Polymarket Settlement Data

The study draws from 8,349 BTC five-minute contracts that settled between 2026-08-26 and 2026-09-23. From those, 298 resolved contracts were sampled, with 17,880 one-second order book snapshots taken across their final windows.

The finding: books ended one-sided at least 90% of the time, the winning side held 0.99 or better in 98.6% of observations, median top-of-book size at that price was 42,254 shares with a 90th percentile of 65,081, and both sides were never simultaneously empty across the sample.

• timestamp at one-second resolution within each final minute
• bid and ask price per frame, on the 0.01 grid
• top-of-book bid depth and ask depth in shares
• cumulative traded volume at each frame
• contract metadata: series, expiry, resolved outcome, reference price
• one-sided flag derivable as askDepth === 0 or bidDepth === 0

The archive serves BTC 5-minute books at 250ms; query the series slug with a time range covering open-to-settlement. For settlement-specific rows, filter to the final 60 seconds and mark one-sided frames from the recorded depths.

A free Starter window reaches 3 days of history — enough to reproduce the methodology on a fresh set of markets; paid windows cover the months a formal study needs. The data-profile of BTC five-minute markets is described in the per-coin page.

• Pull the final five minutes of a resolved BTC 5-minute market at 250ms and separate the last sixty seconds; that is 240 frames for the settlement window you will analyze.
• Tag each final-window frame with its one-sided state (bid-only, ask-only, or crossed) and count the empties; the measured families show 90%+ one-sided frames and the specific 3-of-18 ask-absence is the extreme case.
• Recompute the settlement price independently from the ladder (the best executable bound at the reference time) and compare it to the reported outcome; agreement is the discipline a settlement study should never skip.
• Roll the result into a per-contract settlement table: slug, window, one-sided share, last executable floor, outcome — the sentence-length summary of each market's ending.

The 8,349 BTC five-minute contracts settled between 2026-08-26 and 2026-09-23 are the canonical settlement corpus, with 298 resolved contracts spanning the sample; the size distribution (median 42,254 shares, p90 65,081) means settlement flow, not exotic ordering, is the norm.

Round-grid pricing closes the story: 98.6% of sampled frames rest at 0.99 or 0.995, so a settlement study is most often measuring whether the executable floor actually cleared the final reference — a narrow, measurable question instead of a fuzzy one.

Settlement is the venue's moment of truth, and the archive makes it the most reproducible moment in the study catalog: 240 final frames per contract, one-sided ratios per market, and a floor that either caught the reference or not.

The honest frame for settlement research is: reference, window, executable floor, and outcome — any claim that omits one of the four is a claim being kept from the record it claims to describe.

Start every settlement study from the final-minute flags rather than from the reported outcome; the flags are independent, measured, and they survive even when the reported field does not.

Settlement data is the terminus of every other series: books, prices, and metrics all feed the final-minute truth, which is why this page sits beside the replay and repricing pages rather than inside them.

Each measured case in this page is stored at 250ms with the same field names as every other dataset, so the switch from one page's material to another is a query-parameter change, not a migration.

That shared format is the reason a family can be studied across settlement, repricing, replay, final-seconds, and cross-market frames with one pull and one schema.

• Research blog
• Order book data
• Backtesting data
• Crypto data