Polymarket Data Fidelity

High-fidelity data keeps the three layers aligned — reference, book, and tape on one timestamp grid — and does not silently repair the venue's irregularities. The archive treats one-sided and crossed snapshots as observations, flags them, and leaves them in the series.

The fidelity test is reconstruction: can a study replay the book at a past second and get the same levels, sizes, and outcome the venue printed? Measured where it counts — replay reproduces 36.6% of on-chain settlement outcomes from recorded books alone.

Fidelity shows up as the venue's true texture: one-sided snapshots at 16.9% of 5-minute granularity, crossed states at 3.24%, executable prices on the 0.99/0.995 grid 98.6% of the time, and 76.2% of final-minute books with an empty side.

Low-fidelity data would flatten exactly these numbers into a smooth mid series; the archive's integrity about them is the product.

Every downstream claim in a research paper inherits the fidelity of its data: a backtest on filtered data reports costs that were never real, and a study that keeps one-sided flags can measure that bias directly.

The archive's guarantees — 250ms cadence, full ladder, per-row flags, aligned outcomes — are exactly the properties a reviewer tests, and they are documented per-dataset.

Fidelity is a discipline, not a marketing term: the archive publishes its defect flags, its capture cadence, and its reconstruction rate so reviewers can check — the same standard this site demands of every data source it compares with.

• Data quality
• Replay data
• Footprint