Polymarket Resolution Archive — every resolved market, its outcome, and its on-chain settlement record
by AlertAudit
$49
There's plenty of free Polymarket data. If you want raw bulk trades, Jon Becker's open dataset is excellent and you should use it. So why does this exist?
Because backtests and calibration research don't run on raw data — they run on ground truth: which markets resolved, to what outcome, when it became final, and whether the resolution was contested. That layer is scattered across the Gamma API, the CLOB API, and the UMA Optimistic Oracle's on-chain state — and joining it is a week of work everyone keeps redoing. This archive is that join, done, documented, and current:
markets.csv — 509,357 resolved markets: question, event tags, outcomes, winning outcome, final settlement prices, volume, creation/resolution timestamps. Coverage runs January 2022 through 2026-07-22. The label set for any forecasting benchmark or calibration study. Scope rule, stated plainly: every resolved market with ≥$1,000 lifetime volume that traded for at least 24 hours. Polymarket's recurring sub-day price markets (hourly crypto up/down etc. — 100K+ resolving per month in 2026) are excluded by these two documented floors: they're a different asset class that would bury the event markets this dataset is for. Need that tail for a microstructure study? Email — custom builds are possible.
provenance.csv — 3,424 on-chain UMA Optimistic Oracle settlement records, joined to markets: who proposed the outcome, whether it was disputed and by whom, bond and reward sizes, request→proposal→settlement lags, and the settlement tx hash — verifiable on Polygon. This table exists nowhere else as a download. (Scope note: UMA settles most non-sports markets; markets resolved outside the UMA oracle carry metadata + outcome but no provenance row.)
prices_daily.csv.gz — 1,529,475 daily price rows per outcome token (19,808 markets above a $10K volume floor), so outcome labels come with the price path that led to them. Daily granularity — for tick/orderbook microstructure, use the free/intraday datasets; that's not what this is.
DATA-DICTIONARY.md + sample-analysis.ipynb — every column documented, plus a worked notebook: the calibration curve (price 30 days out vs. realized outcome) and UMA settlement-lag/dispute analysis, running unchanged on teaser and full data.
Why "current" matters: Polymarket's April 2026 contract migration (CTF Exchange V2) broke most community pipelines and stranded static dumps at their build dates. This archive is exported from a live, post-migration pipeline — the one behind calld.io — and refreshed builds are released periodically. Your purchase includes the next refresh.
Who this is for: researchers studying prediction-market calibration or resolution mechanisms, quants backtesting event strategies who need trustworthy outcome labels, builders of forecasting benchmarks (LLM-forecaster evals need exactly this ground truth).
Try before you buy: free teaser slice (top 500 markets by volume, same schema, same notebook) on Hugging Face: https://huggingface.co/datasets/calld/polymarket-resolution-archive-teaser
FAQ
How is this different from the free datasets? They ship raw markets/trades. This ships the resolution layer: outcome labels + the on-chain settlement audit trail, joined and documented. Different job. Use both.
Update cadence? Snapshot builds, refreshed periodically (build date in meta.json). One purchase = current build + next refresh free.
License? Single purchaser, internal use (research, backtesting, modeling, benchmark construction). No redistribution/resale of the files.
Refunds? If the data doesn't load or coverage is materially below what's stated, full refund, no questions.