Jenny Lee

1. Snapshot and testable edge thesis

Lee's proposed edge is a problem-first, cross-market pattern system: map a plausible future, find the bottleneck, test local users and founder capacity, then recruit/board through gaps. Investor-stated with documented applications; support: moderate. It predicts pre-pitch research, field evidence, local ownership and non-outlier results; a lifetime ledger assigning work to partners or showing ordinary returns would falsify personal alpha (EDB; NIU).

2. Core philosophy

Start from a real problem and plausible future rather than trend labels; temporarily enter the founder's ambitious world; then force belief through customer, team, timing and policy evidence. Analytical “brain” and long-working “heart” are complements, not substitutes (EDB, founder-fit blocks; Kellogg profile). Investor-stated; strong articulation, moderate application.

3. Sourcing

Opportunity flow combines founder referrals, thematic hunting, cross-market pattern recognition and local networks. Lei Jun links Xiaomi/WPS/UC/YY; NIU came from smart mobility; hiSoft reportedly came through a technology park (Forbes 2022, cases; 996, 29:26–32:31). The firm reviewed about 4,000 opportunities in 2021 and spent six-to-twelve months around founders/markets, but that is firm—not Lee—flow (RI deck, pp.4–5). Documented/investor-stated; moderate.

4. Picking

At seed/Series A Lee weights learning speed, passion, adaptability, recruiting and local-relative grade; later she shifts toward operating numbers. Separate N-1/N-2 references test influence/truth. Direct product truth came from roughly 200 YY interviews in direct audio versus about 100 in the NIU retelling, and 5,000 NIU surveys (Kauffman, 11:05–15:29; 996, 24:17–29:13; NIU). Direct voice/behavior; strong method, moderate repeatability.

5. Deal and portfolio mechanics

Public evidence supports stage breadth, not a reproducible check/ownership formula. Discovery covered seed/early and Main/Plus later stages; RI proposed $350M Discovery Asia and $2.0B combined Main/Plus. Target ownership and check-level reserves are absent (AVCJ, p.23; RI deck, pp.3–7).

NIU shows follow-on through crisis; YY shows reported: 15x (EDB, introduction, 2018) full GGV exit; EHang shows large A/follow-on; Xiaomi mixes angel/fund capital. Documented cases; moderate; numeric mechanics insufficient. A modern IC must add exposure, pro-rata, milestone and exit rules.

6. Vehicle and incentive fit

GGV's ten-year terms/five-year investment period fit Lee's seven-to-eight-year average first-check-to-exit statement (Kauffman 26:42–27:09). RI reported 2.5% fee, 20% carry and 25% premium carry above 2.5x net MOIC plus $60M GP commitment (RI deck, pp.3–5). Premium carry rewards outliers; multiple vehicles preserve ownership but raise allocation/signaling risk. Public LP record; strong for proposed terms, insufficient for Lee's incentive.

7. Team attribution and work with founders

Post-investment evidence is clearest at hiSoft (CEO recruiting), NIU (Li Yan, crisis roles, quarterly boards) and Ruangguru (support through layoffs/illness) (Forbes 2022, hiSoft; NIU; Ruangguru). Reported/direct behavior; moderate. Board service proves proximity, not causality. Source, sponsor, diligence, approval, negotiation, board, recruiting, follow-on and exit are separately coded; unknown stays unknown.

8. Views on the craft

Lee describes venture as curiosity, resilience, people judgment and long problem-solving rather than prediction alone. Current entrepreneurs are stronger market sensors than books/podcasts (Kauffman 28:21–29:42). Bokee shows staying through shutdown; CNA describes helping a CEO after layoffs. Investor-stated; moderate as behavior, insufficient for efficacy. The craft claim fails if support is selected testimony without counterfactual impact.

9. Signature bets and decision process

Reconstruction; moderate. The analyst-created exercises and sourced outcomes are separated in investments.md.

10. Mental models and maxims

1. Suspend disbelief, then earn conviction.
2. Map the future deductively.
3. Grade founders by stage and ecosystem.
4. Import patterns, localize execution.
5. Price the technology window against duration.
6. Design the team before scaling capital.
7. Close the private–public capital loop.

Each becomes a twelve-field tool in mental-models.md. “Brain and heart,” curiosity and resilience are incorporated rather than padded into overlapping models.

11. Evolution and contradictions

• IoT optimism (2015) precedes cautious wearable maturity scoring (2016), then an explicit five-year window (2022).
• Integrated US/China platform logic (2016) gives way to SEA fragmentation and ultimately separation/multi-asset Asia (2020–2024).
• “Suspend disbelief” conflicts productively with trust-and-verify; no public calibration data shows the balance.
• Long holds enable company-building but Bokee/hardware shutdowns show patience can consume capital.
• EdTech fragmentation/expansion meets a regime shock; public-market price discipline meets venture's tolerance for mispricing.

Dated investor statements plus researcher synthesis; moderate.

12. Failures and limits

Bokee is named and role-attributed but lacks economics. Unnamed social/hardware failures provide causes, not identities. LAIX shows IPO fragility; Spark policy risk; EHang continued-verification risk; 51 Credit firm risk outside the person denominator (996, 12:10–13:49; LAIX 6-K; EHang annual report, p.17). Documented adverse controls; moderate. No failure rate, process-change chain or loss denominator exists.

13. Historical operating environment

Early China offered sparse infrastructure/experienced teams alongside rapid adoption and new listing routes. By 2022 GGV reported 521 firm investments, 56 IPOs and 94 $1B+ leaders, with large unrealized marks; this describes a team/favorable vintages, not Lee alpha (RI deck, p.2). Education/data/outbound-investment rules later altered base conditions. No contemporaneous numerical venture base rate supports a hit-rate comparison.

14. Modern VC translation (as of 2026-08-27)

Researcher translation; moderate for first six, provisional for capital loop. Granite's structure is a current example, but returns/conflicts are not public.

15. Comparison with completed peers

De Rycker isolates local authority plus global transfer (profile); Son isolates outlier attribution and multi-vehicle scale (profile). Researcher comparison; moderate and dimension-bounded, not a ranking.

16. The strongest case against greatness

The visible record mixes GGV team outcomes, Lei Jun access, favorable China/mobile timing and three famous winners. “Led,” board, angel and shared IC power are often collapsed. The denominator is 47 searched families, only eight strictly reconstructed and 39 unresolved; returns are firm/selected/reported. LAIX, Spark, EHang scrutiny and shutdowns show that a listing/ambition is not durable value.

The favorable response is process evidence: NIU field/team chain, YY research, hiSoft recruiting and consistent problem-first language. Researcher inference; moderate. A lifetime role-coded ledger, original memos, audited cash flows, ordinary/failure testimony, passes and board minutes would decide.

17. Analyst study guide, glossary, and go deeper

Path. Read writings, then talks; replay WPS, YY, NIU, EHang and Spark; use models; finish with review.

Glossary. Discovery: seed/early vehicle; Main/Plus: later/larger pools; future map: component thesis; technology window: time to proof; board-control entry: directorship, not investment/return.

Checklist. Sourcing: thesis date/origin/referral. Diligence: paid users, cohorts, independent customers, lieutenant references, regulation. IC: known/unknown/disputed, terms/ownership, kill, next check, bear case. Governance: roles, bad-news cadence, recruiting, reserves, policy tree, stop/exit.

Best sources. NIU for process; Kauffman for procedure; EDB for philosophy; RI for vehicles; SGX for denominator. Full bibliography/access is sources.md.

Pareto 80/20 — what drove the record

No audited Lee-level cash flows support a literal 80/20 calculation. Economic contribution is indeterminate because check sizes, ownership, dilution and proceeds are absent; YY and WPS are visible return clues, not a contribution rank. Historical importance and decision-learning value are ranked separately.

Removing either visible return clue leaves an attributed set whose economic ordering and hurdle performance remain unknowable. “21 $1B+ companies/17 IPOs” are counts, not returns (Granite biography).

Evidence boundary and searched denominator

This is a bounded person-attribution ledger, not a comprehensive lifetime list. The Edge says Lee had been involved in almost 60 GGV companies but names no denominator. The SGX filing lists 75 legal directorship entries: 36 operating entries (26 past + ten current) and 39 exclusions (one past GGV entity + 38 present Temasek/vehicle/non-operating entries).

