mechanism, base/outlier framing, and team history. Ranked second because retrospective summarization can omit adverse material. [researcher inference · strong]
3. Our Investment in Nextdoor. Best compact exposition of Sze's relationship-specific-network taxonomy, grounded use cases, and relational sourcing. It lacks quantitative and safety evidence. [researcher inference · strong]
4. Our Investment in Discord. Best pivot-stage public thesis: behavior shift, incumbent-tool gap, product friction, early adoption, prior founder experience, financing boundary, and board role. Attribution conflict with Josh Elman remains. [researcher inference · strong]
5. Our next fund and our path forward. Best institutional counterweight to individual-deal storytelling: team design, services, stage mix, reserves, and themes. It is fundraising communication without performance/economic evidence. [researcher inference · strong]

Unresolved record that could change the synthesis

• Complete artifacts: full 2004 LinkedIn memo; full July 2018 Roblox memo/model; full IC memos for Facebook, Nextdoor, Discord, Medium, and ordinary/failing deals; partner vote records and dated adverse cases.
• Attribution: sourcing, sponsorship, diligence, negotiation, board, follow-on, and exit roles for Sze versus Hoffman, Rosenberg, Elman, founders, specialists, and the Greylock partnership. Discord lead credit and Roblox joint work are the highest-priority controls.
• Vehicle/economics: actual checks, ownership, dilution, reserves, distributions, write-offs, fund-level DPI/TVPI/IRR, Sze's carry/economics, and the record excluding Facebook. Announced rounds and shared voting control are not personal returns.
• Negative denominator: every pass, write-off, failed network, failed CEO transition, weak service program, mixed-vote loss, and unanimous winner/loser. Without it, the corpus cannot test non-obviousness, relationship sourcing, or platform-service efficacy.
• Dates and versions: original publication dates for migrated Greylock Medium posts; original Nextdoor financing chronology; Roblox's 2004/2005 founding-date discrepancy; historical Greylock author/profile snapshots.
• Governance: board minutes or contemporaneous accounts covering LinkedIn CEO succession, Nextdoor safety/moderation and SPAC decisions, Discord governance, Medium board handoff, and Roblox child-safety oversight.
• Lower-priority archive: four signed administrative posts are inventoried but not deep-analyzed because they are ★ rather than ★★/★★★. A later freshness/firm-building pass may promote them only if new independent evidence makes them decision-material.
• Access boundary: no priority page remains unopened, but two underlying memos remain partial and private materials may exist. No paywall was bypassed, no transcript was treated as writing, and no full copyrighted source text is stored here. [researcher inference · strong]

Spoken record, denominator, and evidence boundary

This corpus contains 22 priority records, reconciled into 21 note units and 20 independent appearances through 2026-08-23. Two record-level duplicates are retained rather than hidden: TechCrunch’s Twitter report and its 21:11 micro-VC video document one 2010-09-27 main-stage appearance, while the 2022 20VC price compilation reuses the 2021 interview. The 2021 Roblox IPO wrapper is a reissue of the 2019 Greymatter appearance. Priority source map [researcher inference · strong]

The strongest spoken evidence is uneven. TechCrunch’s 2010 conversation, Semil Shah’s 2013 interview and Greylock’s 2021 Nextdoor episode expose full speaker-attributed text. Bloomberg’s 2015 Studio 1.0 audio was retrieved and transcribed with OpenAI, then distilled to timestamped locators without committing a verbatim transcript. Several reported interviews preserve named Sze speech but not full recordings. CNBC, Forbes’s 2013 founder video, Roblox, 20VC and four TechCrunch videos remain partial under the explicit retrieval ledger below. [researcher inference · strong]

Attribution is conservative throughout. Reid Hoffman owns his identity, privacy, Airbnb, cloud and organizational remarks; John Lilly owns his Instagram and Dropbox remarks; Sarah Friar, Nirav Tolia and other company leaders own their operating statements. Firm fundraising, fund marks and portfolio-company outcomes are not Sze’s personal returns. Edited articles are legitimate direct-speech evidence only for the words and context they actually preserve.

Chronological source guide

2008 — lower distribution cost as the web-video wedge

In a contemporaneous reported interview, Sze argued that online video was undergoing the same creation-and-distribution cost collapse seen in music and publishing. Revision3’s case therefore required more than traffic: a direct audience, lower production and distribution cost, and a monetization route had to work together. TVWeek, Sze and Revision3 paragraphs [researcher inference · moderate] See appearance note.

2010-04-06 — the clearest contemporaneous investment-process interview

Sze described a deliberately bottoms-up process: roughly 80–90% of Greylock’s time went to listening to entrepreneurs and only 10–20% to trend work, because a consensus trend may already be crowded. He paired this with stage flexibility, citing checks from $200,000 to more than $10 million. TechCrunch transcript, check-size and closing sourcing turns [investor-stated · strong]

His price rule was conditional rather than cavalier. Price can become irrelevant if the company is truly great, but the investor cannot know that ex ante; overpricing also removes degrees of freedom from later financing. He saw location plus mobile data as an enabling change while refusing to predict the exact winner. TechCrunch transcript, valuation and mobile turns [investor-stated · strong]

The same interview freezes an adverse data point: Sze said Digg was doing well. That contemporaneous optimism should remain beside the later outcome rather than being silently removed from the record. TechCrunch transcript, closing Digg turn [investor-stated · strong] See appearance note.

2010-09-27 — Twitter pass and the micro-VC main stage

TechCrunch’s same-talk report preserves Sze calling repeated Twitter evaluations his mistake and placing the company on Greylock’s wall of shame. His comparison with Facebook, LinkedIn and Pandora supports uncertainty tolerance, not a rule that ridicule predicts success. TechCrunch report, opening through final paragraph [investor-stated · strong]

The 21:11 video and report are one appearance. A contemporaneous agenda schedules Sze and Hoffman on stage that day, overriding the migrated page’s 2011 date. The first-party YouTube embed P0tWOxmMoc4 had no subtitle tracks. Disrupt agenda video page [contemporaneous record · strong] See combined appearance note.

2010-09-27 — distinct backstage micro-VC interview

The 08:15 backstage interview is a separate appearance listed by TechCrunch’s same-day Backstage Pass article. Its metadata establishes topic and participants but cannot allocate vehicle design, control or economics to Sze. YouTube k1YoywR0Lko had no subtitle tracks. Backstage Pass listing video page [contemporaneous record · strong] See appearance note.

2010-12-29 — social networks move into physical life and work

AllThingsD attributes three views to Sze: physical and virtual experiences would integrate; technology was proving more powerful at connection than isolation; and network effects could improve work, with LiveOps and Mechanical Turk as examples. Hoffman, not Sze, owns the page’s identity, privacy and Airbnb remarks. AllThingsD, publisher bullets 3, 5 and 7 [investor-stated · moderate] See appearance note.

2012-05-02 — look one level below the obvious objection

Before Facebook’s IPO but after it had become a large winner, Sze urged investors to look one level below what seems crazy. His Facebook case combined Zuckerberg, engagement, passionate users and the persistence of college relationships into later life. He contrasted this with a Baidu pass dominated by reasons to fail: experience should help isolate the decisive reason to say yes. Forbes, paragraphs 2–7 [investor-stated · moderate] See appearance note.

2012-05-18 — IPO as an operating milestone, not an endpoint

On Facebook’s listing day, Sze described a short celebration followed by employees returning to work and linked the posture to continued shipping at LinkedIn and Pandora. Forbes reported more than $1 billion of value to Greylock and a fund-return claim, but neither is audited realized DPI or Sze’s personal proceeds. Sze also framed the investor’s post-investment role as connecting companies to talent. Forbes, paragraphs 2–8 and final paragraph [contemporaneous record · moderate] See appearance note.

2013-01-29 — CNBC archive boundary

Greylock’s migrated wrapper identifies Nextdoor, Facebook, mobile advertising, European privacy and Twitter as topics, but it exposes no usable speaker-attributed positions. The original CNBC segment or transcript remains required before those topics can be turned into Sze claims. Greylock wrapper [contemporaneous record · provisional] See appearance note.

2013-02-05 — operating formation, failure, Pandora and structured dissent

In the most biographically useful full transcript, Sze traced a move from consulting into Electronic Arts product work, Crystal Dynamics and Excite. Excite’s bottoms-up culture and Excite at Home’s top-down merger became his culture case; search-sector capitulation became part of the Pandora opportunity, while he acknowledged losses and royalty-policy risk. Semil Shah transcript, Parts I–II [investor-stated · strong]

He also admitted being a poor enterprise investor in 2002–2003 and described a supported pivot toward consumer businesses. His internal voting claim was narrower than venture folklore: one strong advocate plus relevant domain support could coexist with objections, while unanimous enthusiasm was often merely obvious. No sample, vote table or return series was published. Semil Shah transcript, Parts II–IV [investor-stated · moderate]

Semil’s transcript, the 2013-04-18 TechCrunch article and the migrated 33-minute video are one underlying interview, not independent corroboration. TechCrunch article route [researcher inference · strong] See appearance note.

2013-05-08 — founder-selection video not recovered

The Forbes page establishes an interview about entrepreneur evaluation, but its current route returned HTTP 402 and the official YouTube asset kvAcVSCSz4Q required sign-in. No founder model is inferred from the headline. Forbes page [contemporaneous record · insufficient public record] See appearance note.

2013-06-26 — a provisional Facebook sourcing lead

An attendee recap says Matt Cohler connected Greylock to Facebook and emphasizes valuation uncertainty. It is a useful lead, not a primary record of sourcing, diligence, security, IC approval or economics. Victor Lin recap, full post [retrospective witness · provisional] See appearance note.

2013-09-10 — main-stage dissent process and Fund XIV

The 26:41 Disrupt session repeated Sze’s claim that Greylock’s highest-return cases clustered in debated decisions rather than unanimous enthusiasm. It supports a process of named advocacy plus relevant experience, not an instruction to invest whenever a room disagrees. The conference report also tied the appearance to the $1 billion Fund XIV announcement; fund size is not personal wealth or realized performance. TechCrunch report, selection-process paragraphs [researcher inference · moderate]

The YouTube embed P9jIhPh3Q6Q runs 26:41 and has no subtitle tracks. A four-minute NSA clip is an excerpt of this appearance, not another interview. TechCrunch video [researcher inference · strong] See appearance note.

2013-09-10 — distinct backstage founder-support interview

A separate 11:54 backstage video is paired with a same-day first-party report. Sze described company building as a lonely journey and said useful investor support combines perspective on difficult standards with humility that it remains the founder’s journey. This is a service philosophy, not proof that Greylock caused company outcomes. TechCrunch, paragraphs 8–9 [investor-stated · moderate]

The embed DRI-k5IDVqo had no subtitle tracks, so the note uses only the cited publisher report and metadata. TechCrunch video [contemporaneous record · strong] See appearance note.

2013-09-10 — AllThingsD reported interview before the announcement

In an interview conducted the prior week, Sze said Fund XIV was raised within weeks from the same LPs without new limited partners, described Greylock’s first fully West Coast team, and said 120 of 143 prior-fund companies were seed or Series A. He also named seven current picks. These are institutional positioning and portfolio-count claims, not audited LP demand, investment attribution or DPI. AllThingsD, paragraphs 4–6 [investor-stated · moderate] See appearance note.

2013-09-10 — Forbes interview on stage focus and services

In a separate interview, Sze described Greylock as primarily early stage, with selective growth investing. He rejected a false choice between classic venture work and platform services, citing formal recruiting and executive peer groups, and presented marketplaces as a mobile and broadband-enabled consumer theme. The reported 200 hires lack a denominator or causal test. Forbes, paragraphs 3 and 5–8 [investor-stated · moderate] See appearance note.

Mar/Apr 2014 — small-company orientation in compressed form

The Yale Alumni Magazine profile briefly confirms Excite as a formative improvisational environment, Sze’s attraction to small pre-profit technology companies, and his positive-search rule: the investor’s job is to identify reasons a company could succeed. Because the paragraph selects Facebook, LinkedIn and Pandora after success, it is confirmation of a stated method rather than performance evidence. Yale Alumni Magazine, Sze paragraph [investor-stated · moderate] See appearance note.

2015-07-09 — partnership formation, Facebook dissent and Airbnb miss

In Studio 1.0, Sze credited the partnership rather than himself alone for Greylock’s West Coast development and described a shift toward partners with operating backgrounds. He recalled internal fear that Facebook’s $500 million valuation could damage the firm, then admitted he failed to understand Airbnb while Hoffman advocated for it. Bloomberg audio, 02:18–02:39, 03:33–04:18 and 05:04–05:26 [investor-stated · strong]

The current feed is 21:46, not the listed 20:11, because it includes dynamic advertising at 00:00–01:36; the programme begins at 01:36 and substantive interview at 02:10. Later market remarks are treated as joint or unspecified where diarization could not separate Sze and Lilly. See timestamped appearance note.

2019-06-25 — Roblox appearance, unresolved speech

The 40-minute Greymatter episode is a distinct spoken record from the written Roblox memo, but a reliable transcript was not recovered, so no additional thesis is inferred. The 2021 IPO reissue is the same underlying appearance. Amazon original metadata Greylock reissue [contemporaneous record · strong]

The official Greylock YouTube mirror TQ4HbcWo2v8 is private; the SoundCloud route returned HTTP 404 through yt-dlp; and the Filmot Internet Archive route timed out after 30 seconds with HTTP 000. No privacy bypass was attempted. See appearance note.

2021-04-19 — 20VC agenda without retained answer text

The canonical RSS item is dated 2021-04-19 UTC, while the publisher page labels 2021-04-18. Its 42:37 audio and detailed agenda cover Sze’s career, Facebook and LinkedIn strategy conflict, price discipline, pre-emptive rounds, founder commonalities, partnership safety, younger-investor development and board work. The agenda establishes topics, not Sze’s precise answers. 20VC publisher page [contemporaneous record · strong] See appearance note.

2021-11-08 — Nextdoor pass, pivot, network quality and governance

The full transcript supplies a rare decision sequence. Sze liked the known founders but passed on Fanbase because distributed supply and advertising looked too difficult, then revisited them after a pivot to a geography-specific neighbor graph. Greylock transcript, founder and Fanbase turns [investor-stated · strong]

He contrasted address-book hypergrowth with slow neighborhood-by-neighborhood bootstrapping, treating privacy, density and quality growth as part of the mechanism. He also named Sarah Friar as the top CEO-search candidate and acknowledged racism and profiling as unexpected network problems. The listing-day transcript does not disclose the full board process or independently prove moderation efficacy. Greylock transcript, growth, Friar and moderation turns [investor-stated · moderate] See appearance note.

2022-08-12 — explicit duplicate, not a belief update

The 28:03 20VC price-sensitivity compilation reuses Sze material from the 2021 interview. It remains a priority record with its own duplicate-control note, but it is excluded from the independent-appearance denominator. 20VC compilation page [researcher inference · strong]

Recurring decision frameworks

1. Listen before naming the market

Sze’s most consistent sourcing rule is empirical and bottoms-up: spend most time with entrepreneurs, use trend maps sparingly, and let unexpected company-level evidence revise the taxonomy. His 2010 ratio of 80–90% entrepreneur listening to 10–20% trend work is unusually specific. TechCrunch transcript, closing sourcing turns [investor-stated · strong] The later public themes—mobile, marketplaces, physical and virtual integration, and work networks—show that he still used top-down frames; the distinction is weighting, not abstention. AllThingsD, Sze bullets Forbes, marketplace paragraph [researcher inference · strong]

2. Build the positive case one level beneath the objection

“Crazy” is not itself a signal. Sze’s stronger rule is to identify the concrete mechanism that could invert the obvious objection, then find evidence already visible: Facebook had unusual engagement, passionate users, a founder he rated highly, and a graph that could persist after college. Forbes, paragraphs 2–7 [investor-stated · moderate] Baidu and Twitter show the failure mode: a list of credible risks can obscure an upside mechanism, but hindsight does not reveal which contemporary evidence should have changed the decision.

