Polymarket resolved markets: outcomes on 118,251 crypto up/down markets

Polymarket's crypto Up/Down markets resolve on a Chainlink reference price at a predetermined timestamp: each binary contract settles to Up or Down. Polymarket's Gamma API does serve closed markets, but it defaults to closed markets filtered out of most queries.

In the PolyOrderbooks archive, resolution is metadata on the row: order books, prices, and metrics for resolved markets carry the winning outcome and stay queryable forever.

What resolution looks like in the depth: across 298 resolved BTC 5-minute contracts, 90% of final-second order books had a single side and the winner's side rested a median 42,254 shares at 0.99 (settlement depth post). In a separate 80-contract study, 94% of final-ten-second snapshots had an empty bid or ask side.

Over the full archive, 5-minute books are one-sided 16.9% of the time and 3.24% of snapshots are crossed — both measured from these resolved markets.

• Backtesting — only resolved markets are used for scoring, so strategies are judged against what actually happened (overview).
• Settlement research — study final-second depth and repricing against the stored books.
• Citation — 897,192 snapshots across 805 resolved markets on Zenodo with DOI 10.5281/zenodo.22084114.
• Walk-forward testing — partition resolved markets by window and contract length for out-of-sample evaluation.

A resolved market is the end of the trade but the beginning of the evidence. On Polymarket, crypto Up/Down contracts resolve to Up or Down off a Chainlink reference price at a predetermined timestamp, and the archive keeps every resolved book queryable with the winning outcome attached to every row.

The resolved set is large enough for real scoring: 118,251 markets across eight coins at 5-minute, 15-minute, and 4-hour windows, which is the same corpus the backtest engine scores against. Because only resolved markets are used, every result is measured against what actually happened, not against a prediction.

The settlement record is where the books become readable: across 298 resolved BTC 5-minute contracts, 90% of final-second books had a single side and the winner's side rested a median 42,254 shares at 0.99, while a separate 80-contract study found 94% of final-ten-second snapshots had an empty side.

The citation path exists too: 897,192 snapshots across 805 resolved markets are published on Zenodo under CC BY 4.0, with a live DOI. That is the difference between quoting a vendor number and citing a dataset that an auditor can open.

A resolved corpus also settles the research timetable: because the outcome is a column, last month's markets answer this month's question without waiting for any future resolution event.

Because resolution is metadata rather than a filter, the same query returns live and resolved markets, and the outcome column simply arrives populated for markets that have settled — the archive treats the past and the present as one table.

Resolution metadata should be a filter, not a wall: the archive treats the outcome as a column, so a resolved series joins to its books, prices, and metrics in one query rather than a manual reconciliation.

Use resolved-only scoring as the honesty gate: any strategy report that cannot name its resolved corpus window is scoring against a guess, and the difference from a true corpus study is the entire settlement window.

The final-minute numbers are the ones to repeat with care: 76.2% of 5-minute final-minute snapshots carry an empty side, so a study that needs rests on the winner's side reads the 90%/42,254 facts, and a study that needs quotes should expect the losing side to be absent.

For a first look, pull one resolved contract and read the final-second frame next to its outcome — that single juxtaposition teaches the whole settlement microstructure.

• Crypto data coverage
• Backtest API
• Order book data
• Settlement mechanics
• Backtest AI