Evidence

• — Knowledge Project Ep. #226 (Shane Parrish): 'The #1 characteristic is earnestness — incredibly sincere... what you see is what you get.' [P=3 V=3 A=3 C=3 total=12]
• — My First Million: 'YC picks winners at higher rates because of earnest and formidable founders... Paul Graham and Jessica Livingston fostered both' [P=3 V=3 A=2 C=2 total=10]
• — Recurring across YouTube monologues, Acquired, Vanta — earnestness as the YC selection criterion in every interview Tan gives [≥ 5 cross-context occurrences]

Application

Read the founder, not the deck. Are they doing this because they want to do it, or because they want to have done it? Earnest formidable founders tend to bypass pitch theater in favor of describing what they've built and what they're trying to learn next.

Limits

Earnestness as criterion is hard to falsify and risks post-hoc rationalization — every successful founder gets retroactively labeled earnest, every failure gets relabeled. Doesn't help calibrate between two earnest founders solving similar problems. Less useful for B2B enterprise deals where 'earnestness' is filtered out by procurement.

Evidence

• — Tweet (verbatim): 'Remember startups generally start with a thin edge of the wedge. The early game is going from zero to one... Alphabet wasn't built in a day.' [P=3 V=3 A=3 C=2 total=11]
• — Vanta Frameworks for Growth: Datacurve case study — 'pivoted from undifferentiated AI workflow to their actual expertise, achieving almost eight figures in about nine months' [P=3 V=3 A=3 C=2 total=11]
• — Tweet (verbatim): 'To overturn boulders, find a thin edge of a wedge and then apply force. It's not easy but it is possible.' [P=3 V=3 A=3 C=3 total=12]
• — Recurring framing in YC office hours, Inc. column 'The 2 Things About Your Start-up Idea That Actually Matter' [P=3 V=3 A=2 C=2 total=10]

Application

Ask: who is the first 100 specifically, today? What does the wedge look like — narrow group, 10x better on a sharp axis? Without the wedge there is no zero-to-one motion.

Limits

Some great products started as broad consumer plays without a wedge (Twitter, Instagram) — the wedge frame is high-precision but lower-recall on the consumer-network category. Apply gently to mass-market product / network-effects bets.

Evidence

• — Vanta Frameworks for Growth: PMF signal as 'oh yeah, this is awesome, but it would be even more awesome if you did X, Y, and Z' [P=3 V=3 A=3 C=2 total=11]
• — Recurring framing in YC office hours videos, founder coaching across portfolio post-mortems [P=3 V=3 A=2 C=2 total=10]
• — Implicit in the Posterous post-mortem: Posterous shipped to a small power-user crowd that didn't translate into mass network — vitamin trap [P=3 V=3 A=3 C=2 total=11]

Application

Is this a vitamin or a painkiller? Who is in pain right now? Will they pay this week to remove the pain? If users say 'cool, I'd try it sometimes' — that's a vitamin and it will likely die in slow growth.

Limits

Some great products started as vitamins (Twitter, Instagram, TikTok) and built into painkillers via network effects. The frame is high-precision but lower-recall. Don't reject every social/entertainment product because nobody is 'in pain'.

Evidence

• — Posthaven essay 'Sell? Die? No. Grow profitably' (Ooshma Garg / Gobble): 'Default alive gives you control' [P=3 V=3 A=3 C=2 total=11]
• — Same essay (verbatim): 'one of the biggest lies that startups tell themselves and their boards is that they can turn off marketing and be profitable' [P=3 V=3 A=3 C=2 total=11]
• — Acquired ACQ2: described 'percentage ownership matters a ton... we learned this the hard way' — Initialized's own ZIRP-era expansion was a counter-example he drew lessons from [P=3 V=3 A=3 C=3 total=12]
• — Recurring across YouTube monologues, Knowledge Project, Mixergy [P=3 V=3 A=2 C=2 total=10]

Application

Can this team get to ramen-profitable on a small base of paying users? Are cohort retention numbers separated from aggregate growth? If the team needs the next round to be profitable, they have a fundraising plan, not a business.

