Polymarket Strategy Replay Data

Where a backtest prices fills, a strategy replay prices decisions: for each 250ms frame where the rule fires, the replay records the book state, the fill the rule would get, and the outcome that market eventually settled. The output is a per-decision dataset, not a single equity curve.

That per-decision record is the analytic difference: it lets a study decompose a strategy into its winning and losing decision contexts instead of judging the aggregate.

The corpus is the resolved set of the archive — 118,251 resolved markets across eight coins and the sports families — each with its full 250ms history and outcome. A strategy run across that corpus is a study of ~every tradable instance the venue has printed.

The honest scope statement accompanies every replay: sample of markets, window, and the flag treatment are reported with the results, exactly as the methodology requires.

Run the rule against the books endpoint, store the per-decision rows, then aggregate by context — one-sided frames, bound-resting frames, reprint frames — to find where the edge actually comes from.

The archive ships the per-decision material (books, prices, metrics, tape, outcome), so a strategy replay never has to reconstruct a frame from a summary.

Replay data measures decisions in hindsight; it cannot invent the future. The honest use is calibration — how often the rule's context produced the outcome — and the honest ceiling is the reconstruction benchmark the replay family publishes.

• Replay how-to
• Backtest data
• Fill simulation