3. Preserve informed dissent, not controversy theater

Across the Semil transcript, the 2013 Disrupt report and 2015 Studio 1.0, Sze favors an advocate-plus-domain-support model and a partnership able to disagree safely. Semil Shah transcript, Part II Bloomberg audio, 03:33–04:18 [investor-stated · moderate] Repetition establishes consistency, not causal efficacy. The missing denominator is decisive: losing contested deals, passing unanimous decisions and return measurement are unpublished.

4. Treat price as a two-stage problem

Sze’s 2010 formulation holds two truths together: ownership cost can look trivial in an extraordinary winner, yet the investor cannot know the winner in advance, and excessive price can impair later financing flexibility. TechCrunch transcript, valuation and Goldilocks turns [investor-stated · strong] A practical reconstruction is therefore: first underwrite whether an outlier mechanism exists; then require entry and reserve terms that preserve acceptable returns and future financing paths. [researcher inference · strong]

5. Convert operating experience into bounded service

Sze repeatedly links investor usefulness to product and operating experience, recruiting, peer networks and judgment during difficult decisions. He also insists the company remains the entrepreneur’s journey. Semil Shah transcript, Parts I and IV TechCrunch backstage report, paragraphs 8–9 [investor-stated · strong] This is a posture, not proof that operators outperform or that firm services caused hires, retention or returns.

6. Specify the network and its bootstrapping constraint

The network thesis becomes more precise over time: connection and work effects in 2010; mobile-enabled marketplaces in 2013; and a geography-bound neighbor graph in 2021. AllThingsD, Sze bullets Nextdoor transcript, pivot and quality-growth turns [researcher inference · strong] Nextdoor adds a useful constraint: a slower, locality-by-locality bootstrap may be part of product quality, while privacy, moderation and discrimination are endogenous network costs rather than external footnotes.

7. Keep an anti-portfolio and do not erase embarrassment

Sze publicly retained Twitter and Airbnb misses and admitted early enterprise-investing difficulty. TechCrunch Twitter report Bloomberg audio, 05:04–05:26 Semil Shah transcript, Part III [investor-stated · strong] The right use is diagnostic: preserve the dated rejected thesis, the evidence then available and the observation that would have reversed the decision. “Everyone misses” is not a process improvement.

Biographical turning points and intellectual chronology

• Operator formation: Electronic Arts, Crystal Dynamics and Excite gave Sze product and company-building exposure; he later used the Excite and Excite at Home culture contrast as a partnership and merger lesson. Semil Shah transcript, Part I [investor-stated · strong]
• Early Greylock failure: He described himself as a poor enterprise investor in 2002–2003 and credited Aneel Bhusri and the partnership for supporting a consumer pivot. This is candid retrospective testimony, not a complete loss ledger. Semil Shah transcript, Part III [investor-stated · strong]
• Consumer-network formation: LinkedIn, Facebook and Pandora gave the emerging doctrine recognizable examples: engagement, durable identity or relationship graphs, brand, scale and a reason the apparently broken market could work. The issuer records, not later narration, control board dates and holdings. [researcher inference · strong]
• Institutionalization: By 2013 Sze was publicly describing a West Coast, operator-heavy, primarily early-stage team augmented by recruiting and peer services. AllThingsD, paragraphs 4–6 Forbes, paragraphs 3–6 [investor-stated · moderate]
• More explicit humility: Twitter, Airbnb and Fanbase or Nextdoor make misses and reversibility central rather than peripheral. The 2021 Nextdoor account also brings governance, leadership succession and network harms into a record once dominated by picking. Nextdoor transcript [researcher inference · strong]

Deal rationales: contemporaneous versus retrospective

Attribution ledger and hard boundaries

Evolution, tensions, and skeptical controls

The chronology shows evolution more than wholesale reversal. In 2008–2010 Sze spoke mainly about distribution shifts, networks, price and sourcing. By 2012–2015 he narrated how uncertainty, dissent and operator experience shaped decisions. By the 2021 Nextdoor episode the vocabulary included pivot discipline, quality growth, CEO succession and network harms. That broadening is plausible maturation, but the later record is also more outcome-selected and promotional. [researcher inference · moderate]

Modern translation: a falsifiable 2026 playbook

The following is a researcher reconstruction, not a claim that Sze uses this exact worksheet today. It translates the durable parts of the historical record as of 2026-08-23.

1. Start with the founder-specific anomaly. Write the surprising observed behavior, distribution edge or product loop before assigning a category. A 2026 AI or marketplace deal should not qualify merely because the category is fashionable. [researcher inference · strong]
2. Move one level below the objection. State the obvious reason to reject, the mechanism that could invert it, and the evidence already visible. Add a named falsifier and a date by which it should be observable. This preserves Sze’s positive-search rule without turning it into confirmation bias.
3. Run a disciplined dissent memo. Record the advocate, relevant domain supporters, strongest objections, base rate, counterfactual and kill criteria. Track unanimous and contested decisions across wins, losses and passes.
4. Separate outlier potential from price. First ask whether the mechanism could produce venture-scale value; then model entry ownership, dilution, follow-on reserves and the company’s next-financing flexibility under weak, base and outlier cases.
5. Underwrite the exact network. Identify nodes, relationship type, trust boundary, bootstrap geography or niche, density threshold and abuse surface. Growth that weakens trust is not automatically quality growth.
6. Make service measurable and permissioned. For recruiting, peer networks or board help, specify the founder request, intervention, accountable owner and outcome metric. Preserve the founder’s decision rights.
7. Review the anti-portfolio quarterly. Reconstruct what was knowable at the pass date, what signal was discounted and what would have reversed the decision. Do not score the process solely from later valuation.
8. Keep attribution as a first-class field. Separate introduction, sourcing, diligence, advocacy, approval, board work, follow-on decisions, memo authorship and vehicle economics. A firm logo or shared stage is not a role ledger.

Read or listen to these five first

1. A Conversation with Greylock’s Reid Hoffman and David Sze — best contemporaneous process record: price, check size, bottoms-up sourcing, mobile and adverse Digg evidence; full speaker-attributed transcript.
2. A Career of Company Building — best biography and institutional-process record: operator formation, failure, Pandora risk, consumer pivot and mixed-vote claim; full transcript.
3. Studio 1.0: David Sze and John Lilly — best audio for partnership credit, Facebook dissent and an unusually direct Airbnb miss; use the asset-specific timestamps in the note.
4. David Sze: Nextdoor Reflections — best decision sequence and governance bridge: pass, pivot, network bootstrap, CEO search and moderation; full transcript with listing-day caveat.
5. How to Pick the Next Facebook When People Think You’re Crazy — shortest useful statement of the one-level-down positive case, paired with the Baidu pass; full article but incomplete audiovisual record.

Unresolved spoken record and retrieval priorities

1. TechCrunch video masters: P0tWOxmMoc4 (21:11), k1YoywR0Lko (08:15), P9jIhPh3Q6Q (26:41) and DRI-k5IDVqo (11:54) expose no publisher subtitle tracks. The cited reports and metadata are fully used; creator captions or event-master transcripts would unlock speaker-complete treatment.
2. Forbes founder video: the 2013-05-08 page returns HTTP 402 on the accessible route and kvAcVSCSz4Q requires sign-in. An authorized caption file or licensed transcript is needed.
3. CNBC: the 2013-01-29 Greylock wrapper supplies only topics. Retrieve the original Squawk on the Street master or transcript.
4. AllThingsD 2010 video: the legacy page preserves seven attributed bullets, including three Sze views, but no current asset or captions. Locate the Dow Jones event master.
5. Roblox 2019: the official YouTube mirror is private, the SoundCloud reissue returned HTTP 404 through yt-dlp, and the Filmot archive route timed out after 30 seconds with HTTP 000. Seek an authorized publisher transcript; do not bypass privacy.
6. 20VC 2021 and 2022: official audio and agendas are available, but a full timestamp transcript was not retained. Obtain an authorized transcript and verify whether the 2022 compilation is verbatim or edited.
7. Reported interviews: recover underlying recordings or notes for TVWeek 2008, both Forbes 2012 pieces, the 2013 AllThingsD and Forbes fund interviews, and Yale 2014. Until then, analysis remains bounded to named direct speech and reporter context.
8. Decision evidence: obtain Greylock’s internal voting study, complete deal and pass denominator, dated Twitter, Baidu and Airbnb pass records, Fund XIV cash flows, board minutes and CEO-search or moderation records. These are the missing tests of process efficacy, attribution and return.

Coverage is complete against the frozen public priority denominator; evidence completeness is not. Partial records are retained with exact access boundaries rather than upgraded by inference. [researcher inference · strong]

Listen bottoms-up before naming the market

• Rule — Spend discovery time on founder and product anomalies before imposing a top-down category; write the market map only after identifying the non-consensus behavior. [researcher inference · strong]
• Evidence status — [investor-stated · strong] for Sze's 80–90% entrepreneur-listening versus 10–20% trend-work claim; [documented behavior · moderate] for repeated applications because the deal record is selected.
• Provenance and origin — In a 2010 contemporaneous interview Sze said Greylock spent most time listening to entrepreneurs because a consensus trend may already be crowded. TechCrunch transcript, closing sourcing turn
• Mechanism and reasoning — Founders see local product behavior before it aggregates into a legible market; bottoms-up discovery can therefore surface a changed cost curve, workflow, or graph before price reflects it. It assumes the investor can distinguish an anomaly from noise. [researcher inference · strong]
• Historical and vehicle context — Greylock's West Coast office, operator-partner network, flexible $200,000-to-$10-million-plus check range, and later billion-dollar follow-on pool expanded the anomalies the partnership could pursue. TechCrunch transcript, check-size turns Fund strategy, stage/reserve paragraphs [researcher inference · moderate]
• When to use — Use in emerging consumer behavior, developer/creator ecosystems, and new distribution channels; do not use as an excuse to ignore market structure, regulation, or base rates. [researcher inference · strong]
• How to apply it — Log the anomaly, source, observed behavior, alternative explanations, addressable job, and evidence that would make it a category; kill if independent user data cannot reproduce it or if the market map contradicts required economics. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented decision evidence: Hoffman and the early professional graph preceded a settled social-media category; Sze's memo started from professional utility and user quality. LinkedIn reflection, Context and Business
  • Roblox — documented application: the joint Sze/Rosenberg memo used user interviews, industry calls, and management meetings before its market/outlier case. Roblox memo, Summary
  • Airbnb — counterexample: Sze failed to understand the marketplace evidence Hoffman saw, showing that listening does not ensure interpretation. Hoffman, partner-meeting paragraphs
• Evidence — The rule is directly stated in 2010 and behaviorally compatible with independently documented LinkedIn, Roblox, and Airbnb decisions; those cases do not audit actual time allocation. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Founder narratives can be local, promotional, or non-representative; bottoms-up sourcing can miss regulation and distribution control. Digg shows early community evidence need not compound. TechCrunch 2010 transcript, Digg turn [researcher inference · strong]
• Evolution and contradictions — Sze's later fund essays add sector teams and thematic capital, so bottoms-up listening coexists with, rather than abolishes, top-down work. Fund strategy, team paragraphs [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — In AI and creator tools, begin with repeat behavior and workflow displacement, then test rights, inference cost, platform dependence, and willingness to pay. A hypothetical AI-agent anomaly is [researcher inference · provisional] until cohorts and unit economics are observed; likely misuse is calling every founder demo a new market.

Define the graph, native job, and bootstrap topology

• Rule — Name the nodes, relationship, recurring job, trust boundary, minimum useful density, and route to density before underwriting a network. [researcher inference · strong]
• Evidence status — [researcher inference · strong] reconstructed from repeated investor-stated rationales and documented behavior.
• Provenance and origin — The 2004 LinkedIn memo is the earliest surviving articulation: professional identity and trusted connections support recruiting and business workflows. LinkedIn reflection, Business
• Mechanism and reasoning — Context supplies a reason to join; repeated useful interaction increases graph value; density improves discovery or coordination; native monetization may follow. Each arrow needs evidence rather than the phrase network effect. [researcher inference · moderate]
• Historical and vehicle context — Cheap web/mobile distribution made new graphs possible, while privacy, local verification, moderation, and platform rules varied by context. Nextdoor's slow neighborhood bootstrap and Discord's fast word of mouth demonstrate topology-dependent growth. Nextdoor transcript, growth turns Discord essay, adoption paragraphs [researcher inference · moderate]
• When to use — Use for social, marketplace, communication, creator, identity, and workflow products; do not use when value is primarily single-player or supply does not improve with participation. [researcher inference · strong]
• How to apply it — Draw nodes/edges; specify atomic job and density; instrument retained actions, organic invitations, liquidity, safety, and cross-side value; kill if growth fails to improve member utility or one side captures value without retention. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented application: professional identity and trusted contacts supported recruiting/business use. LinkedIn reflection, Business
  • Nextdoor — documented application: address-verified neighbors required local density and accepted slower rollout. Nextdoor transcript, product and growth turns
  • Roblox — documented application: creators supply experiences while players add discovery, social retention, and payment; Sze/Rosenberg share credit. Roblox 2018 essay, creator/player/network paragraphs
  • Digg — counterexample: a user-driven news graph did not establish durable governance, retention, or disclosed value capture. Digg case
• Evidence — LinkedIn, Nextdoor, Discord, Roblox, and Digg are distinct underlying records; the consistency is strong, but returns and complete IC files are unavailable. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Sparse, unsafe, manipulable, low-trust, or non-monetizable graphs fail; creator concentration and platform dependence can break two-sided loops. Roblox and Nextdoor filings make those risks explicit. Roblox S-1, Risk Factors Nextdoor prospectus, Risk Factors [researcher inference · strong]
• Evolution and contradictions — The rule expands from professional and college identity to geography, gaming communication, and creator/player systems. Fast Discord growth and slow Nextdoor density are not contradictions if topology is specified. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Add AI-agent identity, synthetic-content provenance, data rights, and abuse cost to the graph map. Observable signals remain retained native actions, density, organic acquisition, safety, and cross-side earnings. Misuse is relabeling any collaboration feature a network. [researcher inference · strong]

