About ThesisGap

Over the last two years, discussions in crypto and US politics have leaned increasingly on Polymarket probabilities — not because they're necessarily “more accurate”, but because a consensus backed by hundreds of millions of dollars of real money is harder to fake than any single analyst's forecast. But that signal is nearly unusable for a retail investor directly: Polymarket has thousands of active markets at any time, and bridging the semantic gap between your thesis and the right market often takes 30 minutes of manual searching.

ThesisGap solves that matching problem. You write the investment thesis you actually care about. The AI decomposes it into verifiable sub-claims, retrieves matching contracts from the market pool, and labels each as supporting, contradicting, or neutral. The output is a comparable Gap number — not “here's your answer”, but “here's where you and consensus disagree.”

That distinction matters. Markets aren't always right — but when you disagree with the market by 20 percentage points, you at least know there's something worth re-examining. When you align with it, you know your thesis is in step with hundreds of millions of dollars of independent observers, which is a signal in itself.

I'm Yan, born 1993, based in Shenzhen (Hong Kong Talent visa in progress). ThesisGap is a solo project — no funding, no co-founder, no board. Full-time on this one thing.

Before ThesisGap I spent 5+ years as a full-stack engineer, primarily on B2B SaaS and data-analysis tooling. I'm also a moderately active investor myself, using three frameworks (Polymarket, geopolitics via Jiang Xueqin, Zhou Jintao's Kondratiev long-wave theory). My positions concentrate in uranium / nuclear, AI compute infrastructure, and gold leverage. ThesisGap is a tool I personally use daily, which keeps product decisions aligned with a real investor's actual workflow rather than what sounds clean in a pitch.

Why should you trust me? Honestly, at this stage you can't fully — I don't publish a verifiable track record, and ThesisGap itself is still 0-to-1. But you can verify these things:

• The methodology page spells out every calculation and model choice — no black box.
• The examples page shows anonymized real theses so you see what the input/output looks like.
• Every market the product cites includes its Polymarket URL, live probability, and 24h volume — you can reconcile against Polymarket directly.
• Failed analyses are auto-refunded — I have no incentive to push fake analyses.

1. No buy/sell recommendations, ever. ThesisGap outputs the objective “how much do you disagree with the market” number, not advice on what to do. Trading decisions are yours, responsibility is yours.
2. AI limitations are labeled in the UI. Sub-claims that didn't match a market get an amber AI-only card explicitly labeled “not market-backed”. Users always know which conclusions are $1B-backed versus pure model inference.
3. Failures are transparent. Auto-refund. AI decomposition fails, zero valid sub-claims produced, background timeout — credit goes straight back. You don't pay for bad outputs.
4. No subscriptions. Buy credits per-use, never expire. No recurring charges, no “forgot to cancel” trap, no 7-day-free-trial-then-autobill shenanigans.
5. Your data is only used to improve your own analyses. No external model training, no selling to data brokers. Supabase RLS ensures your theses and results are visible only to you.

For developers who want to know how it's built:

• Frontend + API: Next.js 15 App Router, React 19, TypeScript, Tailwind CSS
• Deployment: Vercel (all public pages SSR, API routes serverless)
• Database + auth: Supabase (PostgreSQL + RLS + magic-link email)
• AI: A dual-model LLM ensemble — the primary model handles decomposition and primary classification, a secondary model runs parallel classification for cross-validation, and a separate model handles fallback analysis for unmatched sub-claims
• Market data: Polymarket CLOB API (/markets endpoint + custom semantic ranker)
• Payments: Stripe Checkout (one-time payment, webhook auto-grants credits)

Product feedback, bug reports, methodology discussions, or partnerships: yan@thesisgap.com. I read and reply to every email personally.

X and GitHub accounts are not yet public; they'll ship in Q2 2026.