Aliases/follow-ons are one family when supported: hiSoft/Pactera, YY/JOYY, Sensoro/ISENSORO, Chukong/Xiamen Yaji. Airlook Airspace and Airlook Aviation remain separate pending proof of aliasing. Fengbian and BeautyVector remain separate. 51 Credit Card is a firm-risk control outside the person denominator. The 12 supported non-board names are WPS, YY, hiSoft, 21Vianet, SinoSun, Xiaomi, XPeng, UCWeb, Glu Mobile, Ruangguru, Geek+ and BlogChina/Bokee. Chukong/Xiamen Yaji is retained as the filing-supported family (SGX schedule, pp.1–2).

Historical, market, and regulatory context

Early China venture combined PC→mobile adoption, scarce local growth capital and fragmented management networks. Mid-2010s devices/AI widened the set. After 2020, education restrictions, fintech/data enforcement, listing friction and outbound-investment scrutiny changed company feasibility and exits. The House report documents GGV-level Megvii scrutiny, not Lee sponsorship (House report, pp.11–13). Granite's Singapore/multi-asset shift is institutional adaptation, not proof risk disappeared.

Vehicle economics and incentives

The 2022 LP deck proposed Fund IX $1.6B, Plus $400M and Discovery IV Asia $350M; $60M GP commitment; 2.5% fee; 20% carry and 25% premium carry above 2.5x net MOIC; five-year investment periods and ten-year terms. Main targeted 40–50% US, 30–40% China, 15–20% SEA; Discovery Asia 40–50% China and 40–50% Southeast/South Asia (RI deck, pp.3–5).

It showed reported: 2.0x (RI deck, p.2, 2022-06-30) aggregate net TVPI and a reported 17% net IRR, with vintage dispersion and unrealized value. These are manager-supplied firm marks. Ten-year duration supports patient board work; premium carry rewards outliers; multiple vehicles raise allocation/signaling questions. Unknown: Lee carry, IC thresholds, checks, ownership, reserves, extensions, currency and DPI.

Material company ledger

SGX board-control ledger and exclusions

The 26 past operating entries comprise LAIX, NIU, EHang, Agora and 22 private legal entities: Wuhan Kubote, Airlook Airspace, Myhug, Immotor, Kascend, Phononic Devices, FarmFriend, Vincross, Beijing Zhichong, Yodo1, Baige, Clobotics, JH, Airlook Aviation, BeautyVector, Fengbian, ISENSORO, InspirAI, Chukong, MegaRobo, Xiamen Yaji and Guangxi Puyuan. Past GGV (Xiaole) is excluded. The ten current operating entries are Cashield, FLT, FUSE, Spark, Palfish, Cloud First, XChange, Treelab, Next Gen Foods and TTMFS Singapore (Thunes). The remaining 38 present entries are excluded Temasek/vehicle/non-operating legal entities. Board presence proves identity, not investment source, date, price or outcome.

No-hindsight case lab: Kingsoft WPS

Ex-ante freeze. In 2014, WPS was a long-lived office suite separated from Kingsoft; Microsoft defined the category. Known: installed history, China workflows, experienced sponsor. Unknown: standalone economics, cloud/mobile transition, governance, pricing and exit. Lee says Lei Jun introduced it and she led (My Investments, §2). Investor-stated retrospective; moderate.

Analyst-created IC exercise before outcome.

• Would you invest then? Conditional yes after clean carve-out governance and retention proof.
• What kills it? Related-party leakage, non-monetizing free use or sponsor dependence without a team.
• What ownership/terms fit? Governance/information rights and meaningful ownership for five years; no historical number invented.
• What unlocks the next check? Retention, conversion, separation economics and standalone leadership.

Sourced outcome. WPS listed in 2019; reported: 55x (Forbes 2022, Kingsoft, 2022) mark. It is not verified realized performance.

No-hindsight case lab: YY / JOYY

Ex-ante freeze. YY was a free low-latency gamer coordination tool before virtual gifts (Kauffman 20:00–21:46). NIU retells about 100 user questionnaires; 996 direct audio says roughly 200 across three cities. Known: high-frequency utility/sticky use. Unknown: willingness to pay, moderation, distribution and founder scaling (NIU, YY comparison; 996, 24:17–29:13).

Analyst-created IC exercise before outcome.

• Would you invest then? Conditional yes if utility persists without subsidized traffic and lawful monetization works.
• What kills it? Weak cohorts, unstable whales, moderation failure or fragile distribution.
• What ownership/terms fit? Stage capital around retention/monetization with board/information rights; historical stake unknown.
• What unlocks the next check? Cohorts, payment conversion, unit economics, moderation and traffic resilience.

Sourced outcome. YY listed in 2012; reported: 15x (EDB, introduction, 2018) full GGV exit in 2017. Ownership and cash remain missing.

No-hindsight case lab: NIU

Ex-ante freeze. In 2015 NIU had a smart-scooter proposal and limited operating history. GGV's case reports 5,000 surveys, Li Yan recruitment, crisis role redistribution and continued investment. Known: mobility pain, design and field interest. Unknown: manufacturing, channels, founder fit, regulation and paid conversion (NIU, §§2–7). Investor-stated/documented behavior; moderate-to-strong.

Analyst-created IC exercise before outcome.

• Would you invest then? Conditional on paid demand, safety and role clarity—not survey intent alone.
• What kills it? Nonconversion, defects, battery safety or unresolved authority.
• What ownership/terms fit? Milestone capital plus board/safety/manufacturing information rights; stake unknown.
• What unlocks the next check? Deposits→delivery, defects/repair, margin, channel repeatability and stable roles.

Sourced outcome. NIU listed in 2018. No return magnitude is public; learning value comes from the linked process chain.

No-hindsight case lab: EHang

Ex-ante freeze. In 2015 passenger drones faced technical, certification, manufacturing and demand uncertainty. GGV's A was then its largest and it followed; Lee's 2016 speech emphasized commercialization (EHang announcement; frontier speech).

Analyst-created IC exercise before outcome.

• Would you invest then? Only as staged frontier risk after independent diligence and duration-compatible certification.
• What kills it? Non-reproducible safety, unverifiable customers, regulatory delay or promotion outrunning facts.
• What ownership/terms fit? Tranches against certification, paid use and manufacturing, with safety/audit rights.
• What unlocks the next check? Independent flight tests, regulator milestones, customer cash and repeatable quality.

Sourced outcome/adverse update. EHang listed in 2019. Wolfpack alleged fabricated revenue; EHang denied it and later disclosed price/litigation effects. Both sides are recorded without adjudication (response; annual report, p.17).

No-hindsight failure lab: Spark Education

Ex-ante freeze. In 2019 online education had large demand, fragmented curricula and service-heavy delivery. Lee later says she led Spark; filing records her director role from July. Known: subject content and measurable retention potential. Unknown: cohorts, acquisition payback, teacher quality, licensed scope, policy durability, terms and reserves (My Investments, §1; Spark F-1, p.174). The 2020 rubric is later narration, not proof of the 2019 memo.

Analyst-created IC exercise before outcome.

• Would you invest then? Only if licensed revenue, retention and renewal cover service labor/CAC without assuming policy continuity.
• What kills it? Unlicensed core subjects, weak renewal, teacher variance, IPO-dependent burn or no lawful adjacency.
• What ownership/terms fit? Tranches against renewal/licensing with board/information/downside rights; no historical terms invented.
• What unlocks the next check? Retention, referrals, contribution margin, CAC payback, licensure by revenue and funded pivot.

Sourced actual decision/outcome. Lee says she led the 2019 investment; the filing confirms the board role. Spark filed in June 2021 and withdrew on 2021-12-01 after core-subject restrictions; the withdrawal cites market conditions (F-1/A; Reuters policy; withdrawal). Board response, impairment and process change are unknown.

Failure and regulatory controls

LAIX: listing is not durable success

The SGX schedule proves Lee's former board, not entry sponsorship. NYSE suspended LAIX after its 30-day average market capitalization fell below $15M; it moved OTC and completed a $1.90/ADS take-private (delisting 6-K; merger 6-K). Original thesis, stake, reserves and board response are unknown. IPO count is not durable value.

Bokee and unnamed shutdowns

Direct 996 audio names BlogChina/Bokee: Lee/Tung co-invested, joined the board and stayed through wind-down (12:10–13:49). The Edge adds unnamed social and hardware shutdowns driven by price, competition, design, capital and timing. None supplies economics; no failure-rate claim is possible.

