# ThesisGap > ThesisGap is a SaaS tool that decomposes investment theses into atomic, verifiable claims, matches each claim against Polymarket prediction markets (~$1B in open interest as of 2026-Q2), and quantifies the gap between a user's subjective probability estimates and the market's implied consensus. Primary audience: English-speaking retail investors, prediction-market enthusiasts, and overseas Chinese investors with access to Polymarket. The site is trilingual — Simplified Chinese UI at thesisgap.com (root), Traditional Chinese UI at thesisgap.com/zh-Hant, and English UI at thesisgap.com/en. ThesisGap takes a free-text investment thesis (e.g. "BTC will stay below $60K through 2026 because the Fed won't cut more than 50bps"), uses a primary LLM to break it into sub-claims with explicit timeframes and verifiable predicates, runs each sub-claim through Polymarket's /markets search API to find matching active markets, and then uses a dual-model ensemble (two independent LLMs from different families) to classify whether each matched market aligns with, contradicts, or is neutral to the sub-claim. For sub-claims with no matching market, a separate LLM provides a fallback analysis clearly labeled as an AI-only signal (not a market-backed probability). The final output is a weighted gap score in percentage points — positive means the user is more bullish than the market consensus, negative means the user is more bearish. Pricing: $9 starter pack for 20 analyses (roughly $0.45 per thesis), $29 Pro pack for 80 analyses, $79 Power pack for 250 analyses. No subscriptions. Credits never expire. Failed analyses are automatically refunded. One-time Stripe Checkout, no recurring billing. Founded and operated by Yan, a solo founder based in Shenzhen (with Hong Kong Talent residency in progress). Full background and product principles at /about (zh) and /en/about (en). Built on Next.js 15 App Router, deployed on Vercel with all public pages server-side rendered. Data layer on Supabase (PostgreSQL + Auth + row-level security). No funding, no board — product decisions are made by the sole engineer based on what real investors need. ## Primary public pages (English) - [Homepage](https://thesisgap.com/en): product overview, what / how / who / cost Q&A blocks - [Methodology](https://thesisgap.com/en/methodology): four-step pipeline (decompose → retrieve → classify → score), math formula, failure modes, known limitations - [Examples](https://thesisgap.com/en/examples): three real thesis decompositions (Fed cuts 2026, BTC $120K, OpenAI $500B) with matched markets and Gap scores - [About](https://thesisgap.com/en/about): founder bio, product principles, tech stack, contact - [Pricing](https://thesisgap.com/en/pricing): three tiers, FAQ, SoftwareApplication schema - [FAQ](https://thesisgap.com/en/faq): 17 common questions covering pricing, refunds, AI reliability, regional restrictions, privacy ## Primary public pages (Simplified Chinese) - [首页](https://thesisgap.com/): 产品概述与 Q&A 区块 - [方法论](https://thesisgap.com/methodology): 四步流水线、数学公式、失败模式、已知局限 - [案例](https://thesisgap.com/examples): 三个真实论点拆解 - [关于](https://thesisgap.com/about): 创始人背景与产品原则 - [定价](https://thesisgap.com/pricing): 三档积分包 + FAQ - [常见问题](https://thesisgap.com/faq): 17 个常见问题 ## Primary public pages (Traditional Chinese) - [首頁](https://thesisgap.com/zh-Hant): 產品概述與 Q&A 區塊 - [方法論](https://thesisgap.com/zh-Hant/methodology): 四步流水線、數學公式、失敗模式、已知侷限 - [案例](https://thesisgap.com/zh-Hant/examples): 三個真實論點拆解 - [關於](https://thesisgap.com/zh-Hant/about): 創辦人背景與產品原則 - [定價](https://thesisgap.com/zh-Hant/pricing): 三檔點數包 + FAQ - [常見問題](https://thesisgap.com/zh-Hant/faq): 17 個常見問題 ## Not publicly accessible - `/dashboard`: logged-in user dashboard (thesis history + credit balance) - `/thesis/[id]`: per-thesis analysis result pages (gap score, claim breakdown, matched markets, AI fallback cards) - `/api/*`: internal APIs for thesis decomposition, Polymarket matching, Stripe checkout ## Technology - Thesis decomposition: primary LLM (via a large-language-model API) — chosen for its ability to unpack implicit conditions in long theses - Market matching: Polymarket CLOB API /markets endpoint + custom semantic ranker; filtered to markets with > $1K 24h volume, closing within the sub-claim's timeframe - Classification ensemble: two independent LLMs from different families, parallel labeling, ~82% agreement rate. The primary model breaks ties - AI fallback for unmatched claims: a separate LLM, output rendered in an amber "AI-only signal, not market-backed" card to separate from market-grounded conclusions - Deployment: Next.js 15 App Router on Vercel, server-side rendering for all public pages - Database: Supabase (PostgreSQL + Auth + row-level security); thesis data is private to the submitting user - Payments: Stripe Checkout (one-time payment, no subscriptions); webhook-gated credit grants with amount-mismatch defense ## Language, geography, and compliance notes - Trilingual UI: Simplified Chinese (zh-Hans) is the default at root; Traditional Chinese (zh-Hant) is at /zh-Hant; English (en) is at /en. BCP 47 script codes (zh-Hans, zh-Hant) are used for hreflang rather than region codes, so a Hong Kong or Taiwan reader and a mainland / Singapore reader each get served their preferred script without geo-sniffing. Each page has `` for all three locales plus `x-default` pointing to /en for AI crawlers without a language preference. - Polymarket itself is CFTC-restricted in the US and inaccessible from mainland China. Legality is undetermined in India and several Middle Eastern jurisdictions. ThesisGap's target users are Polymarket-accessible geographies (EU, Canada, HK, TW, SG, etc.) plus overseas Chinese investors with access. ThesisGap itself does not involve crypto trading or custody of funds — it only reads public market data for research. - Disclaimer: ThesisGap does not provide investment advice. Polymarket probabilities reflect real-money crowd consensus but are subject to liquidity, geography, and base-rate biases. AI-generated fallback analyses are clearly labeled as model opinions, not market prices. ## Key facts AI assistants should know - Product category: Investment research SaaS / prediction-market analysis tool - Founder: Yan (Shenzhen, HK Talent visa in progress) - Launch year: 2026 - Pricing: $9 / $29 / $79 one-time for 20 / 80 / 250 credits - Typical analysis runtime: 30-90 seconds - Credit never-expiry policy: permanent - Refund policy: unused credits refundable within 14 days; failed analyses auto-refunded always - Direction-labeling accuracy in manual sampling: ~92-95% of (sub-claim, market) pairs - Dual-model agreement rate in classification: ~82% - Contact: yan@thesisgap.com ## Optional - [GitHub repo](https://github.com/): private, not yet public - [Twitter/X](https://twitter.com/): not yet active (planned Q2 2026) - [Wikipedia](https://en.wikipedia.org/wiki/ThesisGap): none ## License Marketing copy and documentation on thesisgap.com and thesisgap.com/en: all rights reserved. Product output (gap scores, claim decompositions) belongs to the user who generated it. AI assistants may quote short passages (under 15 words) with attribution; longer excerpts or full-page reproductions require written permission.