Sequence utility and habit before monetization

• Rule — Protect an immature network's native user value until repeated behavior is established; monetize through surfaces aligned with that job, under a dated runway and willingness-to-pay test. [researcher inference · strong]
• Evidence status — [investor-stated · strong] for LinkedIn sequencing; [researcher inference · strong] for the cross-case rule.
• Provenance and origin — The 2004 LinkedIn memo prioritized product value and habit before revenue, while naming recruiting and services as possible native economics. LinkedIn reflection, Path Forward and Challenge
• Mechanism and reasoning — Premature extraction can weaken trust or participation; sufficient utility creates repeated intent that can support aligned transactions, subscriptions, recruiting, advertising, or creator payments. Delay still consumes capital. [researcher inference · moderate]
• Historical and vehicle context — Early web networks could raise private capital while monetization matured; Greylock's pooled reserves could extend the test. Current capital costs and privacy/platform rules make indefinite delay less transferable. Fund strategy, reserves [researcher inference · moderate]
• When to use — Use when monetization predictably harms supply, trust, or growth before density; do not use when regulation, cash runway, or basic willingness to pay requires earlier proof. [researcher inference · strong]
• How to apply it — Define native job, retained cohort, trust guardrail, monetization surface, dated experiment, gross margin, and shutdown threshold; kill if the paid surface degrades the retained job or no cohort pays within the loss budget. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented application: improve utility/habit before layering recruiting and services. LinkedIn reflection, Path Forward
  • Nextdoor — researcher-applied illustration: slow local density prioritized graph quality; later public risks show monetization and moderation remained constraints. Nextdoor transcript, growth/moderation turns Nextdoor prospectus
  • Roblox — researcher-applied illustration: creator payouts and discovery are native to the platform, but the public memo does not connect Roblox to Sze's LinkedIn-era sequencing rule. Roblox memo, structural advantages
  • Path — boundary case: privacy positioning did not guarantee durable use or disclosed value capture. Path case
• Evidence — One early direct memo plus distinct Nextdoor, Roblox, and Path records meet breadth; only LinkedIn directly proves Sze articulated the sequence before investment. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Patient network building can rationalize weak product-market fit, capital burn, or non-native ads forever. No public Sze source supplies universal cohort or time thresholds. [researcher inference · strong]
• Evolution and contradictions — LinkedIn's patience, Discord's rapid word of mouth, and Nextdoor's slow density imply that sequence—not speed—is the stable belief. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Require retention and safety guardrails plus an explicit paid experiment before the next financing. Misuse is treating utility as an alternative to economics. Nextdoor's later filing is a grounded reminder that public scrutiny does not remove the trade-off. [researcher inference · strong]

Build the positive case one level below the objection

• Rule — List credible failure reasons, then identify the specific underlying mechanism that could invert them and the observable evidence already supporting it. [researcher inference · strong]
• Evidence status — [investor-stated · strong] as a selection heuristic; [documented behavior · moderate] because the public decision sample is winner-selected.
• Provenance and origin — Before Facebook's IPO Sze described looking one level below what seemed crazy and contrasted Facebook's engagement/founder evidence with his Baidu pass. Forbes, paragraphs 2–7
• Mechanism and reasoning — Objections may encode consensus and suppress price; a mechanism-level positive case can expose an asymmetric outcome. The method fails when the investor searches for any reason to say yes rather than precommitting falsifiers. [researcher inference · moderate]
• Historical and vehicle context — Venture's capped downside and outlier upside reward uncommon positive cases, while pooled reserves permit follow-on. That payoff shape does not make every contrarian deal attractive. [researcher inference · moderate]
• When to use — Use after a credible failure memo in power-law venture decisions; do not use for legal impossibility, unbounded safety harm, or a thesis with no observable precursor. [researcher inference · strong]
• How to apply it — Write failure reasons, deepest disputed premise, positive mechanism, current evidence, base rate, falsifiers, loss budget, and milestone; kill if evidence only restates founder aspiration. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented decision evidence: no revenue and larger competitors were offset by professional identity, quality users, and native jobs. LinkedIn reflection, Context and memo
  • Facebook — retrospective application: engagement, passionate users, and persistent college relationships formed the positive case. Forbes, Facebook paragraphs
  • Pandora — researcher-applied illustration: search-sector capitulation and royalty risk created both objection and upside, but exact memo evidence is absent. Semil Shah transcript, Pandora turns Pandora S-1/A, Risk Factors
  • Airbnb — counterexample: Sze did not see the marketplace mechanism Hoffman saw. Hoffman reflection
• Evidence — Forbes supplies the direct rule; LinkedIn supplies ex-ante memo excerpts; Pandora and Airbnb provide independent boundary evidence. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Confirmation bias, contrarian theater, omitted base rates, and survivor selection are the core failures. No complete Greylock vote/outcome denominator tests whether controversial investments outperformed. [researcher inference · strong]
• Evolution and contradictions — Sze preserved Baidu, Twitter, and Airbnb misses, showing the heuristic remained corrigible; no sourced formal scoring change follows. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — For AI or regulated networks, require a falsifiable technical/economic mechanism, not a narrative that scale solves safety or rights. Misuse is celebrating disagreement. [researcher inference · strong]

Underwrite people through domain learning and intellectual honesty

• Rule — Prefer founders and investors who show repeated domain learning, product judgment, truth-seeking, recruiting ability, and capacity to operate across the next stage; verify those traits through specific decisions. [researcher inference · strong]
• Evidence status — [investor-stated · strong] in deal and partner writings; outcome efficacy remains moderate.
• Provenance and origin — The LinkedIn memo emphasized Hoffman and team; later Roblox and Duboe pieces made persistence, domain history, quantitative rigor, and people development explicit. LinkedIn reflection, People Roblox memo, Team Duboe essay, criteria
• Mechanism and reasoning — In uncertain markets, learning velocity and honest updating improve response to unmodeled events; domain history supplies causal priors and talent access. Reputation alone can masquerade as proof. [researcher inference · moderate]
• Historical and vehicle context — Greylock increasingly selected operator-investors and surrounded founders with specialists. The institution shares the outcome, making personal attribution harder. Fund strategy, team/service paragraphs [researcher inference · moderate]
• When to use — Use for founder, executive, and investing-partner decisions; do not substitute it for product, market, price, or governance diligence. [researcher inference · strong]
• How to apply it — Build a decision timeline; reference failed and successful choices; ask what changed their mind; test recruiting and delegation; define next-stage gaps and kill if answers are polished stories without verifiable behavior. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented application: Sze underwrote Hoffman/team but later admitted underestimating Hoffman's company-building capacity. LinkedIn reflection, People and Last Reflections
  • Roblox — documented application: the joint memo cites Baszucki/Cassel persistence and team research. Roblox memo, Team
  • Mike Duboe — documented institutional application: Sze cites functional range, data rigor, people judgment, and teaching. Duboe essay
  • Fanbase — counterexample/boundary: knowing capable founders did not rescue an unattractive model. Nextdoor transcript, Fanbase turns
• Evidence — Four underlying records support both company and partnership selection; they are authored/promotional and lack a failed-hire denominator. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Halo effects, affinity, founder myth, and selection bias can turn traits into unfalsifiable praise. Require adverse references and product-specific evidence. [researcher inference · strong]
• Evolution and contradictions — The lens broadens from founder pattern recognition to institutional succession without abandoning product tests. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Add AI/security judgment, data-rights discipline, and ability to manage distributed human/agent teams. Likely misuse is scoring charisma instead of dated decisions. [researcher inference · moderate]

Use relationships as diligence channels, not theses

• Rule — Use prior relationships to inspect integrity, learning, and execution at greater depth, but require a fresh product-specific case and independent evidence for every financing. [researcher inference · strong]
• Evidence status — [documented behavior · strong] across multiple relationships; [researcher inference · strong] as an explicit rule.
• Provenance and origin — LinkedIn, Nextdoor, Discord, Medium, and Roblox writings repeatedly disclose prior human ties. Writings, relational sourcing synthesis
• Mechanism and reasoning — Relationship history reduces information asymmetry and improves references; a fresh thesis prevents familiarity from suppressing downside analysis. [researcher inference · moderate]
• Historical and vehicle context — Greylock's founder/partner network and West Coast presence made repeat relationships an institutional asset; that access is not individually reproducible. [researcher inference · moderate]
• When to use — Use with serial founders, alumni, and follow-on/pivot decisions; do not use if loyalty impairs price, governance, or independent user work. [researcher inference · strong]
• How to apply it — Separate relationship evidence from product evidence; identify what the tie uniquely reveals; obtain independent cohorts/references; assign a dissent owner; kill when only the founder's prior reputation supports the deal. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • Nextdoor — documented application and boundary: Sze knew the team, passed Fanbase, then invested after the mechanism changed. Nextdoor transcript
  • Discord — documented relationship, disputed credit: Citron's prior OpenFeint/Fates Forever history informed the new product; Elman/Sze roles remain unresolved. Discord essay Forbes, Elman
  • Medium — documented relationship: Williams and Elman's prior history opened a second chance after Twitter; economics remain unknown. Medium essay
  • Fanbase — counterexample: familiarity did not justify the original concept. Nextdoor transcript, Fanbase turns
• Evidence — Three separate companies plus the explicit Fanbase pass support the boundary; public writing cannot reveal all relationship-driven deals that failed. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Familiarity, reciprocity, access privilege, weak price discipline, and shared-story bias. Unknown internal introductions and votes remain in the deal responsibility matrix. [researcher inference · strong]
• Evolution and contradictions — The relationship network grows from founder ties into partner/specialist leverage, increasing access and attribution ambiguity together. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Keep separate IC fields for access evidence and thesis evidence. Lumi's 2024 financing is a grounded attribution control: a named personal angel relationship does not prove Greylock participation. Lumi financing [researcher inference · strong]

Preserve informed dissent with named ownership

• Rule — Permit a deal when a named advocate owns the thesis and relevant domain evidence survives explicit objections; preserve the dissent and post-decision responsibilities. [researcher inference · strong]
• Evidence status — [investor-stated · moderate] for Sze's partnership process; [retrospective witness · strong] for Airbnb; efficacy is unproven without the vote/outcome denominator.
• Provenance and origin — Sze's LinkedIn reflection and 2013 interview say mixed votes can accompany non-obvious winners, while unanimity may signal obviousness. LinkedIn reflection, opening Semil Shah transcript, voting discussion
• Mechanism and reasoning — Named advocacy concentrates accountability; recorded objections expose the disputed causal mechanism; trust lets the firm act without false consensus. [researcher inference · moderate]
• Historical and vehicle context — A partnership vehicle can allocate a bounded loss budget across divergent views; partner economics and veto rules remain private. [researcher inference · moderate]
• When to use — Use when uncertainty is irreducible but evidence is decision-ready; do not use to bypass legal, safety, fiduciary, or capital constraints. [researcher inference · strong]
• How to apply it — Record sponsor, supporters, objections, probability/base rate, terms, loss budget, kill criteria, follow-on rights, and dissenter's post-close duty; later score both process and result. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented investor account: mixed vote, Sze advocacy, and later forecast correction. LinkedIn reflection
  • Facebook — retrospective application: Sze recalled fear that the $500 million valuation could damage Greylock. Studio 1.0 audio, 03:33–04:18
  • Airbnb — counterexample and process evidence: Sze opposed, Hoffman proceeded, and Sze later tried to help; investment credit remains Hoffman's. Hoffman reflection
• Evidence — Three independent appearances/cases support practice, but all vote details are selected and retrospective except the published LinkedIn excerpts. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Controversy can reflect missing evidence, poor terms, or bad judgment. No complete Greylock decision/vote/outcome dataset was public after source, talk, archive, and filing sweeps. [researcher inference · strong]
• Evolution and contradictions — Sze's Airbnb error demonstrates that constructive dissent can coexist with being wrong; the model optimizes learning and organizational action, not predictive purity. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Track unanimous, mixed, and rejected decisions across all outcomes; likely misuse is preferring disagreement as theater. [researcher inference · strong]

Separate outlier potential from price and financing optionality

• Rule — First test whether the company could become exceptional; then independently test whether entry price, ownership, dilution, and next-round flexibility leave an attractive fund outcome. [researcher inference · strong]
• Evidence status — [investor-stated · strong] for the two-stage price idea; [documented behavior · provisional] because complete entry/exit economics are missing.
• Provenance and origin — In 2010 Sze said price can become irrelevant for a truly great company but cannot be known ex ante, and overpricing removes future financing degrees of freedom. TechCrunch transcript, valuation turns
• Mechanism and reasoning — Company quality and security return are separate variables; a high price compresses ownership and error tolerance even when upside is large. [researcher inference · moderate]
• Historical and vehicle context — Greylock could invest from seed through growth and reserve follow-ons, but fund size, stage, and competition changed the acceptable ownership/price frontier. Fund essay [researcher inference · moderate]
• When to use — Use on every round; do not turn “great company” into permission to omit dilution, liquidation preference, runway, or alternative allocation. [researcher inference · strong]
• How to apply it — Underwrite company outcomes without price; then model security cash flows, ownership, dilution, reserves, next-round step-up/down, and fund contribution. Kill if no plausible path clears the fund threshold. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • Facebook — retrospective application: internal concern focused on a reported $500 million valuation; later outcome does not prove price never mattered. Studio 1.0, 03:33–04:18
  • Digg — researcher-applied sizing illustration: Sze accepted a smaller-than-normal check; whether that reflected an explicit loss budget is unknown, as is the return. Forbes, Digg paragraphs
  • Roblox — documented recommendation: joint memo recommended up to $50 million, but actual funding/ownership is unknown. Roblox memo, Summary
  • Twitter — counterexample: pass terms and available allocation are unknown, so later valuation cannot price the forgone decision. Twitter report
• Evidence — Direct rule plus distinct deal records provide breadth; none supplies a full comparable cash-flow series. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Public valuations are not realized return; ownership, dilution, timing, reserves, preferences, and carry are missing. The 33/42 ledger prevents price anecdotes from becoming a track record. [researcher inference · strong]
• Evolution and contradictions — Flexible checks and larger funds widen the feasible range but can increase pressure for outliers; no sourced later reversal of price discipline appears. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Model down/flat/up rounds and runway, not only TAM. Likely misuse is quoting “price does not matter” without the ex-ante uncertainty clause. [researcher inference · strong]

Size from weak, base, and outlier cases with reserves

• Rule — Size initial and follow-on exposure from explicit weak, base, and category-creating cases, a loss budget, target ownership, and milestone-gated reserve plan. [researcher inference · strong]
• Evidence status — [researcher inference · moderate] built from the joint Roblox memo and Greylock fund strategy; not proven as Sze's universal rule.
• Provenance and origin — The 2018 Roblox memo separated base and outlier cases and recommended investing up to $50 million. Roblox memo, Summary
• Mechanism and reasoning — Venture outcomes are skewed; scenario sizing controls loss while preserving exposure to outliers. Reserves should buy new information, not merely defend sunk cost. [researcher inference · moderate]
• Historical and vehicle context — Fund XIV paired early-stage concentration with follow-on capacity; exact reserve ratios, ownership targets, and partner allocation are private. Fund essay, stage/reserve paragraphs [researcher inference · moderate]
• When to use — Use for initial checks and follow-ons; do not use company valuation as a proxy for fund contribution. [researcher inference · strong]
• How to apply it — Set weak/base/outlier operating assumptions; map security outcomes, dilution, loss budget, reserves, next-check evidence, and portfolio concentration; kill follow-on when milestones do not change expected value. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • Roblox — documented application: base/outlier memo plus up-to-$50-million recommendation; joint Sze/Rosenberg credit. Roblox memo
  • Nextdoor — boundary rather than application: Sze reports multiple rounds and separately describes locality-by-locality density; no source ties a stated milestone gate to any follow-on allocation or return. Nextdoor transcript
  • Facebook — outcome control: it is the dominant disclosed value marker, yet cash-flow contribution remains unknown. Facebook S-1/A, Principal Stockholders Economic ranking
  • Path — counterexample: financing and private-network promise did not yield disclosed durable value. Path case
• Evidence — Four cases/records meet placement breadth, but only Roblox exposes scenario language and none exposes full fund cash flows. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Model is below a universal-rule proof standard. Searches of fund essays, filings, 20 spoken records, and the 33/42 ledger found no reserve schedule, target-ownership rule, or complete portfolio loss ratio. Talks evidence boundary [researcher inference · strong]
• Evolution and contradictions — A small Digg check, $15-million-class disclosed company allocations, and the up-to-$50-million Roblox recommendation show flexibility, not a stable sizing formula. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Tie each follow-on to a named information gain and recompute opportunity cost against new deals. Misuse is inflating the outlier case to justify sunk-cost reserves. [researcher inference · moderate]