51 Credit Card: outside-denominator firm control

The company disclosed an October 2019 on-site investigation and executive assistance (HKEX announcement). The record does not tie Lee personally, so it is excluded from the 47-family person denominator and retained only as firm-risk context.

Misses and sell discipline

No named pass ledger exists. Kauffman supplies a roughly five-year proof window and seven-to-eight-year average duration, but no numeric sell rule. CNA says unresolved risk disagreement can trigger early exit (15:34–15:50), without a named case. A modern IC should record passes, reserves and exit tranches before outcomes; otherwise survivorship dominates.

Coverage and evidence base

The priority written corpus contains 14 direct-voice or first-party records: six ★★★ and eight ★★, plus one ★ current-strategy lead. All 14 priority records have standardized notes and substantive analysis below: thirteen were accessible in full; Fortune's IoT body was partial but substantive. The separate Forbes, Edge, Midas, filing and regulatory records are controls rather than Lee-authored works. No personal blog, book, public investment memo, LP letter or annual shareholder letter was located.

Evidence is strongest on market mapping, founder/product judgment, localization and board/recruiting work; it is weak on prices, ownership, reserves, passes, write-offs and cash realization. Every derived lesson below separates Lee's statement from evidence it worked.

Source-by-source analysis

Through the lens of VC (2018)

The Singapore EDB republished an edited KrASIA Q&A while Lee was based in Shanghai. Its argument is a chain: envision a future, study the sector, find founder-product-market fit, then combine analytical diligence with founder chemistry (EDB, question blocks 26–48).

Decision rule — investor-stated; strong articulation, moderate application: map before the pitch, “suspend disbelief” long enough to understand the founder's world, then demand market/technology/return logic and ten-year working fit. NIU's field study grounds the analytical side; failures show conviction does not eliminate false positives (NIU, lines 68–106; Edge, “Success and failure”). The edited retrospective does not supply passes or contemporaneous scores.

Small world / AVCJ (2016–2017)

The interview documents Discovery for seed/early risk, Main for Series A through IPO, and services spanning recruiting, finance, legal, marketing, PR and government relations (AVCJ, printed p.23/PDF p.22). Vehicle lesson — contemporaneous investor statement; strong for design, provisional for causality: match capital/services to stage rather than forcing one pool on every risk. The later LP deck supplies ten-year terms, five-year investment periods, 2.5% fees and 20%/25% premium carry; Lee's allocation/carry remain private (RI deck, pp.2–5).

Mobile reduced distribution friction, but consumer behavior, regulation and teams required localization. The 2024 split is a boundary to the then-integrated global-platform belief.

China frontier-technology speech (2016)

Lee treats maturity as investable: early wearables taught users, while “unworn” systems were conceptually possible but only 0.5–1 on a ten-point readiness scale. She decomposes transport through labor/time constraints and automation through wages/manufacturing depth (Jiemian, wearable/transport/robotics sections).

Thesis — investor-stated; moderate: map the constraint and required components, then distinguish prototype feasibility from commercial readiness. EHang reveals the tension between future vision and certification/monetization; appended portfolio names are not person-level sponsorship proof (EHang announcement; Edge, EHang discussion).

NIU diligence and crisis case (2018)

This GGV-origin interview is the richest quasi-memo. GGV led Series A after more than 5,000 surveys, lower-tier-city visits and willingness-to-pay analysis. Lee describes recruiting Li Yan, reallocating founder roles after crisis, spending discipline, quarterly board materials and continued funding (NIU, lines 62–151 and 175–223).

Decision chain — investor-stated/documented behavior; strong: when outside the user base, sample target users directly; test price/pain; then test whether the team can share authority under stress. It is retrospective/firm-produced and omits terms, ownership and proceeds. Its “about 100” YY survey count conflicts with direct 996 audio's roughly 200; the discrepancy is preserved.

Southeast Asia and the next billion (2020)

Lee argues SEA is not one market: China supplies hypotheses about mobile consumption and infrastructure, while licenses, transport, prices, cultures and local champions change execution (YourStory, Lee blocks 177–201). Rule — investor-stated; moderate: import a mechanism as a question, not a product answer. Jixun Foo—not Lee—narrates Grab's first GGV bet, preserving team credit (YourStory, lines 176–182).

Bridging the Gap in Southeast Asia's Capital Markets (2024)

Lee and SGX executive Ng Yao Loong argue that private risk capital, public price discovery, research and liquidity form a recycling loop. Without credible local public markets, smaller SEA companies can be invisible abroad and private markets lose feedback (Granite, “Mind the gap”/“Finding the Big Fish”).

System thesis — investor-stated/co-authored; moderate: invest in ecosystem infrastructure and choose comparables by operating conditions. The 2024 author note says 16 IPOs; live Granite and Forbes 2025 say 17, with no common definition. Firm IPO-market-cap counts are not proceeds.

Granite Asia rebrand (2024)

The announcement names Lee and Jixun Foo as co-leaders, anchors the firm in Singapore and expands beyond venture into growth and credit (Granite rebrand). Evolution — contemporaneous record; strong for structure, provisional for motive: multiple capital forms may fit companies staying private longer. Independent policy context does not prove Lee sponsored Megvii or caused scrutiny (House report, pp.11–13).

Official biography and history (current)

Granite's live page supplies role, education, sectors and firm milestones and reports 21 $1B+ companies/17 IPOs (Granite biography). It is authoritative for identity, promotional for outcomes, and cannot prove one edge caused every result. Its dated totals are reconciled rather than merged with 2024's 16.

Ruangguru founder/firm case (2023)

The firm defines EdTech's post-pandemic execution phase through reach, robustness and AI integration and uses founder experience through illness/layoffs as board-relationship evidence (Ruangguru, “Taking education…”). Rule — investor-stated/founder witness; provisional: after an adoption shock, shift from category possibility toward product, distribution and organizational behavior. Entry, ownership and formal Lee sponsorship are absent.

EHang Series B (2015)

EHang says GGV's Series A was then its largest and that it followed into Series B; Lee links the decision to global UAV growth (EHang announcement). Application — documented firm behavior; moderate: commit materially when technology/team/sector align, then follow. The promotional source and later certification/allegation risk leave technical diligence, solo approval and realized economics unresolved.

Kellogg alumni profile (2016)

Lee connects formation to a problem-first orientation and board help (Kellogg profile, headings). Causal explanation — researcher inference; provisional: engineering may improve technical dialogue, but chronology does not prove selection skill; surveys, component maps and recruiting are stronger observed mechanisms.

Internet of Things means never having to search again (2015)

Lee and Hans Tung predict blurred hardware/sensor/software interfaces and pair US product/market strengths with Chinese manufacturing/engineering (Fortune, accessible indexed body). Investor-stated/co-authored; provisional: the architecture is plausible, but smartphone displacement was early and the 2024 split constrains “global from day one.” Access is partial; no unexposed text is treated as read.

VCs are betting on the great Chinese fitness boom (2016)

Lee and Haojun Li map fitness/community apps, connected devices/data and outdoor/mobility services from one behavior shift; Keep and NIU illustrate the stack (TechCrunch, “Wide-ranging investment opportunities”). Investor-stated/co-authored; moderate: behavior creates demand, apps coordinate, devices measure and services monetize. Later hardware shutdown admissions supply price/competition/design/capital contradictions (Edge, “Success and failure”).

My Investments and Exits (2020)

Lee names Geek+, Sensoro, Spark and Fengbian as investments she led in 2019; calls WPS her ninth led IPO; dates WPS to 2014; credits Lei Jun's introduction; and narrates carve-out, business-model change, failed internationalization and domestic refocus (Qianzhan/GGV, §§1–3).

Documented behavior/investor-stated; moderate: repeated relationships improve access/context but do not prove alpha. EHang is described as too early and forced to save itself. Spark's later policy shock constrains winner selection; terms and proceeds remain absent.

Current strategy lead: DBS and Granite Asia (2026)

The partnership is a freshness control for multi-asset direction (Granite/DBS announcement). It is ★ metadata/context, not counted among the 14 priority writings and not used to retroactively validate the venture record.

Cross-source synthesis

Three beliefs recur. Curiosity means replacing self-reference with user/employee evidence (EDB, lines 45–48; NIU, lines 87–106). Global patterns predict sequences, but local prices, people and institutions decide whether they travel (AVCJ, p.23; YourStory, lines 183–201). Investor work extends through recruiting, boards, later rounds and exits, but long vehicles—not temperament alone—enable it (RI deck, pp.2–6).

