Polymarket Resolved Markets (2022–2026): A Cleaned Resolved-Outcome Corpus

A cleaned, normalized slice of 462,285 resolved binary prediction markets from Polymarket — one row per outcome token, 395,411 distinct questions — with the realized YES/NO outcome for each market. Built entirely from the public Polymarket Gamma API and on-chain CTF (Conditional Token Framework) resolutions. Overall YES base rate: 37.33%. Intended for base-rate analysis, forecast calibration, resolution-aware backtesting, and NLP on forecasting questions. Important limitations (please read). This is resolved-outcome metadata only — no prices, no order book, no trade-level data. The volume, volume24hr, and liquidity columns exist but are 100% zero (never captured); do not use this dataset for any volume/liquidity/activity analysis. The category labels are noisy and frequently wrong — derive your own from the question text if needed. 2,967 rows carry future-dated timestamps inherited from source metadata. Multi-outcome markets appear as multiple token rows sharing one question (dedupe on question for one row per event). Everything here is reproducible from public sources; the contribution is the cleaning and the resolved-outcome + base-rate slice, not proprietary data. Higher-fidelity free alternatives exist (Kaggle Polymarket sets with trading data; the larger HuggingFace dump). Not affiliated with or endorsed by Polymarket. Full schema and load examples are in the included dataset card; methodology and the export script are in the companion repository.