Follow cost collapse to the new bottleneck

• Rule — When creation or distribution cost collapses, identify which scarce layer moves to discovery, trust, rights, audience, or monetization, then underwrite value capture there. [researcher inference · strong]
• Evidence status — [investor-stated · moderate] for web video; [researcher inference · provisional] as a reusable cross-case model.
• Provenance and origin — In 2008 Sze compared online video's production/distribution cost decline with music and publishing and tied Revision3 to direct audience plus monetization. TVWeek, Sze/Revision3 paragraphs
• Mechanism and reasoning — Lower supply cost increases content/products; abundance shifts scarcity to trusted discovery, distribution, rights, or conversion. Value accrues only if the company controls the new bottleneck. [researcher inference · moderate]
• Historical and vehicle context — Broadband, mobile, and user-generated content reduced distribution/creation costs while platform dependency and rights remained. Pandora demonstrates that rights costs can remain binding. Pandora S-1/A, content-cost risks [researcher inference · moderate]
• When to use — Use during technological cost discontinuities; do not assume cheaper supply creates profitable demand or defensibility. [researcher inference · strong]
• How to apply it — Map old/new cost stack; measure supply growth; identify scarce discovery/trust/right; test control and willingness to pay; kill if distribution remains rented or supply abundance destroys margins. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • Revision3 — documented application: direct web audience and lower production/distribution cost were paired with monetization. TVWeek
  • Medium — researcher-applied illustration: abundant publishing shifted the question toward discovery and durable reader/writer relationships. Medium essay, product/discovery paragraphs
  • Roblox — researcher-applied illustration: creator production shifts supply economics, but discovery, payout, safety, and platform control remain scarce. Roblox essay Roblox S-1
  • Pandora — counterexample/boundary: internet distribution did not remove royalty dependence. CRB Web II docket
• Evidence — Revision3 is the only direct articulation; Medium, Roblox, and Pandora are labeled researcher applications rather than retrofitted Sze rationales. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Attempted routes across 12 writings, 20 independent appearances, archive searches, filings, and the deal ledger found no second direct Sze statement of the rule. Keep provisional pending a recovered 2008 interview master or other contemporaneous memo. Talks retrieval ledger [researcher inference · strong]
• Evolution and contradictions — The theme moves from media distribution to discovery and creator systems, but rights/safety can become more—not less—important. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Generative AI lowers creation cost; an analyst-created example should underwrite provenance, distribution, trust, and paid intent. Misuse is equating more supply with more value. [researcher inference · provisional]

Make founder service bounded, permissioned, and measurable

• Rule — Deliver help only against a founder-defined bottleneck, with an accountable specialist, preserved founder decision rights, and an observable outcome. [researcher inference · strong]
• Evidence status — [investor-stated · moderate] for founder humility and services; [documented behavior · moderate] because counterfactual outcomes are missing.
• Provenance and origin — Sze described company building as lonely and useful investors as offering standards/perspective while remembering it is the founder's journey. TechCrunch, founder-support paragraphs
• Mechanism and reasoning — Pattern recognition and pooled specialist labor can reduce recruiting or scaling friction; permission and metrics reduce board overreach and platform theater. [researcher inference · moderate]
• Historical and vehicle context — Fund XIV coupled investment capital with recruiting, executive peer groups, domain partners, and platform services. Fund strategy, service paragraphs [researcher inference · moderate]
• When to use — Use for hiring, introductions, organization, governance, and crisis support requested by founders; do not substitute investor activity for operator accountability. [researcher inference · strong]
• How to apply it — Define founder request, bottleneck, owner, intervention, decision right, outcome metric, deadline, and handoff; stop if help is unwanted or lacks measurable effect. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • Dan Portillo — documented institutional application: Greylock described a dedicated recruiting capability and outputs, while causal founder outcomes remain unknown. Portillo essay
  • LinkedIn — retrospective witness: Hoffman called Sze his most helpful director; exact interventions are not public. Hoffman reflection, opening
  • Nextdoor — documented board context: Sze names Sarah Friar as the top CEO candidate and discusses moderation problems; full search/intervention record is absent. Nextdoor transcript, Friar/moderation turns
  • Airbnb — boundary: Sze tried to help after opposing the investment, proving service is not picking credit. Hoffman reflection
• Evidence — Firm essays, founder/partner witness, and Nextdoor transcript provide independent support; none supplies counterfactual service ROI. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Platform headcount, introductions, or hires can be activity without causation; board power can make permission ambiguous. Credit the specialist, founder, and operator. [researcher inference · strong]
• Evolution and contradictions — Sze's record broadens from individual board work to specialists and partner development, increasing scalable support and team-credit complexity. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Measure time-to-hire, accepted introductions, decision quality, or operating milestone against founder request; likely misuse is using “platform” as fundraising marketing. [researcher inference · strong]

Treat CEO succession as a role-system decision

• Rule — Decide founder-versus-professional leadership from the company's next-stage role requirements, founder ability/commitment, candidate fit, governance design, and an explicit transition system. [researcher inference · strong]
• Evidence status — [investor-stated · moderate] for the framework; [documented behavior · provisional] because public examples are selected and incomplete.
• Provenance and origin — Sze's professional-CEO essay rejects a universal answer and frames succession through company needs, founder fit, role design, and commitment. Professional CEO essay, comparison and governance paragraphs
• Mechanism and reasoning — Stage changes can outgrow one role while preserving founder strengths elsewhere; explicit responsibilities and commitment reduce authority ambiguity. [researcher inference · moderate]
• Historical and vehicle context — Board investors could recruit executives through Greylock's platform, but directors owe company duties and founders retain different control rights by issuer/round. [researcher inference · moderate]
• When to use — Use before a scale, crisis, or capability transition; do not use as a generic preference for founders or professional managers. [researcher inference · strong]
• How to apply it — Define 18-month company needs, founder capabilities/desire, role boundaries, candidate scorecard, control/board rights, 90/180-day metrics, and reversal path; kill a transition without founder commitment or role clarity. [researcher inference · strong]
• Examples — Grounded applications, researcher-applied illustrations, and adverse controls are distinguished below. [researcher inference · moderate]
  • LinkedIn — documented role-system evidence: Sze says he was one participant in Hoffman's decision to recruit professional leadership and stresses founder role design and commitment; he separately admitted underestimating Hoffman's company-building capacity. Professional CEO essay, LinkedIn and governance paragraphs LinkedIn reflection, Last Reflections
  • Nextdoor — documented retrospective application: Sze says Sarah Friar was the top CEO-search candidate; the search process, board vote, alternatives, and causal outcome are not public. Nextdoor transcript, Friar turns
  • Greylock succession — researcher-applied analogy: management duties shifted toward Hoffman/Slavet while Sze returned time to investing, showing role redesign at firm level rather than company proof. Forbes, succession paragraphs
• Evidence — One framework essay plus LinkedIn, Nextdoor, and firm-succession records; no failed company CEO transition is deeply documented. [researcher inference · moderate]
• Limits, failure modes, and exception ledger — Model shortfall: success-selected examples and no full board process, candidate slate, conflict, or failed transition. Writings, talks, issuer filings, CEO-search references, archives, and the 33/42 ledger were searched; no additional Sze-specific case met the threshold. Sources assessment Talks retrieval ledger [researcher inference · strong]
• Evolution and contradictions — Strong founder underwriting coexists with contingent replacement; LinkedIn's correction narrows confidence in ex-ante founder ceilings. [researcher inference · moderate]
• Modern VC translation (as of 2026-08-23) — Add distributed-team, AI/security, regulatory, and public-company capability to the role scorecard. This is an analyst-created hypothetical extension; misuse is weaponizing “stage needs” to centralize investor control. [researcher inference · provisional]

Keep an anti-portfolio and reopen only on changed evidence

• Rule — Preserve the pass-date or pre-pivot thesis, terms, rejected evidence, and opportunity cost; review without hindsight and reopen only when the product mechanism or another decision-relevant premise changes. [researcher inference · strong]
• Evidence status — [investor-stated · strong] for Fanbase/Nextdoor and for publicly preserving Twitter/Baidu/Airbnb errors; [documented behavior · moderate] for Discord; the formal review procedure is [researcher inference · moderate].
• Provenance and origin — Sze called repeated Twitter passes a mistake in 2010; his later Nextdoor account explicitly contrasts passing Fanbase with backing a changed neighborhood mechanism. TechCrunch, Twitter report Nextdoor transcript, founder/Fanbase/pivot turns [investor-stated · strong]
• Mechanism and reasoning — A durable record separates process error from outcome luck, while a mechanism diff prevents familiarity or later valuation from masquerading as new evidence. Team quality preserves option value but cannot repair an unchanged product thesis. [researcher inference · strong]
• Historical and vehicle context — Greylock's multi-stage funds could finance changed products and follow-ons, but venture portfolios still ration attention, ownership, and reserves; counterfactual returns remain unknowable without allocation, terms, timing, and alternatives. Fund essay, stage/reserve paragraphs [researcher inference · moderate]
• When to use — Review passes, pivots, lost rounds, dissent, and write-offs; do not reopen for cosmetic rebranding, a new financing, social proof, or later valuation alone. [researcher inference · strong]
• How to apply it — Freeze the old nodes, job, distribution, economics, terms, risks, kill criteria, and rejected signal; diff the new causal chain; require new behavior; reopen only if a prior premise changes; track the subsequent action. [researcher inference · strong]
• Examples — Six bounded cases, including two changed-product applications and adverse controls. [researcher inference · strong]
  • Nextdoor — documented application: distributed Fanbase supply/advertising gave way to an address-verified neighborhood graph. Nextdoor transcript
  • Discord — documented behavior, inferred rule: Fates Forever gave way to a communication layer with organic adoption; Sze/Elman credit remains shared or uncertain. Discord essay Forbes, Elman
  • Twitter — documented admission: repeated evaluations and pass; exact rationale/terms unavailable. Twitter report
  • Baidu — investor-stated miss: Sze says failure reasons dominated the decisive positive case. Forbes, Baidu paragraphs
  • Airbnb — retrospective witness: Sze opposed while Hoffman sponsored; Sze later revised and helped, but the investment remains Hoffman's credit. Hoffman reflection
  • Digg — counter/control: a funded deal was still described positively in 2010 before a weak/disputed sale, so anti-portfolios must include owned failures; Greylock proceeds and causal board actions remain unknown. TechCrunch transcript, Digg turn TechCrunch sale report
• Evidence — Nextdoor and Discord establish changed-product behavior; Twitter, Baidu, Airbnb, Digg, and Medium independently establish error preservation or its limits. Only Nextdoor directly documents pass-to-pivot reinvestment. Medium essay, founder paragraph [researcher inference · strong]
• Limits, failure modes, and exception ledger — Public anti-portfolios are survivor-selected and reputation-enhancing; complete passes, terms, opportunity costs, and formal review actions are missing. The research checked all 12 priority writings, 20 independent appearances, the 33/42 deal ledger, Greylock portfolio/archive pages, filings, and transcript/caption routes; ToyTalk/PullString and other pivots lacked a Sze decision rationale. Sources coverage assessment Talks retrieval priorities [researcher inference · strong]
• Evolution and contradictions — Nextdoor and Discord show changed products can reopen judgment; Medium shows memory of a miss can motivate another deal without proving the mechanism changed or the correction worked. [researcher inference · strong]
• Modern VC translation (as of 2026-08-23) — For AI pivots, require changed workflow, cost, data rights, distribution, or economics plus new cohorts; for any pass, preserve reason codes and opportunity-cost bounds. Misuse is public humility theater or funding the same thesis under a new model label. [researcher inference · strong]

Evidence gaps and candidate models

Inventory reconciliation

Retrieval boundary

The inventory was reconciled against all 12 writing notes, 21 talk-note units representing 20 independent appearances, the writing and talk syntheses, the 33-included/42-searched deal record, current and archived Greylock routes, issuer filings, and the named transcript/caption/archive attempts. Writings coverage Talks coverage Deals denominator [researcher inference · strong]

Highest-value missing proof is: the full LinkedIn and Roblox memos; Facebook/Pandora/Discord IC files; complete Greylock vote outcomes; board/CEO-search records; a reserve/ownership policy; failed service and succession cases; the 2008 web-video master; and a career-wide attributed investment/pass/write-off denominator. Until those records surface, provisional models must not be promoted and excluded slogans must not be restored merely to enlarge the list.

What the evidence supports

The strongest supported claim is narrower than a greatness ranking. In the visible US consumer/network record from roughly 2004–2018, Sze repeatedly asked whether a specific relationship graph served a recurring native job, then combined product judgment with relationship access, partnership dissent, board work, and Greylock follow-on capital. LinkedIn's selected 2004 memo excerpts and the Fanbase-to-Nextdoor sequence are the best decision records; the 2018 Roblox summary is the best diligence/sizing artifact but is joint Sze/Seth Rosenberg work. LinkedIn reflection, memo headings Nextdoor transcript, founder/Fanbase/pivot turns Roblox memo, Summary and Team [researcher inference · moderate]

Facebook is the only defensibly dominant disclosed value marker, but filing holdings and reported listing value are not realized fund cash flows or Sze's wealth. Digg and Path show that a network label does not secure retention, governance, monetization, or value capture; Twitter and Airbnb show that preserved dissent does not prove superior selection. Facebook S-1/A, printed pp. 144–145 Failure table [researcher inference · strong]

What could falsify the thesis

The thesis predicts that unseen Sze-sponsored memoranda should name the nodes, relationship, repeated job, bootstrap path, density/retention evidence, advocate, objections, and milestone-gated follow-on logic. A complete denominator should show stronger outcomes for context-specific graph/job cases than generic enterprise or content bets. [researcher inference · moderate]

It would be weakened by complete memoranda showing brand-following rather than mechanism underwriting; labor records assigning celebrated work elsewhere; a full denominator showing no graph/job advantage; or audited cash flows showing Facebook alone drove the record. The strongest alternative is privileged institutional access plus survivor selection in a favorable social/mobile regime. Bhusri and the partnership built the platform; Hoffman sponsored Airbnb; Elman has reported Discord/musical.ly credit; Rosenberg coauthored Roblox; founders built companies; pooled vehicles held securities. Responsibility matrix [researcher inference · strong]

Attribution and denominator audit

The ledger mechanically reconciles: 33 included relationships (31 Greylock-attributed plus personal angels Treehouse and Lumi), 42 searched candidates, 8 sufficiently reconstructed, 25 included-unresolved, and 9 excluded. Thus 8 + 25 = 33 and 33 + 9 = 42. Aliases/follow-ons are deduplicated; 8/33 is correctly rejected as a career coverage fraction. Count reconciliation [researcher inference · strong]

The source map contains 134 item rows. A corrected final parity pass compared 106 unique HTTP(S) URLs across the seven non-source core documents—and 116 after adding all source notes—against 143 unique URL tokens in the source map; zero content URLs were absent. Mapped-only controls remain deliberate rather than citation orphans. [researcher inference · strong]