Limits

Some categories (deep tech, regulated industries, infrastructure) require multi-year capital before unit economics close. The default-alive frame works best for SaaS, consumer, and dev tools; weakens on hardware, biotech, science.

Evidence

• — Vanta Frameworks for Growth (verbatim): 'You go out there and the universe will smack you' [P=3 V=3 A=3 C=2 total=11]
• — Posthaven essay 'How founders can build Trust & Safety teams' — Steve Kirkham/Eric Levine of Airbnb: 'Knowing who you're doing business with is the first step' [P=3 V=3 A=2 C=2 total=10]
• — Recurring across YouTube founder interviews (Apoorva Mehta / Instacart, Brian Armstrong / Coinbase, Ooshma Garg / Gobble) — the manic listening pattern [P=3 V=3 A=2 C=3 total=11]
• — YC's canonical 'How to Talk to Users' (Eric Migicovsky talk) is the YC-orthodox version Tan repeats and amplifies [P=2 V=3 A=3 C=2 total=10]

Application

Does the founder have evidence of weekly user contact? Are decisions traceable to user signal? When the founder describes the user's problem, do they use the user's words or their own paraphrase? Beware the founder who 'knows' what the user wants without asking.

Limits

User research can be over-applied — some products (radically new, behaviorally novel) require Henry-Ford-like 'faster horses' overrides. Tan acknowledges this implicitly through his Brian Chesky / Founder Mode admiration — the founder-as-creator can override user input when the conviction is earned.

Evidence

• — Tweet (verbatim, March 2025): 'For 25% of the Winter 2025 batch, 95% of lines of code are LLM generated. That's not a typo. The age of vibe coding is here.' [P=3 V=3 A=3 C=2 total=11]
• — Knowledge Project Ep. #226: 'The model itself is not the moat — the evals are the moat' [P=3 V=3 A=3 C=3 total=12]
• — Mixergy: 'YC companies routinely get to $10-20M ARR in 10-20 months — literally never happened before in software' [P=3 V=3 A=2 C=2 total=10]
• — Tweet (verbatim): 'It's not just about the prompting, or your understanding of systems. It's about your understanding of your users, and your tenacity with which you pursue making something people want.' [P=3 V=3 A=3 C=2 total=11]

Application

Is this a small AI-native team or a 50-person legacy build? What are the golden evals — the truth-telling test set? When the next model drops, does this product get better or get killed? Does the team understand its users better than the model can without them?

Limits

AI-native shape thesis is most current; could be obsoleted by a foundation-model release that absorbs the application layer. Tan publicly bets it won't, but the lens has the shortest temporal half-life of his frame set.

Evidence

• — Social Radars 'Founder Mode' (Jessica Livingston): jazz analogy — 'managers usually can't play jazz because once they deviate from the scales or do something that breaks convention, they are fired' [P=3 V=3 A=3 C=2 total=11]
• — Pre-publication reviewer of Paul Graham's 'Founder Mode' essay (Sep 2024); co-organized Founder Mode event for YC alumni [P=3 V=3 A=3 C=3 total=12]
• — Vanta Frameworks for Growth: 'high agency is something that can be learned and is learned quite frequently' — water-around-obstacles framing of agency [P=3 V=3 A=3 C=2 total=11]
• — Recurring in YouTube monologues post-Sep-2024 [P=3 V=3 A=3 C=2 total=11]

Application

Does this founder operate as creator-with-legitimacy, or as a process manager? Are they in the unglamorous details? Are they playing jazz, or playing scales? In an AI-native team, are they doubling down on hands-on or delegating to managers who weren't there at Day 1?

Limits

Founder Mode can romanticize founder cruelty / micromanagement when applied without judgment. Tan implicitly acknowledges this by citing the Posthaven essay 'Micromanagement is toxic: Delegation is the cure (6 simple steps)' — the same author who champions founder-hands-on also champions delegation. The tension is unresolved.