Polymarket liquidity data: depth, spread, and what settlement shows

Liquidity data here means what was actually resting on the book — every bid and ask price level with its size, captured every 250ms. PolyOrderbooks stores that as full L2 ladders and pre-aggregates it into spread, total bid depth, total ask depth, and volume-weighted average price on the metrics endpoint.

Because the underlying books are stored, every metric can be reproduced from the raw snapshots rather than taken on trust. The open 897,192-snapshot dataset exists so the rates below can be verified.

• Spread — best ask minus best bid on the 0–1 probability scale, resample-friendly (60-second resampling removes transient quote noise).
• Depth — total bid and ask size, from top-of-book to cumulative ladders (depth guide).
• VWAP and alignment — volume-weighted average price, with prices and books sharing the same timestamps so everything joins in one DataFrame (metrics guide).
• All at 250ms-to-1d query resolution, on every plan.

Across 298 resolved BTC 5-minute contracts (17,880 one-second snapshots), 90% of final-second order books had a single side, and 87% were already one-sided in the last ten seconds; the winner's side rested a median 42,254 shares at 0.99 (settlement depth post).

A separate 80-contract study found 94% of snapshots in the final ten seconds had an empty bid or ask side (final seconds post). This is why fill models that assume two-sided depth misprice the close.

Liquidity data on this venue means what actually rested on the book: every bid level and ask level with its size, captured every 250ms, and pre-aggregated into spread, total bid depth, total ask depth, and volume-weighted average price on the metrics endpoint. Because the raw books are stored, those aggregates are reproducible rather than handed down.

The measured rates give the shape: 5-minute markets are one-sided 16.9% of the time, 3.24% of snapshots are crossed, and across 298 resolved BTC contracts 90% of final-second books had a single side, with the winner's side resting a median 42,254 shares at 0.99. Those numbers, not an opinion page, are what liquidity really looks like here.

Put the same lens on any market: pull the metrics for one contract, watch the spread compress into the final minute, and count how often a side simply disappears. The settlement window — 76.2% of final-minute snapshots have an empty side — is where the book stops being a market and becomes a verdict.

The honest workflow is to read metrics as pointers and then verify in the raw ladders: the endpoint answers "how deep and how wide", and the stored snapshot answers "was the executable side actually there". Both take the ten-minute route on the free Starter plan.

The free download is the fastest orientation: one CSV of snapshots opens in any tool and already answers how many seconds were one-sided, crossed, or bound-hugging, before a single API call is made.

Separate the two liquidity facts that marketing merges: tight spreads say the two-sided book is near, while one-sidedness says the venue is frequently only half there — the pair explains more failures than either alone.

Use 60-second resampling to strip transient quote noise when comparing across days; on a 250ms grid, that resample is cheap and it makes the comparison legible.

Settlement concentration is the strongest signal in the set: a winner side resting 42,254 shares at 0.99 is not depth in the economic sense, it is the market conceding.

Whenever a spread seems suspiciously stable across an event, diff the underlying snapshots — a metric computed from a reconstructed book inherits the 67.8% divergence risk.

• Order book data
• Order book archive
• Resolved markets
• Microstructure data
• Trades data
• Final seconds study