Rhetoric and observed behavior align at NIU, YY diligence, hiSoft recruiting and EHang follow-on. These are selected cases. Bokee and unnamed shutdowns prevent a claim that the playbook reliably avoids loss (996, 12:10–13:49; Edge).

Intellectual chronology, evolution, and contradictions

The hardest tension is openness versus discipline: suspension expands the opportunity set; field checks, timing and policy narrow it. No public IC memo proves calibration.

Decision playbook and modern translation

1. Map the constraint, technology window and disconfirmers before meeting. Investor-stated; moderate.
2. Replace self-reference with users and separate lieutenant references; kill when independent evidence contradicts the pitch. Documented behavior; strong in NIU/YY.
3. Import mechanisms, then re-underwrite local regulation, price, product and leadership. Investor-stated; moderate.
4. Match capital to evidence gates; do not make pro rata automatic when proof slips. Researcher synthesis; moderate vehicle support.
5. Plan recruiting, governance and liquidity across a long hold; do not rely on IPO availability. Investor-stated/documented vehicle; moderate.

Modern VC translation (as of 2026-08-27)

User/employee checks, local regulation maps and staged capital transfer. The 2016 assumption of platform convergence does not: export controls, data localization and cross-border policy affect product and fund structure. Applicability signals are reachable users, accountable local regulation, a five-year proof window and matching reserves/duration; likely misuse is relabeling analogy as evidence. Researcher inference; moderate, grounded in the shift from AVCJ's integrated structure to Granite's regional form.

Best writings and unresolved record

1. EDB 2018 — clearest full decision rubric.
2. NIU 2018 — richest diligence/recruiting/crisis application.
3. AVCJ 2017 — best vehicle/service/localization record.
4. My Investments and Exits 2020 — strongest first-person sponsorship ledger.
5. Capital-markets essay 2024 — clearest current ecosystem/exits evolution.

Unresolved: the Fortune article is partial; DBS is metadata-only; no personal blog/book/archive, original IC memo, pass note, board minute, LP letter, complete cohort or realized cash-flow record emerged. The retired EdTech transcript was mitigated by legitimate OpenAI audio transcription, not treated as a writing.

Coverage denominator

Eight distinct priority appearances comprise Analyse Asia (2015), CNA (2017), 996 (2018), Next Billion EdTech (2020), GGV/YourStory Southeast Asia (2020), Kauffman Fellows (2022), and two DealStreetAsia market-reset appearances (March and December 2022). Each has its own note: five full-audio transcripts, one edited transcript and two partial access records. They are grouped into seven analytical sections below only because the two distinct DealStreet appearances share one comparative section; they are not deduplicated as one appearance. Full public audio for Analyse, CNA, 996, EdTech and Kauffman was acquired and transcribed through the OpenAI API; media/full transcripts remained temporary. YourStory supplies an edited speaker-attributed transcript. DealStreet bodies remain partial; DBS is a lower-priority discovery route.

Searches covered English and 李宏玮, Apple/Amazon/podcast indexes, CNA/Brightcove, conference and firm archives, transcripts and captions. No oral history, classroom lecture, LP meeting audio or testimony was found.

Source-by-source guide

Analyse Asia at RISE (2015)

The surviving 11:42 published audio—publisher says roughly half the interview was lost to a technical interruption—contrasts a mature US ecosystem with then first-time-founder China and says imported patterns still require local context (01:50–04:29). Leadership must evolve with scale (04:53–05:59); “clone” is not a substitute for user/commercialization analysis (06:09–08:36); PC software gives way to vertical internet, mobile, sensors and cloud (08:44–11:40) (Analyse episode). Investor-stated; strong as dated articulation, insufficient as efficacy. Malay orthography in the transcript limits quotation.

CNA Power List Asia (2017)

Lee describes China-office friction (01:14–03:15), a technical/product/sacrifice/recruiting founder test (05:16–06:27), and seven-to-ten-year holds. She states targets of reported: 3–5x (CNA, 06:27–07:33, 2017) on fund capital and reported: >10x (CNA, 06:27–07:33, 2017) for a deal—not realized results. She recalls supporting a CEO through shutdown (07:42–08:46), historical trust-and-verify checks (10:40–13:43), and early exit when risk disagreement persists (15:34–15:50). Investor-stated; moderate. Brightcove had no subtitles; full audio was OpenAI-transcribed. A narrator's Didi attribution at 03:50 is rejected.

996: Being a VC in China (2018)

This 65:18 record connects engineering, the 2001 move, Morgan Stanley/JAFCO and the 2005 GGV build (03:50–08:49). It names BlogChina/Bokee as a failure: Lee and Hans Tung co-invested, joined the board, and stayed through wind-down (12:10–13:49). Lee describes roughly 200 YY interviews across three cities, private investigators and 70–76% sample use (24:17–29:13), then recursive founder/technology-park sourcing (29:26–32:31). She discusses policy direction but admits no special government insight (39:37–51:13) (996 page/audio route). Investor-stated; strong role/process evidence, moderate selected application. Publisher enclosure 404; Apple supplied legitimate audio.

Next Billion: EdTech is not winner-take-all (2020)

The 43:20 audio describes roughly 15 GGV EdTech companies—a category, not Lee denominator (01:47–04:27). Product-market fit joins content, technology, teachers, operations and acquisition (10:08–15:01); age, curriculum, format and licensing fragment the market (10:08–17:37). Lee ties painkillers/vitamins to frequency, ARPU and LTV (26:22–30:01), then requires unit economics, retention, renewal and referrals after monetization (37:01–38:18) (episode). Investor-stated; strong rubric, provisional outcome efficacy. The 2021 policy shock/Spark withdrawal is the adverse control.

Southeast Asia and the next billion (2020)

Lee says SEA is fragmented, local founders retain initial advantages and financial-system infrastructure matters; Jixun Foo—not Lee—narrates Grab (YourStory, lines 172–205). Investor-stated; strong as speech, moderate application. Co-appearance does not create sponsorship.

Kauffman Fellows: emerging technology ecosystems (2022)

Emerging markets differ by founder depth, regulation/infrastructure and exits (01:50–06:22). At seed, Lee tests learning and influence through repeat meetings and separate lieutenant references; later narrative must become product/number proof (10:25–15:29). Global foresight remains bounded by local people, product, regulation and willingness to pay (16:10–18:09) (Kauffman episode).

YY first solved low-latency gamer coordination; virtual gifts monetized later (20:00–21:46). Technology should become investable within roughly five years (23:55–24:40); Lee states a seven-to-eight-year GGV average from first check to final exit (26:42–27:09). Investor-stated; strong articulation, moderate selected application; performance denominator absent.

DealStreetAsia market-reset appearances (2022)

March preserves “Mr. Market is always right”; December frames the funding winter as introspection (March report; December video). Investor-stated; provisional: partial bodies cannot reveal how price overrides operating conviction.

Recurring frameworks

Consistency is not causal alpha; ordinary/failure evidence remains incomplete.

Evolution and contradictions

• 2015–2018: platform transitions, China learning, field verification and recursive sourcing dominate.
• 2020: transfer becomes explicitly constrained by SEA fragmentation, licenses and service operations.
• 2022: founder references, stage proof, timing/duration and valuation feedback become procedural.
• 2024–2026: writing expands toward Singapore-centered capital-market architecture. Researcher inference; moderate: unit of analysis widened from company to ecosystem.

“Suspending disbelief” risks charisma; references/surveys are the proposed brake. China-to-world convergence meets the later split; EdTech optimism meets policy shock; patience meets Bokee and unnamed shutdowns.

Deal rationales and attribution checks

YY's utility-first logic and field work are direct; NIU's chain is first-party; Xiaomi began as Lee's personal angel check before GGV; hiSoft combines technology-park sourcing and CEO recruiting; EHang combines vision with incomplete commercialization. No spoken record supports assigning Grab, Alibaba, Didi, Megvii or all GGV China outcomes to Lee. Chukong/Xiamen Yaji is retained as one filing-supported board-company family.

Listen to these five first

1. 996 (2018) — sourcing, diligence, policy boundary and named failure.
2. Kauffman (2022) — integrated founder/localization/timing rubric.
3. EdTech (2020) — whole-system PMF and stage metrics.
4. CNA (2017) — duration, shutdown support and sell clue.
5. Analyse (2015) — contemporaneous platform/local-context framework.