Material credit controls passed: Sze was Facebook observer, not lone director; LinkedIn securities belonged to Greylock entities; Discord lead credit is shared/unresolved with Elman; Roblox work is joint with Rosenberg; Airbnb belongs to Hoffman's sponsorship; Palo Alto Networks is hard-excluded under Asheem Chandna; Treehouse/Lumi remain personal angels without invented checks or economics. Meta release LinkedIn S-1/A, pp. 115–117 Forbes, Elman Roblox memo Airbnb reflections PANW proxy, pp. 50–51 [researcher inference · strong]

Reported economics remain bounded: Pandora's preferred-share arithmetic is reproducible, but no snapshot is a realized multiple; acquisition value is not Greylock proceeds; the narrator-level $2.5bn claim stays reported/provisional and outside the ranking. Pandora S-1/A, pp. 112–113 [researcher inference · strong]

Transfer limits for modern VCs

The transferable unit is the test: name graph/job/topology, measure retained native actions, separate access from thesis, preserve dissent and attribution, and model price/ownership/reserves independently from company upside. The package dates translations to 2026-08-23. Modern translation [researcher inference · strong]

Cheap web distribution, emerging social graphs, different privacy/platform regimes, and billion-dollar follow-on pools do not transfer wholesale. Modern work must add abuse cost, privacy, supplier concentration, data rights, content provenance, platform dependence, and willingness-to-pay/runway tests. Roblox and Nextdoor filings are risk controls, not proof that Sze used today's checklist earlier. Roblox S-1, Risk Factors Nextdoor prospectus, Risk Factors [researcher inference · strong]

Unresolved evidence

Material gaps remain: a career-wide investment/pass/write-off denominator; complete LinkedIn/Roblox memos; Facebook/Pandora/Nextdoor/Discord/Medium/Digg/Path/Vessel IC and board records; checks, securities, ownership, dilution, reserves, cash flows; partner labor/carry; and failure response. Retrieval ledger [researcher inference · strong]

Direct-voice exhaustion is documented. WRITINGS has 12 priority page wrappers, 14 textual units, and 12 notes, with two partial memos. TALKS has 22 records, 21 notes, and 20 independent appearances after duplicate reconciliation. Captionless, private, HTTP-402, dead-embed, archive, catalog, annual-report/LP, and oral-history routes have exact barriers. Writing coverage Talk coverage Source coverage [researcher inference · strong]

Ten claim-critical URLs were rechecked through normal web access on 2026-08-24; all ten resolved, including LinkedIn’s canonical Medium page. The earlier command-line HTTP 403 was an access-control event rather than a dead source. No control was bypassed. [researcher inference · strong]

Review method and central-claim audit

The package was read biography → writings → talks → deals → philosophy → models. Every central class required by the field manual was reopened.

Reproducible ordinary-claim audit

The deterministic selector is: the first materially factual claim after the first H2 in each of the seven prerequisite core documents, plus the first materially factual claim after the second H2 in _profile.md, writings.md, talks.md, and investments.md—11 claims total. Headings, local-navigation sentences, methodology declarations, and purely prescriptive/inferential statements are ineligible; a bounded negative-evidence or corpus-count claim is eligible because it can be mechanically audited.

Twenty-point acceptance evidence

Verdict

PASS. All 20 gates clear after adding nine previously absent core-cited URLs to sources.md, removing the stale moving global count of 76 priority rows from writings.md, repairing three Twitter/Medium anchors plus one stale review source-coverage anchor, standardizing the Roblox 2019 note’s URL metadata key, and confirming canonical evidence-label vocabulary. The package supports a falsifiable, moderate-confidence account of Sze's visible process—not audited personal alpha, a complete win rate, or lone-investor causation. Remaining gaps are an exhausted-public-record retrieval agenda rather than filler or unfinished accessible work. [researcher inference · strong]

Direct writings and firm-authored record

• ★★★ | David Sze official biography and portfolio | https://greylock.com/team/david-sze/ | current; accessed 2026-08-23 | official profile/selected portfolio; reliability A for role and firm attribution, C for completeness/performance; access full; locators: Biography, Portfolio Companies, Previous Investments | baseline chronology, current advisory title, direct consumer thesis, board roles, and a selected—not complete—deal denominator
• ★★ | Greylock article archive filtered by David Sze | https://greylock.com/blog-old/ | current; accessed 2026-08-23 | official archive route; reliability A for displayed metadata, C for completeness; access full route; locator: Author filter “David Sze” | enumerate bylined posts and detect migrated/dead Greylock URLs before WRITINGS
• ★★★ | LinkedIn Secures $10 Million in Series B Funding Led by Greylock | https://news.linkedin.com/2004/10/linkedin-secures-10-million-in-series-b-funding-led-by-greylock | 2004-10-13 (page displays Oct. 14) | contemporaneous issuer release with Sze statement; reliability A/C; access full; locator: announcement and final paragraph | earliest public LinkedIn thesis, round participants, reported $10 million size, and board appointment
• ★★★ | Facebook Secures $25M Investment | https://about.fb.com/news/2006/04/facebook-secures-25m-investment-2/ | 2006-04-19 | contemporaneous issuer release with Sze statement; reliability A/C; access full; locator: full announcement | freezes Facebook's seven-million-user information set, Greylock-led round, co-investors, and Sze's user/advertiser thesis
• ★★★ | Our Investment in Discord | https://greylock.com/blog/our-investment-in-discord/ | 2016-01-27 | Sze-authored investment essay; reliability A; access full; locator: “People play games” through closing | contemporaneous category thesis, pivot-stage timing, traction, founder history, Greylock lead, and Sze board role
• ★★★ | Our Investment in Roblox | https://greylock.com/blog/our-investment-in-roblox-2/ | 2018-09-15 | Sze-authored investment note; reliability A; access full; locator: full post | founder persistence, developer/user ecosystem, school-science-fair observation, and original public rationale
• ★★★ | Congratulations Roblox! (NYSE: RBLX) and summarized July 2018 investment memo | https://greylock.com/blog/congratulations-roblox/ | 2021-03-10; embedded memo dated 2018-07 | Sze essay plus Sze/Seth Rosenberg memo summary and audio route; reliability A with retrospective-curation caveat; access full text/partial audio locator; locators: “Roblox Investment Memo,” Summary, Team | highest-value ex-ante decision artifact, diligence method, up-to-$50 million recommendation, coauthor attribution, and IPO-era retrospective
• ★★★ | Congrats, Nextdoor! and David Sze “Nextdoor Reflections” transcript | https://greylock.com/blog/congrats-nextdoor/ | 2021-11-08 | Sze-bylined, third-person listing article plus full interview transcript; article authorship/editorial composition uncertain; reliability A with promotional/retrospective caveat; access full; locators: article and Episode Transcript | sourcing relationships, Fanbase pass, post-pivot investment, repeated follow-ons, governance, network density, and leadership transition
• ★★ | Mike Duboe Is Greylock's New General Partner | https://greylock.com/blog/mike-duboe-is-greylocks-new-general-partner/ | 2020-09-22 | Sze-authored firm essay; reliability A/C; access full; locator: full post | succession, stated partnership values, operator-investor model, board/service claims, and consumer-platform staffing
• ★★★ | Greylock consumer portfolio | https://greylock.com/consumer/ | current; accessed 2026-08-23 | official thematic portfolio; reliability A for current firm attribution, C for completeness; access full; locator: company entries | links Sze to Facebook, LinkedIn, Pandora, Nextdoor, Discord with Josh Elman, musical.ly with Elman, and Roblox with Seth Rosenberg
• ★★ | Greylock full portfolio | https://greylock.com/portfolio/ | current; accessed 2026-08-23 | official selected portfolio route; reliability A/C; access full; locator: investor/company filters | denominator candidate and cross-check for current, public, acquired, and team-attributed relationships
• ★★★ | Airbnb Reflections | https://greylock.com/portfolio-news/airbnb-reflections/ | 2020-12-10 | Reid Hoffman first-person Greylock account; reliability A for witness recollection; access full; locator: partner-meeting and David Sze paragraphs | teammate evidence that Sze was Hoffman's primary recruiter/helpful LinkedIn director and a dissenting voice on Airbnb; essential anti-hero-worship control; not Sze-authored and excluded from the WRITINGS corpus denominator
• ★★★ | Investing in LinkedIn in 2004: “That’ll Never Work” / Early Principles of LinkedIn | https://medium.com/%40davidsze/investing-in-linkedin-in-2004-bba2be414119 | 2013-11-03 canonical author copy; Greylock migration dated 2014-03-01 at https://medium.com/%40greylockvc/early-principles-of-linkedin-a-reflection-of-our-series-b-investment-20f95ff5feb5 | Sze-authored reflection with excerpts from his 2004 investment memorandum; reliability A with retrospective-selection caveat; access full; locators: “The Context of the Investment,” “Reid’s Publication,” and “Some Last Reflections” | strongest surviving LinkedIn decision artifact; preserves ex-ante objections, network/monetization sequence, founder assessment, and Sze’s admitted forecasting errors without treating the excerpts as the complete memo
• ★★★ | Our Investment in Nextdoor | https://medium.com/%40greylockvc/our-investment-in-nextdoor-c6bec078c21e | migrated 2014-02-28; underlying investment context 2012 and original publication date unresolved | Sze-signed investment essay; reliability A with migration-date caveat; access full; locators: paragraphs beginning “We’re very excited,” “I have personally,” and “However” | contemporaneous-to-early-investment thesis for relationship-specific social graphs, founder familiarity, neighborhood use cases, and private-network differentiation
• ★★★ | Our investment in Medium | https://medium.com/%40davidsze/our-investment-in-medium-b1c5e7350595 | 2014-01-28 | Josh Elman/David Sze co-signed investment essay hosted on Sze’s account; reliability A with team-attribution control; access full; locators: paragraphs beginning “All of us,” “We believe,” and “Beyond our belief” | publishing-network thesis, discovery/engagement mechanism, founder-pattern evidence, Sze’s Twitter-pass admission, and explicit joint board attribution
• ★★★ | Our next fund and our path forward | https://medium.com/%40greylockvc/our-next-fund-and-our-path-forward-fd26dec852bb | migrated 2014-02-28; underlying Greylock XIV announcement 2013-09-10 | David Sze/Reid Hoffman coauthored firm-and-fund essay; reliability A for stated strategy, C for unsupported efficacy; access full; locators: paragraphs beginning “Today,” “We also focus,” and “Moving forward” | vehicle, stage, reserve, team-building, platform-service, network/marketplace, and technology-background theses; all claims belong jointly to Sze and Hoffman/Greylock
• ★★★ | Why Greylock is Committing $100M to Marketplaces | https://medium.com/%40greylockvc/why-greylock-is-committing-100m-to-marketplaces-d695d2300b86 | migrated 2014-02-28; underlying program date unresolved | David Sze/Reid Hoffman coauthored marketplace-strategy essay; reliability A for stated strategy, C for efficacy and unsupported market-size claims; access full; locators: opening commitment, Simon Rothman paragraphs, support-program paragraph, and sign-off | announced capital, named operator expertise, advisory/content/relationship support, and team/vehicle attribution; the page does not explain marketplace mechanics or prove efficacy
• ★★ | If, Why, and How Founders Should Hire a “Professional” CEO | https://medium.com/%40greylockvc/if-why-and-how-founders-should-hire-a-professional-ceo-b977a391eff9 | migrated 2014-02-28; underlying post date unresolved | Sze-signed governance reflection introducing a Reid Hoffman essay; reliability A for Sze’s stated view, C for the selected nine-company count; access full; locators: paragraphs beginning “Being one,” “Replacing a founder,” and company footnote | contingent founder/CEO succession rule, LinkedIn role evidence, and explicit rejection of a one-model-fits-all governance doctrine
• ★★ | Please Help Us Congratulate Dan Portillo! | https://medium.com/%40greylockvc/please-help-us-congratulate-dan-portillo-37b60c6a93e0 | migrated 2014-02-28; underlying promotion period 2013, exact original date unresolved | Reid Hoffman/David Sze coauthored talent-platform essay; reliability A for stated firm practice, C for unverified output counts; access full; locators: paragraphs beginning “Dan first joined,” “Dan has done,” and “Finally” | concrete portfolio-recruiting service design, internal role attribution, claimed output measures, and the firm’s stated leave-founders-better service test
• ★ | Welcome Ali Rosenthal | https://medium.com/%40greylockvc/welcome-ali-rosenthal-13b4f0de8646 | migrated 2014-02-28; original date unresolved | Sze-signed firm announcement; reliability A/C; access full | lower-value archive item on operator hiring and social/mobile distribution; inventoried outside the priority denominator
• ★ | Welcome Joseph Ansanelli | https://medium.com/%40greylockvc/welcome-joseph-ansanelli-efeabdd67072 | migrated 2014-02-28; original date unresolved | Sze-signed firm announcement; reliability A/C; access full | lower-value archive item on operator-investor selection and enterprise market conditions; inventoried outside the priority denominator
• ★ | Super Deal-of-the-Day for Groupon and Congratulations to David Thacker! | https://medium.com/%40greylockvc/super-deal-of-the-day-for-groupon-and-congratulations-to-david-thacker-d4bfa760b307 | migrated 2014-02-28; original date unresolved | Sze-signed personnel announcement; reliability A/C; access full | lower-value archive item allocating analyst/associate advocacy and deal leads; inventoried outside the priority denominator
• ★ | Welcome Kristin Richards and Elisa Schreiber | https://medium.com/%40greylockvc/welcome-kristin-richards-and-elisa-schreiber-to-greylock-partners-f9a73ca2fa07 | migrated 2014-02-28; original date unresolved | Greylock archive announcement attributed to Sze in the migrated corpus; reliability A/C; access route identified, content not independently material to the WRITINGS gate | lower-value recruiting announcement inventoried outside the priority denominator
• ★ | Our Investment in Lyra Health | https://greylock.com/blog/our-investment-in-lyra-health/ | 2015-10-15 | James Slavet-authored investment essay; reliability A for witness statement, C for promotion; access full; locator: opening founder-diligence paragraphs | documents Sze's long personal connection to David Ebersman but attributes the Lyra investment to Slavet/Venrock rather than Sze