Modern decision translation (as of 2026-08-27)

A modern IC should score target-user evidence, founder learning/references, local policy ownership, whole-system economics, technology-window proof, next-check gates and duration-compatible exits. Historical investigator/family checks do not transfer without lawful proportional privacy controls. AI/robotics misuse is equating a demo with five-year commercialization; cross-border misuse is importing a playbook without local authority. Researcher inference; moderate.

Unresolved spoken record

DealStreet and DBS lack complete stable transcripts. Five priority audio families had legitimate acquisition/OpenAI transcription; Analyse limits quotation and CNA had no caption track. No audio reveals pricing/ownership discipline, IC votes, reserve decisions, write-off mechanics or a full failed-company cohort.

Suspend disbelief, then earn conviction

• Rule — Enter the founder's ambitious future, then convert belief into independently testable product, customer, team and timing evidence.
• Evidence status — Investor-stated with documented applications; strong for articulation, moderate for repeatability/efficacy.
• Provenance and origin — EDB supplies the phrase and founder-fit test; earlier IoT/frontier work shows the posture without proving results (EDB, founder traits).
• Mechanism and reasoning — Immediate skepticism rejects category creators; unbounded optimism funds stories. Separate possibility from proof.
• Historical and vehicle context — Early China lacked settled categories; ten-year multi-stage funds supported experimentation but enlarged timing exposure (RI deck, pp.3–7).
• When to use — Seed/frontier/new behavior with weak comps; never relax integrity, safety, regulation, terms or milestones.
• How to apply it — Write the strongest future; list assumptions; choose independent tests/disconfirming reference; set technical/regulatory gates and next-check learning.
• Examples —
  • Documented application — NIU: 5,000 surveys and team recruitment bounded a new scooter vision (NIU, §§2–7).
  • Documented application — YY: roughly 200 interviews in direct audio preceded monetization, then reported: 15x (EDB, introduction, 2018) full GGV exit (996, 24:17–29:13).
  • Boundary — EHang: ambition led to investment/listing; allegations/rebuttal show verification must continue (EHang response).
• Evidence — EDB states the rule; NIU/996 document field tests; Kauffman supplies references; EHang records the adverse control.
• Limits, failure modes, and exception ledger — Charisma can mimic vision, survey intent payment, demos certification. LAIX/Spark add company/regime risk. No calibration/pass cohort or verification protocol was found.
• Evolution and contradictions — 2015 interface optimism meets 2016 low wearable-maturity scoring, suggesting timing discipline without proven process change (Fortune; Jiemian).
• Modern VC translation (as of 2026-08-27) — For AI/robotics require production retention, independent customers, reproducible evaluation and funded regulation. Likely misuse: calling scrutiny unimaginative. Researcher inference; moderate.

Map the future deductively

• Rule — Begin with a plausible future, decompose its technical/commercial stack, then source the binding bottlenecks.
• Evidence status — Investor-stated across multiple writings; moderate-to-strong as process, insufficient for causal returns.
• Provenance and origin — The frontier speech maps transport/robotics/automation by maturity; EDB explicitly calls the process future-led/deductive (Jiemian; EDB, sourcing).
• Mechanism and reasoning — Decomposition turns vague TAM into required components, timing and disconfirmers and supports proactive sourcing.
• Historical and vehicle context — PC→mobile→sensor shifts opened new stacks; China's manufacturing base changed which bottlenecks were locally advantaged (TechCrunch fitness).
• When to use — Platform transitions, hardware/AI, mobility and infrastructure; avoid when the “future” has no falsifiable timing or customer.
• How to apply it — State future date; draw component graph; grade maturity; identify scarce link; source companies; require a test that could kill each link.
• Examples —
  • Documented thesis — fitness: apps→devices/data→mobility services (TechCrunch, opportunity sections).
  • Documented application — NIU: smart-mobility map led to field work and Series A (NIU).
  • Counterexample — IoT interface: smartphone displacement was too early (Fortune).
• Evidence — Three dated direct sources independently establish method/application; EHang supplies a commercialization boundary (EHang).
• Limits, failure modes, and exception ledger — Future maps invite narrative confirmation, category double counting and premature timing. No complete map→deal/pass/outcome cohort or forecast scorecard exists.
• Evolution and contradictions — Broad IoT optimism becomes explicit maturity scoring and later a five-year proof window, an articulation change with unknown performance effect.
• Modern VC translation (as of 2026-08-27) — In AI, map data, compute, model, workflow, distribution and governance; kill theses whose scarce link is already commoditizing. Misuse: stack diagrams as diligence. Researcher inference; moderate.

Grade the founder by stage and ecosystem

• Rule — At seed weight learning, integrity and recruiting; with scale shift weight toward metrics, organization and execution, using local-relative context without lowering absolute integrity.
• Evidence status — Directly stated with recruiting/team cases; strong articulation, moderate application.
• Provenance and origin — Kauffman gives repeat meetings, N-1/N-2 references, local grading and later numeric proof (11:05–15:29) (episode).
• Mechanism and reasoning — Early metrics are noisy; learning/recruiting preserve option value. Later narrative persistence without numbers destroys it.
• Historical and vehicle context — Early China had fewer experienced managers, raising the value of GGV recruiting/services (RI deck, pp.4–5).
• When to use — Founder selection, references, stage transitions and new geographies; never relativize dishonesty, abuse or controls.
• How to apply it — Define stage; score truth/learning/recruiting/product/metrics; interview lieutenants separately; name missing role; set next-round proof.
• Examples —
  • Documented application — NIU: Li Yan recruitment and crisis role redistribution (NIU).
  • Reported application — hiSoft: Lee reportedly recruited a CEO after sourcing (Forbes 2022, hiSoft).
  • Boundary — LAIX: board/IPO did not prevent delisting; founder/intervention record absent (LAIX 6-K).
• Evidence — Kauffman process, NIU/Forbes cases and Ruangguru founder witness provide independent placements (Ruangguru).
• Limits, failure modes, and exception ledger — Charisma/cultural bias, coached references and investor-credit inflation. No scorecard, negative-reference case or management-replacement record emerged.
• Evolution and contradictions — “Brain/heart” becomes more procedural reference/stage weighting by 2022; adverse EdTech outcomes show no documented rubric revision.
• Modern VC translation (as of 2026-08-27) — For AI founders, require fast product/eval/safety learning, then retention/margin/reliability. Misuse: demos as learning speed. Researcher inference; moderate.

Import patterns, localize execution

• Rule — Transfer causal mechanisms across markets, not surface templates; rebuild product, pricing, people, distribution and regulation locally.
• Evidence status — Repeated investor-stated principle with cases; moderate-to-strong.
• Provenance and origin — Analyse covers US/China context and platform shifts; AVCJ formalizes globalization/localization (Analyse; AVCJ, p.23).
• Mechanism and reasoning — Prior-market patterns compress learning; local re-underwriting prevents false analogy in price, regulation or distribution.
• Historical and vehicle context — GGV spanned US/China then SEA with differing vehicle geography ranges (RI deck, pp.3–5).
• When to use — Cross-border SaaS, consumer, fintech, mobility; avoid where sovereignty/licensing/culture eliminates the mechanism.
• How to apply it — State imported cause; list local assumptions; interview local users/regulators/operators; assign local authority; kill if economics disappear at local prices.
• Examples —
  • Documented application — NIU: lower-tier willingness-to-pay rather than affluent-user assumption (NIU).
  • Documented thesis — SEA: fragmentation/local champions before giant advantage (YourStory).
  • Boundary — IoT: manufacturing bridge real, interface timing early (Fortune).
• Evidence — Analyse/AVCJ establish principle; EDB/YourStory/NIU apply across dates and regions.
• Limits, failure modes, and exception ledger — Analogy bias and copied metrics. No country-by-country hit rate or named full failed-localization postmortem exists.
• Evolution and contradictions — US/China comparison dominates 2015–2018; SEA fragmentation/public markets dominate 2020–2024; separation constrains seamless-global assumptions.
• Modern VC translation (as of 2026-08-27) — Test language, data residency, procurement, labor substitution and inference economics by country. Misuse: partnerships called localization without product authority. Researcher inference; moderate.