Talks, interviews, podcasts, and video

• ★★ | Investors Target Web Video | https://www.tvweek.com/in-depth/2008/05/investors-target-web-video/ | 2008-05-12 | reported interview; reliability B; access full article/no recording; locator: Sze/Revision3 paragraphs | processed direct speech on distribution-cost change
• ★★★ | A Conversation with Greylock’s Reid Hoffman and David Sze | https://techcrunch.com/2010/04/06/mike-arrington-talks-to-reid-hoffman-and-david-sze-full-video/ | 2010-04-06 | full video/page transcript; reliability A; access full; locator: speaker turns/topics | fund cycle, price, check size, bottoms-up sourcing, mobile and Digg
• ★★★ | Twitter Was the Mistake on Our Wall of Shame | https://techcrunch.com/2010/09/27/twitter-investment/ | 2010-09-27 | report from Disrupt main-stage talk; reliability A/B; access full report | same appearance as 21:11 micro-fund video; explicit pass
• ★ | TechCrunch Disrupt agenda | https://techcrunch.com/2010/09/20/techcrunch-disrupt-next-week-heres-the-agenda-of-awesome/ | 2010-09-20 | contemporaneous publisher agenda; reliability A for event/date metadata; access full; locator: Monday schedule, David Sze/Reid Hoffman entry | fixes the 2010 main-stage appearance date against migrated video metadata; event metadata only, not content evidence
• ★★ | Reid Hoffman and David Sze on Their Micro VC Fund | https://techcrunch.com/video/reid-hoffman-and-david-sze-on-their-micro-vc-fund/ | recorded 2010-09-27; migrated metadata incorrectly says 2011-10-09 | YouTube P0tWOxmMoc4, 21:11; reliability A/B; no subtitles | same appearance as Twitter report; date controlled by contemporaneous agenda
• ★★ | Reid Hoffman and David Sze on Microlending to Startups | https://techcrunch.com/video/reid-hoffman-and-david-sze-on-microlending-to-startups/ | recorded 2010-09-27; migrated metadata conflicts | distinct YouTube asset https://www.youtube.com/watch?v=k1YoywR0Lko, 08:15; reliability A/B; no subtitles | backstage vehicle route processed to exact boundary
• ★ | TechCrunch Disrupt Backstage Pass, day one | https://techcrunch.com/2010/09/27/techcrunch-disrupt-the-backstage-pass-day-one-tctv/ | 2010-09-27 | contemporaneous publisher listing; reliability A for event/date/participant metadata; access full; locator: Reid Hoffman/David Sze listing | establishes the separate backstage appearance and date; metadata does not establish vehicle control or economics
• ★★ | Future of the Social Web | https://allthingsd.com/20101229/video-greylocks-reid-hoffman-and-david-sze-on-the-future-of-social/ | 2010-12-29 | 11-minute publisher video/report; reliability A/B; full attributed bullets/no current embed | physical-virtual integration, connection and work network effects
• ★★★ | How to Pick the Next Facebook When People Think You’re Crazy | https://www.forbes.com/sites/tomiogeron/2012/05/02/greylocks-david-sze-how-to-pick-the-next-facebook-when-people-think-youre-crazy/ | 2012-05-02 | Forbes video/full article; reliability A/B; full prose/partial audiovisual | one-level-down, Facebook persistence, Baidu pass
• ★★★ | Hackers Celebrated IPO, Now They’re Back to Work | https://www.forbes.com/sites/tomiogeron/2012/05/18/facebook-investor-david-sze-hackers-celebrated-ipo-now-theyre-back-to-work/ | 2012-05-18 | contemporaneous interview/report; reliability A/B; access full prose | operating focus, reported firm value and talent connection
• ★★ | David Sze on Squawk on the Street | https://medium.com/%40greylockvc/david-sze-on-squawk-on-the-street-jan-29-2013-3c05918055a2 | 2013-01-29 | migrated CNBC wrapper; reliability A/B/C; access partial | topic route processed without unsupported content assignment
• ★★★ | A Career of Company Building | https://semilshah.com/2013/10/02/a-career-of-company-building-text-of-my-discussion-with-david-sze/ | recorded 2013-02-05; transcript published 2013-10-02 | full transcript; reliability A | operating history, failures, Pandora, voting process and career advice
• ★ | In the Studio with David Sze | https://techcrunch.com/video/in-the-studio-with-greylock-partners-david-sze/ | recorded 2013-02-05; migrated 2023-06-05 | 33-minute video; reliability A; partial | same underlying work as Semil transcript; excluded from priority denominator
• ★ | In the Studio with Greylock Partners' David Sze article route | https://techcrunch.com/2013/04/18/david-sze-greylock/ | 2013-04-18 | TechCrunch article/video wrapper for the 2013-02-05 interview; reliability A/B for metadata; access full; locator: opening and embed | duplicate route for the Semil/TechCrunch interview; not independent corroboration
• ★★★ | What Makes the Best Entrepreneurs? | https://www.forbes.com/sites/tomiogeron/2013/05/08/what-makes-the-best-entrepreneurs-greylocks-david-sze-midas-list/ | 2013-05-08 | Forbes video/report; reliability A/B; HTTP 402; official YouTube asset https://www.youtube.com/watch?v=kvAcVSCSz4Q requires sign-in | exact access failure; no headline-derived model
• ★★ | Q&A: The $250MM Difference in Facebook’s Valuation | https://iamvictorio.us/post/53985025665/qa-with-david-sze-the-250mm-difference-in | 2013-06-26 | attendee recap; reliability B; full summary/no transcript | Matt Cohler sourcing lead remains uncorroborated
• ★★★ | Behind the Velvet Rope with Greylock Partners | https://techcrunch.com/video/behind-the-velvet-rope-with-greylock-partners/ | 2013-09-10 | YouTube asset https://www.youtube.com/watch?v=P9jIhPh3Q6Q, 26:41 plus report https://techcrunch.com/2013/09/10/greylock-disrupt-sf/; reliability A/B; no subtitles/full report | fund XIV and mixed-vote process
• ★★ | Greylock Partners on Raising a New $1 Billion Fund | https://techcrunch.com/video/greylock-partners-on-raising-a-new-1-billion-fund/ | 2013-09-10 | distinct YouTube asset https://www.youtube.com/watch?v=DRI-k5IDVqo, 11:54; reliability A/B; no subtitles | backstage fund route processed to exact boundary
• ★ | Greylock Partners Raises $1 Billion Fund XIV | https://techcrunch.com/2013/09/10/greylock-14/ | 2013-09-10 | contemporaneous publisher report quoting Sze; reliability A/B; access full; locator: paragraphs 8–9 | founder-service philosophy and founder decision-right boundary; not proof Greylock services caused outcomes
• ★★★ | Greylock Partners Raises Another Billion Dollars | https://allthingsd.com/20130910/greylock-partners-raises-another-billion-dollars/ | 2013-09-10; interview conducted prior week | reported direct interview; reliability A/B; access full article | LP continuity, West Coast team, early-stage count and current picks
• ★★★ | Facebook, LinkedIn Investor Greylock Partners Raises $1 Billion for 14th Fund | https://www.forbes.com/sites/tomiogeron/2013/09/10/facebook-linkedin-investor-greylock-partners-raises-1-billion-for-14th-fund/ | 2013-09-10 | reported direct interview; reliability A/B; access full article | early-stage concentration, services, marketplaces and mobile
• ★ | What it Takes to Be a Good Partner to Founders | https://medium.com/%40greylockvc/what-it-takes-to-be-a-good-partner-to-founders-cb7f7052f69b | migrated 2014-02-28; source date unresolved | one-quote wrapper; reliability A/C; partial | excluded from priority denominator
• ★ | David Sze on Investment Strategy | https://medium.com/%40greylockvc/david-sze-on-investment-strategy-8e0fcc54f3cb | migrated 2014-02-28; source date unresolved | one-minute wrapper; reliability A/C; dead embed | excluded from priority denominator
• ★★ | 12 Internet Players You’ll Want to Know About — David Sze profile | https://www.yalealumnimagazine.com/articles/3842-12-internet-players-youll-want-to-know-about?page=2 | Mar/Apr 2014 | short reported profile/interview; reliability B; access full | Excite formation, small-company preference, positive-search rule
• ★★ | Studio 1.0: David Sze and John Lilly | https://www.spreaker.com/episode/studio-1-0-david-sze-and-john-lilly-audio--61806888 | aired 2015-07-09; page 2015-11-06; listed 20:11/current feed 21:46; direct audio asset https://api.spreaker.com/v2/episodes/61806888/play.mp3 | Bloomberg audio; reliability A/A-B; OpenAI-transcribed with timestamp segments | partnership, Facebook dissent and Airbnb miss
• ★★★ | Building the Next-Gen Entertainment Platform with Roblox | https://music.amazon.co.uk/podcasts/9a49c1ea-53e7-4852-a8bd-a20c0714d603/episodes/54c54ce9-dd88-423f-a0f4-cc8707e579d5/greymatter-building-the-next-gen-entertainment-platform-with-roblox-greymatter | 2019-06-25; 2021-03-10 reissue at https://soundcloud.com/greylock-partners/roblox-next-generation-gaming-goes-public/s-S9RQOCRr2kU | 40-minute audio; reliability A/C; YouTube TQ4HbcWo2v8 private; Internet Archive long-shot https://web.archive.org/web/1oe_/http%3A//wayback-fakeurl.archive.org/yt/TQ4HbcWo2v8 timed out | one 2019 appearance; reissue duplicate; no bypass attempted
• ★★★ | 20VC with David Sze | https://www.thetwentyminutevc.com/david-sze | canonical RSS route https://thetwentyminutevc.libsyn.com/20vc-greylocks-david-sze-on-his-biggest-lessons-working-with-mark-zuckerberg-and-reid-hoffman-the-rise-of-pre-emptive-rounds-how-to-think-through-price-discipline-today-and-how-truly-strong-venture-partnerships-function; 2021-04-19 UTC; publisher page 2021-04-18; runtime 42:37 | audio/detailed agenda; reliability A/C; indexed transcript route https://www.tapesearch.com/episode/20vc-greylock-s-david-sze-on-his-biggest-lessons-working-with-mark-zuckerberg-and-reid-hoffman-the-rise-of-pre-emptive-rounds-how-to-think-through-price-discipline-today-and-how-truly-strong-venture-partnerships-function/kwcdELUUN7koYzbhcU9i3Q was Cloudflare-blocked | exact answers not content-analyzed without retained transcript
• ★★★ | David Sze: Nextdoor Reflections | https://greylock.com/blog/congrats-nextdoor/ | 2021-11-08 | Greymatter audio/full embedded transcript; reliability A/C | pass/pivot, local network, quality growth, CEO search, privacy and moderation
• ★★ | 20VC Ultimate Guide to Price Sensitivity | https://www.thetwentyminutevc.com/price-sensitivity | canonical RSS 2022-08-12; publisher page 2022-08-11 | 28:03 compilation; reliability A/C; partial text | duplicate of 2021 material, not independent evidence

Filings, institutional records, and deal data

• ★★★ | David Sze, Yale trustee biography | https://www.yale.edu/board-trustees/current-trustees/david-sze | current; accessed 2026-08-23 | official institutional biography; reliability A/C; access full; locator: full biography | best single title, education, senior-managing-partner, Facebook observer, LinkedIn/Pandora/Nextdoor/Discord, and public-service chronology
• ★★ | The Yale Endowment 2015 | https://zoo.cs.yale.edu/classes/cs458/lectures/old2016/Yale_Endowment_15.pdf | 2015 | official Yale Investments Office annual report mirror; reliability A/C for biography and selected portfolio, not complete performance; access full PDF; locator: printed p. 9, “David Sze” | contemporary institutional profile, direct quotation, and selected investments; use as a bounded list rather than a career denominator
• ★★ | Venture capitalist David Sze named to Rockefeller board | https://www.rockefeller.edu/news/18178-venture-capitalist-david-sze-named-to-board | published 2012-12-14; elected 2012-06-06 | official institutional release; reliability A/C; access full; locator: opening appointment paragraph | exact election date and contemporaneous career snapshot
• ★★★ | LinkedIn S-1/A | https://www.sec.gov/Archives/edgar/data/1271024/000119312511064249/ds1a.htm | 2011-03-11 | SEC issuer filing; reliability A; access full; locators: Management—David Sze, printed p. 89; Principal Stockholders, printed pp. 115–117, footnote 3 | September 2004 board start, biography, pre-IPO vehicle holdings, and shared voting/dispositive-control language [contemporaneous record · strong]
• ★★★ | LinkedIn Form 3 for David L. Sze | https://www.sec.gov/Archives/edgar/data/1339221/000118143111030831/xslF345X03/rrd311779.xml | 2011-05-18 | SEC insider filing; reliability A; access full; locator: Tables I–II and footnotes | Greylock XI/XII vehicle-control map; establishes indirect reporting, not personal economics
• ★★★ | Facebook S-1/A | https://www.sec.gov/Archives/edgar/data/1326801/000119312512232582/d287954ds1a.htm | 2012-05-16 | SEC issuer filing; reliability A; access full; locators: Principal and Selling Stockholders, Greylock row and footnotes 27–28 | Greylock vehicle stake and IPO sale baseline plus observer-role controls; a holding value is not Sze personal return [contemporaneous record · strong]
• ★★ | Curtains for the Observer Role on the Facebook Board | https://allthingsd.com/20080725/curtains-for-the-observer-role-on-the-facebook-board/ | 2008-07-25 | contemporaneous reported governance account; reliability B; access full; locator: David Sze/Paul Madera observer paragraphs | corroborates Sze non-voting observer status and separates observation rights from formal director control
• ★★★ | Pandora S-1/A | https://www.sec.gov/Archives/edgar/data/1230276/000119312511162497/ds1a.htm | 2011-06-10 | SEC issuer filing; reliability A; access full; locators: Management—David Sze, printed p. 94; Certain Relationships—Series F and investor offer; Principal Stockholders, footnote 5 | May 2009 board start, Series F as first Greylock investment, vehicle breakdown, risks, and pre-IPO ownership [contemporaneous record · strong]
• ★★★ | Pandora 2012 proxy | https://www.sec.gov/Archives/edgar/data/1230276/000119312512181990/d307982ddef14a.htm | 2012-04-25 | SEC issuer filing; reliability A; access full; locator: Security Ownership, printed p. 17, Greylock row and footnote | 13.03% Greylock-affiliated beneficial-owner row and shared-control boundary [contemporaneous record · strong]
• ★★ | Pandora 2014 proxy | https://www.sec.gov/Archives/edgar/data/1230276/000104746914004048/a2219637zdef14a.htm | 2014-04-28 | SEC issuer filing; reliability A; access full; locator: Proposal No. 1—David Sze | continuing board biography and role as of 2014; useful for board-tenure endpoint research
• ★★★ | Nextdoor announces $18.6 million financing | https://about.nextdoor.com/press-releases/nextdoor-announces-18-6-million-in-funding-from-benchmark-capital-dag-ventures-greylock-partners-and-shasta-ventures | 2012-07-24 | contemporaneous issuer release with Sze statement; reliability A/C; access full; locator: announcement and Sze paragraph | round participants, pre-Series-B adoption baseline, and specialized-network thesis
• ★★★ | Nextdoor post-combination prospectus | https://www.sec.gov/Archives/edgar/data/1846069/000162828021024323/nextdoorinc424.htm | 2021-11-05 | SEC issuer filing; reliability A; access full; locators: Management—David Sze, Principal Stockholders, Risk Factors | November 2021 public-company board start, role biography, business/competition baseline, and holder structure
• ★★★ | Nextdoor 424B3 | https://www.sec.gov/Archives/edgar/data/1846069/000162828022006792/nextdoorinc424b3.htm | 2022-03-18 | SEC issuer filing; reliability A; access full; locators: Principal Securityholders and footnotes 5/10 | Greylock IX/XIII/XIV/Discovery entity breakdown and 21,196,977-share shared-control row
• ★★ | Nextdoor 2024 proxy | https://www.sec.gov/Archives/edgar/data/1846069/000184606924000086/kvsb-20240416.htm | 2024-04-16 | SEC issuer filing; reliability A; access full; locator: Proposal No. 1—David Sze | current board/committee biography and governance baseline
• ★★ | Nextdoor 2025 proxy | https://s28.q4cdn.com/517578190/files/doc_downloads/2025/04/80ce2442-18d2-4c7a-afe3-b15ca99cd599.pdf | 2025-04 | issuer-hosted proxy PDF; reliability A; access full; locator: printed pp. 8–9, Proposal No. 1 | continuing board role, age/title snapshot, and director qualification
• ★★★ | Nextdoor 2025 Form 10-K | https://www.sec.gov/Archives/edgar/data/1846069/000184606926000023/kind-20251231.htm | 2026-03 filing for year ended 2025-12-31 | SEC issuer filing; reliability A; access full; locators: Overview, key business metrics, Risk Factors | current weekly-active-user, ARPU, monetization and moderation control; not evidence of Sze-specific causation
• ★★ | Nextdoor Form 4 for David L. Sze | https://www.sec.gov/Archives/edgar/data/1339221/000120919122034927/xslF345X03/doc4.xml | 2022-06-07 | SEC insider filing; reliability A; access full; locator: transaction table and footnotes | security-level vehicle holding/transaction route; Greylock 16 shares are not Sze-direct ownership
• ★★★ | Roblox S-1 | https://www.sec.gov/Archives/edgar/data/1315098/000119312520298230/d87104ds1.htm | 2020-11-19 | SEC issuer filing; reliability A; access full; locators: Business, Risk Factors, Principal Stockholders | independent company, market, governance, and ownership baseline for testing the later-published 2018 memo
• ★★★ | Roblox 2026 proxy | https://www.sec.gov/Archives/edgar/data/1315098/000110465926044362/rblx-20260528xdef14a.htm | 2026-05-28 | SEC issuer filing; reliability A; access full; locator: Founder and CEO letter | current issuer-reported daily-active-user and engagement context; promotional letter, not an audited Sze return record
• ★★★ | Palo Alto Networks 2022 proxy | https://www.sec.gov/Archives/edgar/data/1327567/000120677422002637/panw4099631-def14a.htm | filed 2022-11-03 | SEC issuer filing; reliability A; access full; locator: Proposal No. 1, Non-Continuing Directors—Asheem Chandna, printed pp. 50–51 | hard attribution control: Chandna served on the board from April 2005 and is the documented Greylock enterprise/security investor; PANW must not be attributed to Sze [contemporaneous record · strong]
• ★★ | Greylock XIV Limited Partnership Form D | https://www.sec.gov/Archives/edgar/data/1584740/000158474013000001/0001584740-13-000001-index.htm | 2013-09-10 | SEC exempt-offering filing; reliability A; access full; locator: filing detail/primary document | legal vehicle/date/capital-offering route; not fund performance or Sze economics
• ★★ | Greylock Capital Management IAPD route | https://adviserinfo.sec.gov/firm/summary/161175 | current; accessed 2026-08-23 | SEC/IAPD regulatory route; reliability A; access full route; locator: firm summary and filings | adviser/entity-control and disclosure search; does not supply deal-level DPI/TVPI/IRR
• ★★★ | Rockefeller University Form 990, related-party disclosure | https://docs.candid.org/990/131/131624158/131624158_2024_202530449349301203_990.pdf | filing year 2024 | IRS return mirror; reliability A; access full; locator: Schedule L, Part IV–V | discloses endowment investments in partnerships connected to trustee Sze and states he never served on the investment committee; governance boundary, not misconduct