Price the technology window against fund duration

• Rule — Back frontier technology only when decisive proof can arrive within a fundable window—roughly five years in Lee's statement—and stage exposure to milestones.
• Evidence status — Directly stated once and supported by frontier cases; moderate, with insufficient numeric application evidence.
• Provenance and origin — Kauffman supplies five-year proof and seven-to-eight-year average exit duration (23:55–24:40; 26:42–27:09) (episode).
• Mechanism and reasoning — A correct distant future can still destroy venture returns if certification/manufacturing/adoption outlast runway and vehicle life.
• Historical and vehicle context — Ten-year funds/multiple stages permit patience but reserves/extensions are finite; aviation differs from software (RI deck, pp.3–5).
• When to use — Robotics, aviation, climate, semiconductors and regulated science; do not impose one cutoff on low-capital software or non-venture capital structures.
• How to apply it — Build technical/regulatory/manufacturing/customer clocks; estimate cash; require each next reserve to buy discriminating proof; predefine stop/license/partner paths.
• Examples —
  • Documented thesis/decision — EHang: unmanned aviation with certification central (EHang).
  • Documented application — WPS: Lee frames a 2014→2019 five-year path (My Investments, §2).
  • Counterexample — unnamed hardware: design, price, competition, capital and late timing caused shutdown (Edge).
• Evidence — Kauffman time heuristic, RI duration, frontier maturity and three cases triangulate the model (Jiemian).
• Limits, failure modes, and exception ledger — Five years is not a base rate; syndication/licensing may extend duration. No reserve schedule, extension history or forecast-error cohort exists.
• Evolution and contradictions — Early IoT optimism and cautious 2016 maturity precede explicit 2022 timing, improved articulation but unproved calibration.
• Modern VC translation (as of 2026-08-27) — Match robotics/AI infrastructure certification to strategic/project/credit capital. Misuse: rejecting science on a software clock or funding indefinitely. Researcher inference; moderate.

Design the team before scaling capital

• Rule — Treat company-building as role architecture: identify missing leaders, test shared authority/bad news, then use recruiting/board cadence before large follow-ons.
• Evidence status — Documented in hiSoft/NIU and founder witness; moderate-to-strong behavior, insufficient causal effect.
• Provenance and origin — Forbes reports hiSoft CEO recruitment; NIU documents Li Yan, crisis redistribution and quarterly boards (Forbes 2022; NIU).
• Mechanism and reasoning — Complementary operators convert founder vision; explicit roles/bad-news cadence let capital respond to reality.
• Historical and vehicle context — Scarce early China managers raised network value; long funds supported boards, while shared platforms obscure credit (AVCJ, p.23).
• When to use — Founder transition, hardware scale, multi-country expansion and crisis; avoid reflexive replacement/foreign templates.
• How to apply it — Map owners; independently reference; name next roles; agree bad-news packet/cadence; define founder versus board decisions; gate follow-on on team proof.
• Examples —
  • Documented application — hiSoft: outside-network sourcing and CEO recruitment (Forbes 2022).
  • Documented application — NIU: role redistribution, continued funding and quarterly boards (NIU).
  • Boundary — Ruangguru: founder empathy is selected witness, impact unquantified (Ruangguru).
• Evidence — Forbes/NIU support recruiting; Kauffman references; Ruangguru witness; SGX confirms board scale but not action (SGX, pp.1–2).
• Limits, failure modes, and exception ledger — Investor credit inflation, micromanagement and founder trust loss. No board minutes, executive attrition cohort or Lee-specific distress intervention at LAIX/Spark/EHang.
• Evolution and contradictions — Stable from hiSoft through NIU/Ruangguru; founder partnership conflicts with board power, with no public arbitration rule.
• Modern VC translation (as of 2026-08-27) — Separate AI research, product, infrastructure, safety and sales authority; measure whether leaders recruit stronger successors. Misuse: micromanagement as help. Researcher inference; moderate.

Close the private–public capital loop

• Rule — Ecosystems compound when private companies, local listings, research, institutional buyers, secondaries and later capital reinforce one another.
• Evidence status — Directly articulated in 2024 but forward-looking; moderate diagnosis, provisional outcome model.
• Provenance and origin — Lee/co-authors argue SEA needs local price discovery, comparable public companies and private/public feedback (Granite essay).
• Mechanism and reasoning — Listings create comps/liquidity; research/buyers improve price discovery; exits recycle talent/capital; growth/credit bridge timing.
• Historical and vehicle context — Earlier US/HK routes enabled exits; later geopolitics weakened them. Granite's venture/growth/credit platform responds structurally (rebrand).
• When to use — Ecosystem strategy, growth, secondaries and listing underwriting; not to hide weak equity/postpone write-downs.
• How to apply it — Map venues, coverage, buyers, secondaries and credit; require each vehicle solve a gap; define conflicts; track realized liquidity.
• Examples —
  • Documented thesis — SEA: essay diagnoses thin local comps (Granite essay).
  • Documented outcome — Geek+: HKEX trading supports a regional route, not the whole model (HKEX allotment results, summary).
  • Counterexample — Spark/LAIX: filed IPO can withdraw; completed IPO can delist (Spark; LAIX).
• Evidence — Granite essay/rebrand are related primary sources; HKEX/SEC independently show endpoints; RI supplies earlier economics/context.
• Limits, failure modes, and exception ledger — Multi-asset breadth may create allocation conflicts or support weak equity. No full cycle, realized returns or conflict policy.
• Evolution and contradictions — Unit of analysis expands from product globalization to market infrastructure; maturation/geopolitics are plausible motives, not proved.
• Modern VC translation (as of 2026-08-27) — Measure IPO quality, coverage, ownership, discounts, repeat founders and DPI. Misuse: AUM breadth as ecosystem impact. Researcher inference; provisional.

Evidence gaps and candidate models

“Brain and heart,” curiosity, persistence and entrepreneur learning are incorporated above rather than separate overlapping rules. Pricing/ownership, reserves, sell discipline and failure-rate models were excluded because writings, eight appearance notes, LP/SGX/SEC/HKEX records and adverse searches yield no reproducible thresholds or full denominator. Timestamped 996 audio confirms policy awareness but also Lee's admission of no special insight; without case-level applications, government alignment is a boundary, not a model.

What the evidence supports

The strongest support is for a repeatable method, not audited personal alpha: Lee states a future-led, problem-first process; direct audio adds user/employee checks, stage-adjusted proof and local re-underwriting; NIU documents 5,000 surveys, executive recruitment, crisis role allocation, boards and follow-on capital (EDB, sourcing and founder sections; NIU, §§2–7; Kauffman, 10:25–18:09). Investor-stated with documented applications; moderate: public evidence selects memorable cases and does not expose the passes or ordinary-company cohort needed to establish selection alpha.

Her China/Asia career, selected sponsorship, public boards, vehicle structure and several outcomes are independently bounded. EDB attributes hiSoft, 21Vianet, SinoSun and YY leadership; Lee's own 2020 essay names Geek+, Sensoro, Spark and Fengbian as 2019 investments she led; SGX establishes legal-entity board service but not source, price or economics (EDB, introduction; My Investments, §1; SGX schedule, pp.1–2). Contemporaneous record/investor-stated; strong for identity and bounded roles, insufficient for personal economics.

What could falsify the thesis

A complete role-coded ledger could show that partners or network access supplied most sourcing and decisions, that field checks rarely changed an IC outcome, or that non-outlier results underperformed a comparable opportunity set. Audited cash flows could show that YY/WPS dominate otherwise ordinary economics; contemporaneous passes could show weak calibration; ordinary board cases could show no systematic recruiting or governance effect. Those findings would falsify the favorable edge thesis. Conversely, dated IC records connecting maps and disconfirming evidence to repeatable decisions, together with role-adjusted cash outcomes across wins and losses, would weaken the skeptical brand/team/era explanation. Neither record is public.

The visible record already contains pressure tests: Bokee and unnamed social/hardware shutdowns, LAIX's delisting/private transaction, Spark's post-filing policy shock and EHang allegations/company rebuttal (996, 12:10–13:49; The Edge, “Success and failure”; LAIX 6-K; EHang response). These do not establish misconduct or process failure; they do disprove an inference from listing count to durable value.

Attribution and denominator audit

The 2025 SGX attachment contains 75 legal entries. The accepted split is 36 operating legal entries—26 past and ten current—and 39 exclusions—past GGV (Xiaole) plus 38 current non-operating entries, consisting of Temasek and 37 GGV/Granite/vehicle entities (SGX schedule, pp.1–2). The ten current operating entries are Cashield, FLT, FUSE, Spark, Palfish, Cloud First, XChange, Treelab, Next Gen Foods and TTMFS Singapore/Thunes. GGV (Petkit) is a vehicle/legal-entity exclusion, not an operating inclusion.