Press, books, and historical context

• ★★★ | Late-blooming Greylock thrives in venture capital | https://www.sfgate.com/tech/article/late-blooming-greylock-thrives-in-venture-capital-4690152.php | 2013-07-28 | long-form institutional profile; reliability B; access full; locator: West Coast transition, Facebook, Pandora sections | independent Greylock history, early losses, Bhusri/Sze roles, relocation, and reported Facebook/Pandora checks
• ★★★ | Facebook investor David Sze: Hackers celebrated IPO, now they're back to work | https://www.forbes.com/sites/tomiogeron/2012/05/18/facebook-investor-david-sze-hackers-celebrated-ipo-now-theyre-back-to-work/ | 2012-05-18 | contemporaneous interview/report; reliability A/B; access partial/full page text; locator: opening and Greylock stake paragraphs | Facebook decision recollection and reported $12.5 million-to-$1.5 billion firm-stake value; not realized or personal return
• ★★★ | Facebook bet paid off for Greylock Partners | https://www.bostonglobe.com/business/2012/05/18/facebook-bet-paid-off-for-greylock-partners/uku3rxf6CWsADw2qZMDdbP/story.html | 2012-05-18 | contemporaneous report; reliability B; access partial/paywalled; locator: opening and IPO-sale paragraphs | independently reported firm check, stake value, partial IPO sale, and Sze attribution with explicit access boundary
• ★★ | The Midas Top 10 | https://www.forbes.com/2011/04/06/breyer-thiel-andreessen-moritz-top-10_slide.html | 2011-04-06 | ranked press profile; reliability B/C; access full; locator: No. 9 David Sze | contemporary selected deal list and external reputation; ranking is not an audited return measure
• ★★ | David Sze Forbes profile and linked interview archive | https://www.forbes.com/profile/david-sze/ | last updated 2017-04-17 | press profile/archive; reliability B/C; access partial; locator: profile and linked 2012–2013 items | Midas history, selected portfolio, role snapshots, Facebook founder/selection interviews, and retrieval leads
• ★★ | Greylock Preaches Continuity as It Raises Another $1 Billion Fund | https://www.forbes.com/sites/alexkonrad/2016/10/11/greylock-raises-1-billion-fifteenth-fund/ | 2016-10-11 | reported firm/fund profile; reliability B; access full; locator: succession and fund-strategy paragraphs | senior-managing transition, partnership design, investment cadence, and vehicle scale
• ★★ | Greylock announces fresh $1 billion fund | https://venturebeat.com/ai/greylock-partners-announces-fresh-1-billion-fund | 2013-09-10 | contemporaneous conference/fund report; reliability A/B; access full; locator: Sze/Hoffman remarks | fund XIV, 84% seed/A claim, follow-on reserve logic, and mixed-vote result claim
• ★★★ | Nextdoor closes $21.6 million Series B led by Greylock | https://techcrunch.com/2013/02/12/nextdoor-closes-21-6-million-in-new-series-b-funding-to-take-its-neighborhood-focused-social-network-global/ | 2013-02-12 | contemporaneous financing report; reliability B; access full; locator: round, Greylock amount, and board paragraphs | reported roughly $15 million Greylock contribution, Sze board appointment, and his then-largest check
• ★★ | Googlespawn | https://www.forbes.com/forbes/2006/0410/050.html | 2006-04-10 | contemporaneous Web 2.0 feature; reliability A/B; access full; locator: Sze and Digg paragraphs | period consumer-startup volume, small-check context, and Digg sourcing before the outcome
• ★★ | Kleiner Perkins Bets the Farm | https://fortune.com/2008/07/21/kleiner-bets-the-farm/ | 2008-07-21 | cycle/strategy feature; reliability A/B; access partial/paywalled; locator: Web 2.0 versus cleantech paragraphs | Sze's contemporaneous Web 2.0 return confidence and concentrated consumer exposure before later exits
• ★★ | Venture Firm Greylock Partners Outshines Silicon Valley Rivals | https://www.bloomberg.com/news/articles/2013-07-25/venture-firm-greylock-partners-outshines-silicon-valley-rivals | 2013-07-25 | long-form institutional profile; reliability B; access partial/paywalled | independent firm history, West Coast transition, reported fund/deal outcomes, and team-credit leads
• ★★ | Twelve Internet Players You'll Want to Know About | https://www.veeva.com/wp-content/uploads/2014/03/12-Internet-players-you%E2%80%99ll-want-to-know-about-Features-Yale-Alumni-Magazine.pdf | captured 2015-09-18; original issue/date to verify | alumni feature PDF; reliability A/B; access full; locator: PDF page containing David Sze | short direct statements on Excite, garage-stage companies, Facebook skepticism, Pandora, and focusing on why a venture can succeed
• ★★★ | The 2016 Midas Brink List: Josh Elman | https://www.forbes.com/sites/truebridge/2016/03/23/2016-midas-brink-list/ | 2016-03-23 | contemporaneous investor profile; reliability B; access full; locator: Josh Elman section | says Elman led Greylock's Discord investment, creating a material attribution tension with Sze's board-announcement post that later phases must resolve
• ★★★ | Musical.ly raising $100 million at a $500 million valuation | https://techcrunch.com/2016/05/06/musical-ly-raising-100-million-at-500-million-valuation-for-social-music-videos/ | 2016-05-06 | contemporaneous financing report with Elman statement; reliability A/B; access full; locator: prior-round and Elman paragraphs | credits Greylock's musical.ly work to Josh Elman and provides a check against the current firm page's Sze/Elman joint attribution
• ★★ | The Facebook Effect by David Kirkpatrick | https://www.simonandschuster.com/books/The-Facebook-Effect/David-Kirkpatrick/9781439102121 | 2010 | book; reliability B; access partial, publisher metadata/excerpts only in SCOUT | reconstructs Facebook's founding and investor network; relevant chapters must be accessed legitimately and checked against filings
• ★★★ | The Power Law by Sebastian Mallaby | https://www.penguinrandomhouse.com/books/576124/the-power-law-by-sebastian-mallaby/ | 2022-02-01 | reported history/book; reliability B; access partial, publisher metadata/previews only in SCOUT | comparative venture history, power-law/firm context, and interview trail; locate Greylock/Sze passages before use
• ★★ | AI Valley by Gary Rivlin | https://www.harpercollins.com/products/ai-valley-gary-rivlin | 2025-03-25 | reported history/book; reliability B; access partial, publisher metadata/searchable excerpts only | recent Greylock institutional history, early Bhusri/Sze losses, LinkedIn/Facebook dissent, and reported firm economics requiring source checks
• ★★★ | Masters of Scale: Tristan Walker transcript | https://mastersofscale.com/wp-content/uploads/2018/09/mos-episode-transcript_-tristan-walker.pdf | 2018-09 | podcast transcript; reliability A for Reid Hoffman recollection, B for narration; access full; locator: PDF search “David Sze” | Hoffman witness evidence on Sze's Airbnb objection and a reported Greylock-return claim that needs vehicle/cash-flow corroboration
• ★★ | FRED Nasdaq Composite series | https://fred.stlouisfed.org/series/NASDAQCOM | historical daily series; accessed 2026-08-23 | official Federal Reserve data route; reliability A; access full | reproducible dot-com boom/crash context for Sze's 2000 Greylock entry; later analysis must select decision-date windows explicitly
• ★★ | Pew: Social Media Usage 2005–2015 | https://www.pewresearch.org/internet/2015/10/08/social-networking-usage-2005-2015/ | 2015-10-08 | survey synthesis; reliability A/B; access full; locators: adoption charts/methodology | later adoption baseline for separating what was knowable at LinkedIn/Facebook decision dates from hindsight
• ★★★ | Copyright Royalty Board Web II proceeding | https://www.crb.gov/rate/05-1/ | 2005–2009 proceeding archive; accessed 2026-08-23 | official regulatory docket; reliability A; access full route; locator: docket documents/orders | primary route for Pandora's royalty-rate uncertainty and 2009 financing context
• ★★★ | FTC COPPA Final Rule Amendments | https://www.ftc.gov/legal-library/browse/federal-register-notices/16-cfr-part-312-coppa-final-rule-amendments | 2025-04-22 | official U.S. regulator rule page; reliability A; access full; locator: final-rule summary | dated modern regulatory premise for childrens platforms; not evidence of Roblox compliance or Sze present practice
• ★★★ | European Commission DSA protection-of-minors guidelines | https://digital-strategy.ec.europa.eu/en/library/commission-publishes-guidelines-protection-minors | 2025-07-14 | official European Commission guidance; reliability A; access full; locator: guidelines summary | dated modern regulatory premise for platform safety; not historical evidence about the 2018 Roblox decision

Skeptical, failure, adverse, and archive routes

• ★★★ | Digg sold in parts after raising a reported $45 million | https://techcrunch.com/2012/07/12/digg-sold-to-linkedin-and-the-washington-post-and-betaworks/ | 2012-07-12 | contemporaneous exit report; reliability B; access full; locator: opening and financing paragraph | adverse outcome, disputed $0.5 million versus roughly $16 million aggregate consideration, and denominator evidence; no exact Greylock cash-flow allocation
• ★★ | After Raising $45M, Digg Sells for Pennies on the Dollar | https://www.wired.com/2012/07/digg-sells-for-pennies/ | 2012-07-12 | contemporaneous adverse report; reliability B; access full; locator: opening | independent check on the failed-outcome framing; sale figures remain reported and disputed
• ★★ | Nextdoor CEO pledges to address racial profiling and censorship | https://www.axios.com/2020/07/02/nextdoor-ceo-censorship-racial-profiling | 2020-07-02 | reported product/governance criticism; reliability B; access full | tests costs and governance failure modes of hyperlocal trust; does not by itself establish Sze's personal action or inaction Adverse discovery was exhausted on 2026-08-23 through name plus enforcement/fraud terms, exact-name litigation terms, and loss/miss/write-off terms. The routes found no substantiated personal enforcement/fraud matter, rejected unrelated same-name results, and surfaced the documented Twitter, Digg, and early-enterprise leads but no complete loss ledger. Search absence is not proof. [researcher inference · provisional]
• ★ | SEC EDGAR full-text search for “David Sze” | https://efts.sec.gov/LATEST/search-index?q=%22David%20Sze%22 | searched 2026-08-23 | official filing-search route; reliability A for filings, search coverage imperfect; access full route | enumerate issuer, insider, fund, nonprofit, and ownership filings while rejecting unrelated names
• ★★ | Greylock/David Sze Wayback CDX route | https://web.archive.org/cdx/search/cdx?url=greylock.com%2Fteam%2Fdavid-sze%2F*&output=json&filter=statuscode%3A200 | attempted 2026-08-23 | web-archive index; reliability A for captured pages, C for capture completeness; access route intermittent | recover historical profile versions, former investments, title changes, and deleted writing links; snapshot dates must be recorded when used
• ★ | Computer History Museum catalog search for David Sze | https://www.computerhistory.org/collections/search/?q=David%20Sze | attempted 2026-08-23 | archive/catalog route; reliability A for catalog metadata; access full route; no Sze oral history identified | negative direct-voice route; recheck Excite/@Home and Greylock collections by company/colleague names
• ★ | Stanford SearchWorks search for “David Sze” | https://searchworks.stanford.edu/?q=%22David%20Sze%22 | attempted 2026-08-23 | library/catalog route; reliability A for records; access full route; no named personal collection identified | search Silicon Valley Archives, GSB recordings, Excite/@Home records, and event video; catalog record/call number required if found
• ★ | Yale Archives search for “David Sze” | https://archives.yale.edu/search?q=%22David%20Sze%22 | attempted 2026-08-23 | archive/finding-aid route; reliability A for catalog metadata; no named personal papers identified | trustee/alumni context and possible Digital Yale records; no contact or restricted-access request authorized
• ★ | 25iq search for David Sze | https://www.google.com/search?q=site%3A25iq.com+%22David+Sze%22 | searched 2026-08-23 | secondary meta-source route; reliability C; access full results; no dedicated profile found | required methodology sweep completed; do not substitute quote compilations for original sources
• ★ | Podcast venue sweep: Colossus, Acquired, Tim Ferriss, Knowledge Project, and Stanford eCorner | https://www.google.com/search?q=%22David+Sze%22+%28site%3Ajoincolossus.com+OR+site%3Aacquired.fm+OR+site%3Atim.blog+OR+site%3Afs.blog+OR+site%3Aecorner.stanford.edu%29 | searched 2026-08-23 | discovery query; reliability C, search route only; access full results | no dedicated Sze appearance found; 20VC, TechCrunch, AllThingsD, and Greylock remain the priority spoken corpus

Additional DEALS evidence

• ★★ | Red Hat fiscal 2003 Form 10-K / annual report | https://media.corporate-ir.net/media_files/NSD/RHAT/reports/rhat_051804.pdf | fiscal 2003 | issuer annual report; reliability A; access full; locator: business-combination/acquisition note, “NOCpulse” | acquisition consideration baseline; company value is not Greylock proceeds

• ★★ | EarthLink 2007 Form 10-K | https://www.sec.gov/Archives/edgar/data/1102541/000104746908001943/a2183055z10-k.htm | 2008-02-29 filing | SEC issuer filing; reliability A; access full; locator: New Edge acquisition paragraphs | $144.8 million purchase-price composition; not a Greylock return

• ★★★ | QVC filing response — Oodle asset acquisition | https://investors.qvcgrp.com/investors/sec-filings/pre-split-liberty-interactive/content/0001104659-12-085975/filename1.htm | 2012-12-21 | issuer filing correspondence; reliability A; access full; locator: Oodle response | signed agreement to acquire substantially all assets, expected closing by 2012-12-31, and less-than-1% materiality boundary; no Greylock proceeds

• ★★★ | Viacom fiscal 2018 results — WhoSay acquisition | https://ir.paramount.com/static-files/031e9319-13b0-4c19-9937-286aa1fd234a | 2018-02-08 | issuer results release; reliability A/C; access full; locator: acquisition references | confirms WhoSay acquisition; grouped company economics do not isolate WhoSay or Greylock

• ★★★ | SiriusXM completes acquisition of Pandora | https://investor.siriusxm.com/news-events/press-releases/detail/1084/siriusxm-completes-acquisition-of-pandora | 2019-02-01 | acquirer release; reliability A/C; access full; locator: opening and exchange-ratio paragraph | confirms completion and 1.44-share exchange ratio; does not reveal Greylock distributions

• ★★ | Cisco completes acquisition of OpenDNS | https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2015/m08/cisco-completes-acquisition-of-opendns.html | 2015-08-27 | acquirer release; reliability A/C; access full; locator: terms paragraph | $635 million cash/equity-award transaction baseline; no Greylock allocation

• ★★★ | Path announces $30 million-plus round | https://techcrunch.com/2012/04/16/path-announces-30m-round-of-funding/ | 2012-04-16 | contemporaneous financing report; reliability B; access full; locator: announcement and investor list | explicitly pairs Greylock Partners/David Sze and identifies Redpoint-led round

• ★★ | Path raising more than $30 million | https://venturebeat.com/mobile/path-raising-40m-round-led-by-redpoint-no-it-will-not-be-the-next-instagram | 2012-04-16 | contemporaneous financing/adverse report; reliability B; access full; locator: round, investor, and activity-quality paragraphs | ex-ante privacy/mobile thesis from Redpoint and warning that registrations may not equal active use

• ★★ | Path seeks Series C amid reported usage decline | https://techcrunch.com/2013/07/16/path-is-raising-50m-at-a-500m-valuation/ | 2013-07-16 | contemporaneous financing report; reliability B; access full; locator: valuation, prior round, and U.S. iPhone-use paragraphs | warning signal and follow-on route; does not disclose Sze's response

• ★★ | Kakao acquires Path and Path Talk assets | https://www.kakaocorp.com/page/detail/7753 | 2015-05-29 | acquirer release in Korean; reliability A/C; access full; locator: opening acquisition announcement | primary asset-sale confirmation; consideration not disclosed on page

• ★★ | Photo-Sharing App Path Sells to Korean Internet Firm | https://time.com/3901106/path-kakao-korean-photo-sharing-buy-dave-morin-google/ | 2015-05-29 | Fortune report hosted by Time; reliability B; access full; locator: financing and undisclosed-terms paragraphs | reported ~$77 million total financing and explicit terms boundary

• ★★ | Mobile social network Path is closing down | https://techcrunch.com/2018/09/17/rip-path/ | 2018-09-17 | contemporaneous shutdown report; reliability B; access full; locator: opening and closure dates | final adverse product outcome; no investor cash-flow allocation

• ★★★ | Verizon acquires Vessel and will shut service | https://techcrunch.com/2016/10/26/verizon-acquires-vessel/ | 2016-10-26 | contemporaneous acquisition report; reliability B; access full; locator: acquisition, capital raised, and shutdown paragraphs | adverse outcome and reported >$130 million financing; terms/returns unknown

• ★ | Dealroom David Sze profile | https://dealroom.co/people/david_sze/ | current; accessed 2026-08-23 | third-party role database; reliability C; access full | discovery-only TheFind/Kongregate/Synapse/ambiguous NewCo leads; Wink was promoted only after the preserved historical Greylock-bio quote

• ★ | NFX Signal David Sze profile | https://signal.nfx.com/investors/david-sze | current; accessed 2026-08-23 | third-party investor database; reliability C; access full | discovery-only portfolio/board leads; excluded absent official, filing, or credible contemporaneous corroboration

• ★★★ | Microsoft to acquire LinkedIn | https://news.microsoft.com/source/2016/06/13/microsoft-to-acquire-linkedin/ | 2016-06-13 | acquirer/issuer announcement; reliability A/C; access full; locator: opening terms paragraph | $26.2 billion all-cash value inclusive of LinkedIn net cash; company value, not Greylock proceeds

• ★★ | Medium adds to board and bank account | https://medium.com/blog/medium-adds-to-board-bank-account-ada7f0059b9b | 2015-09-29 | issuer-authored board/financing update; reliability A/C; access full; locator: final side note | direct record that Sze stepped down from Medium's board; no reason or economics disclosed

• ★★ | Discovery acquires Revision3 | https://techcrunch.com/2012/05/03/discovery-channel-acquires-revision3/ | 2012-05-03 | contemporaneous acquisition report; reliability B; access full; locator: terms/capital paragraphs | terms undisclosed; source estimate near $30 million after roughly $10 million raised; not Greylock proceeds

• ★★ | Pinterest acquires Jelly | https://www.axios.com/2017/12/15/pinterest-acquires-qa-app-from-twitter-co-founder-1513300825 | 2017-03-08 | contemporaneous acquisition report; reliability B; access full; locator: transaction characterization | source called it small/all-stock; issuer confirmation and Greylock allocation unavailable

• ★★ | Preserved 2007 Greylock biography quotation | https://sourcewatch.org/index.php?title=David_Sze | accessed/edited 2007-07-31 | secondary preservation of then-Greylock bio; reliability B/C; access full; original page/Wayback capture not recovered | expressly lists Glimpse, Wink, Troux and Claria; provisional inclusion with archive caveat

• ★★ | Troux $10 million Series B | https://www.gaebler.com/VC-Funding-8DAB8848-2438-4F91-AC55-4EB875451B54-Troux-Technologies-09-15-2003 | 2003-09-15 | financing database; reliability C; access full; locator: round/investor rows | Greylock lead, co-investors and Sze attribution; terms/economics unavailable

• ★ | David Sze positions and network | https://www.marketscreener.com/insider/DAVID-SZE-A04LT7/ | current; accessed 2026-08-23 | third-party role database; reliability C; access full; locator: former positions | Troux 2003–06 and Glimpse 2004–07 directorship dates; role evidence, not terms

• ★★ | If You Can raises $6.5 million Series A | https://www.finsmes.com/2014/02/if-you-can-company-raises-6-5m-in-series-a-funding.html | 2014-02-26 | contemporaneous financing report; reliability B; access full | Greylock-led round and Sze board appointment; exact allocation/economics unavailable

• ★★★ | Complete history of Treehouse | https://www.ryancarson.com/articles/ryan-carson-treehouse | 2024-03-04 | founder retrospective; reliability A/B; access full; locators: 2011–12 funding, 2016–21 outcome | names Sze in $600,000 angel syndicate, separates later Greylock Discovery A, documents decline/sale

• ★★★ | Latham advises Lumi Story AI seed | https://www.lw.com/en/news/latham-watkins-advises-lumi-story-ai-in-seed-funding | 2024-08-01 | company-counsel notice; reliability A/B; access full; locator: financing paragraph | $4 million seed and Sze named among angels; personal check/security/economics unavailable

Coverage assessment

Guiding questions

The sweep was framed around eight questions before deep searching: (1) which operating experiences actually preceded Sze's consumer judgment; (2) how Greylock's Boston-to-Silicon-Valley transition and vehicles shaped his opportunity set; (3) who performed each deal function rather than who narrated it later; (4) what Sze said before major outcomes versus afterward; (5) what the complete investment, pass, loss, and write-off denominator contains; (6) how the record changes without Facebook and under vehicle/team attribution; (7) what credible adverse, legal, governance, and company-level criticism complicates the success story; and (8) which period-specific market, regulatory, and technological conditions made the strategy possible.

Inventory and source-category coverage

This map contains 134 source-map rows; item-level notes identify reliability, access, and duplicate status. It covers direct writings and firm-authored records; full and partial transcripts; podcasts and video; official biographies; SEC issuer, fund, insider, and ownership filings; a nonprofit related-party filing; books and long-form press; deal and failure reports; market, adoption, royalty-regulation, and current platform-safety context; adverse searches; web archives; oral-history/library catalogs; and required venue/meta-source sweeps. Duplicate underlying works are marked: the TechCrunch video is the audiovisual original of the Semil transcript, and the 2022 20VC compilation may reuse the 2021 interview.

Direct voice is concentrated rather than voluminous [researcher inference · strong]: the 2013 Semil transcript, 2010 TechCrunch joint transcript, 2021 20VC audio, 2021 Nextdoor transcript, 2016 Discord post, 2018/2021 Roblox posts and memo summary, and short contemporaneous LinkedIn/Facebook/Nextdoor statements. No dedicated Sze book, sustained personal blog beyond the two-post personal Medium account, public LP letter, complete shareholder letter, testimony, or named personal oral history was found. Greylock's migrated archive is author-filterable but may omit old pages; Wayback recovery remains necessary for stable historical versions.

Names, dates, archives, and access boundaries

Identity variants searched were David Sze, David L. Sze, David Sze Greylock, and combinations with interview, transcript, podcast, video, essay, memo, filetype:pdf, oral history, lawsuit, complaint, fraud, SEC, loss, miss, write-off, and company names. Birth-date claims from aggregators were not promoted into the chronology; official sources establish education years and age snapshots but SCOUT did not locate a primary birth record. Career date conflicts were preserved: Greylock confirms a 2000 join, Yale supplies 2012–2017 senior-managing-partner service, and the current advisory title has no verified transition date.

The Computer History Museum, Stanford SearchWorks/Silicon Valley Archives, Yale Archives, SEC full-text search, IAPD, Greylock/Wayback, 25iq, Colossus, Acquired, Tim Ferriss, Knowledge Project, and Stanford eCorner routes were attempted. No named Sze papers or oral history appeared in the public catalog results. Spreaker supplies the full 2015 Studio 1.0 audio; Boston Globe, Fortune, book, and some other video/audio routes are partial or paywalled; none was bypassed. Publicly accessible audio/captions should be processed in TALKS, with timestamps. No archive or person should be contacted without separate authorization.

Denominator, vehicle, team, and adverse controls

Candidate denominator sources are Greylock's current Sze profile, consumer page, complete portfolio page, preserved historical-profile evidence, LinkedIn/Facebook/Pandora/Nextdoor/Roblox filings, financing records used with explicit reliability limits, founder/company-counsel records, and contemporaneous press. DEALS reconciles 33 public relationships: 31 Greylock-attributed companies and two separately sourced personal angels (Treehouse and Lumi Story). The recovered Greylock set includes Glimpse, Wink, Troux, Claria, and If You Can; aliases/follow-ons remain deduplicated. This selected set still omits unnamed early enterprise losses, cannot enumerate every pass/write-off, and cannot support a career coverage fraction.

Vehicle controls must cover Greylock XI/XII/XII-A, XIII/XIII-A, XIV and affiliated discovery/management entities wherever the filings name them. Sze's reported voting/dispositive power is shared control, not direct ownership; an announced round is not a Greylock-only check; a market value is not a realized return; and a fund result is not Sze's personal economics. [researcher inference · strong] Team controls are especially important for Facebook (founder, Accel/Jim Breyer, Meritech/Paul Madera, Peter Thiel), LinkedIn (Reid Hoffman and cofounders), Discord (Josh Elman), Roblox (Seth Rosenberg), musical.ly (Elman), and Greylock's West Coast institution (Aneel Bhusri and the partnership). Palo Alto Networks belongs to Asheem Chandna's documented firm attribution and must not be attributed to Sze merely because Sze used it in a list of Greylock companies. [researcher inference · strong] Greylock portfolio, Palo Alto Networks entry

Adverse searches found no substantiated personal enforcement or fraud matter. They did identify a reported adverse Digg outcome, Sze's admitted Twitter pass, early enterprise-investing difficulty, Airbnb dissent, and product/governance criticism around Nextdoor. Company criticism is not evidence of personal wrongdoing or of a specific director act [researcher inference · moderate]; later phases should search board records, litigation, moderation/safety reports, write-offs, and founder/employee accounts with that boundary intact.

Historical context and completed-peer comparisons

The initial context set [researcher inference · moderate] covers the dot-com reversal through FRED Nasdaq data; social-network adoption through Pew; contemporaneous LinkedIn and Facebook issuer releases; Pandora's Web II royalty-rate docket and filing risks; and Roblox's later issuer risks against the dated 2018 memo. Later DEALS work must add decision-date financing alternatives, comparable-company valuations, mobile/broadband adoption, privacy and platform regulation, gaming/user-generated-content economics, and period venture base rates without hindsight leakage.

The closest eligible completed peers [researcher inference · moderate] are Jim Breyer for co-investment and formal-board contrast at Facebook, Fred Wilson for consumer-network thesis and high-public-writing contrast, and Bill Gurley for network/marketplace underwriting, price discipline, and partnership/governance comparison. Reid Hoffman is the closest Greylock teammate but is ineligible as a canon comparator because investor 060 remains queued in the investor index.

The completed-peer comparison uses five underlying controls for framework parity, not as David Sze evidence: Jim Breyer's contemporaneous Forbes profile; Union Square Ventures' formation thesis; Fred Wilson's venture-fund arithmetic; Bill Gurley's marketplace framework; and Gurley's career interview. They calibrate attribution, public-thesis, fund-math, marketplace, and governance questions; they do not establish Sze decisions or economics. [researcher inference · moderate]

Highest-value retrieval leads

1. Full, legitimately accessible audio/captions with timestamps for the 2021 20VC, 2010 AllThingsD, 2010 micro-fund, 2013 Disrupt, 2015 Studio 1.0, and Greylock Roblox appearances.
2. Historical Greylock profile/author snapshots and the full, unedited 2004 LinkedIn and 2018 Roblox IC memoranda, including adverse sections and vote records.
3. Deal-by-deal Greylock checks, ownership, dilution, follow-ons, distributions, and attribution for LinkedIn, Facebook, Pandora, Nextdoor, Discord, musical.ly, Roblox, and ordinary/failing deals.
4. Complete early Greylock enterprise-loss and consumer-deal ledgers, including passes, write-offs, and the record excluding Facebook.
5. Founder, co-investor, executive, and fellow-director testimony allocating sourcing, diligence, approval, recruiting, board interventions, follow-ons, and exits.
6. Exact dates and role records for Marakon, BCG, HBO, Electronic Arts, Crystal Dynamics, Excite/Excite@Home, the advisory-partner transition, and current private boards.

Source notes