Adding 12 source-supported non-board names produces 48 searched legal/name cases. Combining the two legal entries Chukong Holdings and Xiamen Yaji Software into one supported company family produces 47 searched company families. Twenty received material case research; only eight meet the strict reconstruction bar—WPS, YY, hiSoft, Xiaomi, NIU, EHang, Spark and Geek+—so 39 remain unresolved. Airlook Airspace and Airlook Aviation remain separate pending alias proof; Fengbian and BeautyVector remain separate. 51 Credit is outside the person denominator and appears only as a GGV firm-risk control. A lifetime coverage fraction is not computable: SGX is a current/five-year directorship schedule, while The Edge's “almost 60” is approximate and neither is a lifetime investment ledger (The Edge).

This conservative count prevents board title from filling unknown source, sponsor, diligence, approval, negotiation, recruiting, follow-on or exit roles. The Glu filing likewise identifies Lee as one of nine GGV II investment-committee members with shared power, not a unilateral decision-maker (Glu Schedule 13D/A, Item 2 and notes 2–3). Xiaomi remains a Lee personal angel investment followed by a GGV investment; it is not collapsed into fund economics (Forbes 2022, Xiaomi section).

A filing-identity check resolves a prior conflation: Chukong Holdings F-1, management pp.148–149 (CIK 1596883) names Jenny Hong Wei Lee as a director since August 2012; the separate Sungy Mobile F-1/A, Management/Board of Directors (CIK 1584133) omits Lee from its directors and says JAFCO Asia Technology Fund III was not exercising its appointment right. Sungy is excluded from the person denominator; Chukong/Xiamen Yaji remains the supported company family.

Transfer limits for modern VCs

Transferable: decompose a future into bottlenecks; collect target-user and separate lieutenant evidence; change the founder rubric with stage; localize price, product, people and regulation; design roles before scaling capital; and match technical proof to fund duration. These are process hypotheses with observable inputs, not licenses to copy portfolio sectors. Researcher inference; moderate. Modern AI/robotics teams should replace demo enthusiasm with production retention, independent customer references, reproducible evaluations, safety/regulatory milestones and cash needed to reach the next discriminating proof.

Non-transferable without re-underwriting: early China mobile distribution, the integrated US–China platform, Lei Jun/GGV access, historical investigator-style personal checks, and an assumed IPO route. Granite's 2024 regional multi-asset design itself shows that market and vehicle conditions changed (Granite rebrand). The likely misuse is calling a future map diligence, treating local partnership as local authority, or using patient capital to avoid a write-down.

Unresolved evidence

Missing: a lifetime investment/pass denominator; check sizes, entry prices, ownership/dilution, reserves and proceeds; personal carry; IC votes and dissent; board minutes; complete executive-recruiting results; named shutdown economics; negative references; Spark/LAIX/EHang response records at Lee level; and systematic comparison of map predictions with passes and outcomes. No public source establishes a personal loss ratio. The Business China award biography is inaccessible/index-only—original HTTP 404 and derivative HTTP 502—so its indexed “fewer than five failed” phrase is not evidence.

Spoken evidence is unusually strong but bounded. Five public audio families were legitimately acquired and OpenAI-transcribed; no raw media or full transcript is retained. The Analyse publisher says a technical interruption lost roughly the entire second half, so 11:42 is the complete surviving published audio, not the complete interview (Analyse episode). The two distinct DealStreet 2022 appearances, grouped into one comparative analytical section, and DBS remain partial discovery routes. No public audio resolves pricing, reserves, IC votes or write-off mechanics.

Central-claim citation audit

Ordinary-claim citation audit

Selection rule, frozen before source re-opening: in learning order (_profile → sources → writings → talks → investments → investmentphilosophy → mental-models → review), take the first externally checkable non-thesis factual assertion carrying or requiring a source in each root; exclude claims already in the central table. Then continue in the same order with the next eligible assertion in _profile and talks to reach ten. Table rows and bullet clauses count as assertions; headings and citations alone do not.

Verification

The root set, note templates, source arithmetic, return syntax, attribution controls, denominator wording and relative links passed mechanical checks. The Astro site built T057 and all 15 repository tests passed. Eight canonical high-risk links were checked: Granite, NIU, SGX, Rhode Island, Analyse and EDB returned HTTP 200; SEC blocked bare curl with HTTP 403, but both filings opened through EDGAR/web extraction and the cited management sections were rechecked. Access failures and partial routes remain labeled rather than silently treated as read. No raw audio, media, transcript, cache or dependency artifact is retained.

Twenty-point acceptance evidence

Acceptance judgment

The package is fit as an evidence-bounded curriculum on thesis formation, local verification, team design and duration—not proof of Jenny Lee's personal alpha. The decisive discipline is attribution: filings and direct voice reconstruct several decisions much more confidently than they reconstruct ownership, cash returns, partner contribution or the ordinary-company denominator. Future IC records and cash-flow evidence should revise that boundary, not be backfilled from reputation.

Direct writings, speeches, and firm records

• ★★★ | Through the lens of VC | 2018-10-25 | edited Q&A/direct voice | A/B | full; deductive sector work, founder-product-market fit, brain-and-heart test, China/SEA comparison; question headings.
• ★★★ | My Investments and Exits | 2020-01-21 | GGV-origin Chinese essay | A/B | full; four personally led 2019 investments, WPS ninth led IPO, EHang self-rescue and attribution; §§1–3.
• ★★★ | Small world | recorded 2016; published 2017-03-07 | Q&A PDF | A/B | full; Discovery/Main design, portfolio services, localization and mobile/data thesis; printed p.23/PDF p.22.
• ★★★ | Bridging the Gap in Southeast Asia's Capital Markets | 2024-06-06 | co-authored essay | A | full; private/public loop, local price discovery and self-reported outcome counts; headings.
• ★★★ | China frontier-technology speech transcript | speech 2016; page 2016-08-11 | translated/reposted speech | A/B | full Chinese transcript; wearable maturity, transport, robotics, automation and commercialization; numbered sections.
• ★★★ | NIU diligence and crisis case | 2018-10-21 | GGV-origin Chinese interview | A/B | full; Series A lead, 5,000 surveys, recruiting, crisis, boards and follow-on; lines 56–223.
• ★★ | VCs are betting on the great Chinese fitness boom | 2016-07-07 | co-authored column | A/B | full; apps, device/data and outdoor-mobility market map; headings.
• ★★ | Internet of Things means never having to search again | 2015-03-18 | co-authored column | A/B | partial but substantive indexed body; cross-border IoT and device-interface forecast.
• ★★ | Southeast Asia and the next billion | 2020-01-15 | edited podcast transcript | A/B | full relevant blocks; local champions, fragmentation, infrastructure and fintech; speaker blocks 172–205.
• ★★ | Granite Asia rebrand | 2024-03-30 | firm announcement | A | full; regional scope, leaders and multi-asset transition; headings.
• ★★ | Granite Asia biography and timeline | current, accessed 2026-08-27 | official biography | A for identity/B for outcomes | full; career, focus, boards and current self-reported totals; profile/history headings.
• ★★ | Kellogg alumni profile | 2016 | direct-voice alumni profile | A/B | full; problem-first orientation and board help; headings.
• ★★ | How to choose the right VC | 2023 | firm/founder case | A/B | full; EdTech execution phase, founder empathy and board relationship; headings.
• ★★ | EHang Series B announcement | 2015-08-24 | company announcement with Lee quotation | A/B | full; GGV's then-largest Series A, follow-on and UAV thesis; announcement paragraphs.
• ★ | DBS and Granite Asia strategic partnership | 2026 | firm announcement | A | current multi-asset strategy lead; metadata/context only.

Talks, podcasts, and video

• ★★★ | Jenny Lee of GGV Capital on Being a VC in China | 2018-06-08 | 996 podcast | A | full 65:18 audio acquired through public Apple route and OpenAI-transcribed; publisher enclosure returned HTTP 404; timestamped note.
• ★★★ | The Shortest Interview with Jenny Lee | recorded 2015; posted 2015-08-02 | podcast | A | full surviving 11:42 published SoundCloud HLS audio acquired and OpenAI-transcribed; publisher says roughly half the interview was lost to a technical interruption; Malay orthography limits quotation fidelity.
• ★★★ | Secrets of Emerging Technology Ecosystem | 2022-04-06 | Kauffman Fellows podcast | A | full public 30:10 MP3 OpenAI-transcribed; timestamped note.
• ★★★ | Why EdTech is not winner-take-all | 2020 | podcast | A | full 43:20 audio acquired from public iTunes enclosure and OpenAI-transcribed; retired publisher transcript failed.
• ★★ | Power List Asia: Jenny Lee | 2017 | television interview | A | full 22:48 Brightcove audio acquired and OpenAI-transcribed; player explicitly reported no subtitles.
• ★★ | Funding winter introspection | 2022-12-21 | summit video | A/B | accessible video/summary; no full transcript exposed.
• ★★ | “Mr. Market is always right” | 2022-03-14 | summit report/interview | A/B | partial/paywalled; title, event context and accessible excerpt only.
• ★ | DBS Two in a Pod | current archive | podcast hub | A | discovery route; no stable episode-level record located.

Biography and independent press

• ★★★ | GGV Capital's Jenny Lee rises | 2022-09-08 | reported profile/interview | B | full; career, WPS/hiSoft/Xiaomi cases, team and reported economics; sections.
• ★★★ | How to invest in winners in China | 2013-05-10 | interview/profile | B | partial; accessible facts cover YY, hiSoft, 21Vianet and SinoSun.
• ★★ | Women in tech: Jenny Lee | 2019-05-25 | interview/profile | A/B | accessible body; boards as people work and early-company IPO count.
• ★★ | Capital for Asia, rooted in Asia | 2024-10-07 | profile/interview | B | partial; engineering background, bond penalty, platform and split.
• ★★ | How Jenny Lee became Asia's “Iron Woman” | 2025 | translated profile/interview | A/B | accessible; current AI/geopolitical reflection and 2005 scene.
• ★★ | The power of curiosity | 2020 | profile/interview | A/B | full; almost 60-company involvement and unnamed social/hardware shutdowns; “Success and failure.”
• ★★ | Current Forbes profile | updated 2025-12-10 | profile | B | current 21 $1B+/17 IPO claims; contains an erroneous 2012 Top-10 sentence, rejected against contemporaneous lists.
• ★★ | Business China award biography — original / indexed derivative | 2016 | award biography | B/C | inaccessible/index-only: original HTTP 404, derivative HTTP 502; “fewer than five failed” is an unverified search-index clue only.
• ★★★ | The women of the 2012 Midas List | 2012-06-01 | contemporaneous Forbes list feature | B | full/indexed body; Lee's debut at #94 and then-career basis.
• ★★★ | Cracking the boys club | 2015-03-25 | contemporaneous profile/interview | A/B | full/indexed body; #10 and highest-ever female placement; paired with Top 10 list identifying the first woman in the Top 10.

Deal, vehicle, governance, and regulatory records

• ★★★ | Glu Mobile Schedule 13D/A | 2011-07-22 | SEC filing | A | GGV II ownership; Lee one of nine IC members with shared power; Item 2 and notes 2–3.
• ★★★ | Temasek offering circular | 2024-07-22 | SGX-hosted filing | A | Temasek director since 2022 and named Xiaomi/WPS/JOYY/XPeng/NIU/Agora/EHang guidance; director biography.
• ★★★ | Rhode Island Treasury GGV LP presentation | data 2022-06-30; meeting 2023 | public LP deck | A/B | terms, construction, team, funnel and fund marks; pp.2–7.
• ★★★ | Spark Education draft F-1 | 2021 | SEC draft filing | A | Lee director since July 2019; directors/printed p.174.
• ★★★ | SGX Jenny Lee directorship schedule | 2025 | exchange attachment | A | exact past/present legal-entity board lists; pp.1–2.
• ★★★ | US House Select Committee VC report | 2024-02-08 | congressional report | A as official allegation/record | GGV/Megvii scrutiny; pp.11–13; no Lee sponsorship finding.
• ★★ | EHang Series B | 2015-08-24 | company announcement | A/B | GGV capital sequence and Lee quotation.
• ★★ | 2025 HKEX draft filing | 2025 | exchange filing | A | GGV VII entities and shared control among five partners; printed p.207.
• ★★ | 2023 HKEX filing | 2023-06-30 | exchange filing | A | shared control of Select, V and VII entities; substantial-shareholder note 2.
• ★★ | GGV closes $1.88B in new funds | 2018-10-18 | firm announcement | A | six-partner team, vehicles and founder witness.
• ★★ | GGV split coverage | 2023-09-22 | independent press | B | non-geographic funds complicated separation; policy context.
• ★★ | GGV China separation | 2023 | press | B | partial; platform and US-policy pressure; no individual-causation claim.
• ★ | NFX Signal profile | current | database | C | displayed/claimed count inconsistency; discovery only.
• ★★ | Geekplus final offer price and allotment results | 2025-07-08 | HKEX allotment-results announcement | A | stock code 2590 and dealings commencement on 2025-07-09; dated reconciliation clue for 16/17 IPO counts.
• ★★ | LAIX delisting 6-K / merger 6-K | 2022 | SEC filings | A | sub-$15m threshold and $1.90/ADS take-private; adverse board control.
• ★★ | 51 Credit Card investigation | 2019-10-22 | HKEX filing | A | official on-site investigation; firm-risk control outside person denominator.
• ★★ | EHang response / 2021 annual report | 2021–2022 | company/SEC-filed materials | A for company position | denial, price fall and shareholder suits without adjudication.
• ★★ | Spark IPO withdrawal record | 2021-12-01 | filing tracker | A/B | offering withdrawal after proposed listing.
• ★★ | China core-subject tutoring policy report | 2021-07-23 | Reuters report | B | regime-shock control for the EdTech thesis.
• ★★ | Cliffwater RISIC recommendation memo | 2023 | LP consultant memo | A/B | independent LP-side recommendation/risks; complements manager deck.

Coverage assessment

Searches covered Jenny Lee, Jennifer Lee, Jenny Hong Wei Lee, Lee Hongwei Jenny, 李宏玮, JAFCO/JAFCO Asia, Morgan Stanley, GGV/Jiyuan/纪源资本, Granite Asia, writings, transcripts, podcasts, filings, portfolio, failures, lawsuits, investigations, criticism and policy. The retained map has 53 catalog rows / 52 deduplicated source families after the repeated EHang announcement is deduplicated: 15 writing/firm records, eight spoken routes, ten biography/press controls and 20 deal/vehicle/regulatory entries. The priority writing denominator is 14: six ★★★ and eight ★★; all are deeply analyzed, thirteen in full and Fortune substantively in part. Five audio families were acquired and OpenAI-transcribed; YourStory supplies a sixth spoken family through an edited transcript. DealStreet and DBS remain partial discovery routes.

The SGX schedule supplies 36 operating legal entries: 26 past plus ten current. Its 39 exclusions are past GGV (Xiaole) plus 38 present non-operating entries (Temasek and 37 GGV/Granite/vehicle entities). Adding 12 supported non-board legal/name cases gives 48 searched legal/name cases; deduplicating Chukong Holdings and Xiamen Yaji Software as one company family gives 47 searched company families. Eight are strictly sufficiently reconstructed, leaving 39 unresolved. Airlook Airspace and Airlook Aviation remain separate pending proof of aliasing; Fengbian and BeautyVector remain separate; 51 Credit is a firm-risk control outside the person denominator. Coverage fraction is not computable because The Edge's “almost 60” is approximate and the SGX schedule is a five-year/current directorship record, not a lifetime investment ledger.

A filing-identity control rejects an earlier issuer conflation: Chukong Holdings F-1, management pp.148–149 (CIK 1596883) names Jenny Hong Wei Lee as a director since August 2012, whereas the separate Sungy Mobile F-1/A, Management/Board of Directors (CIK 1584133) omits Lee from its directors and says JAFCO Asia Technology Fund III was not exercising its appointment right. Sungy is therefore excluded; Chukong/Xiamen Yaji remains the supported family. This control is part of coverage, not an added catalog row.

Outcome counts require dates and definitions: the June 2024 co-authored essay says 16 IPOs; Granite's live biography (accessed 2026-08-27) and Forbes's 2025 profile say 17. Geekplus began HKEX trading on 2025-07-09, consistent with but not proof of the one-count increase. Midas chronology is controlled by contemporaneous Forbes pages: Lee debuted in 2012 at #94, then reached #10 in 2015 as the first woman in the Top 10; the current Forbes profile's “2012 Top 10” sentence is erroneous. Defensible completed peers are Sonali De Rycker and Masayoshi Son; comparisons remain dimension-bounded.

Source notes