Polymarket Market Making

Sports books carry 35–36 price levels per side on a connected ladder, but the useful depth sits at the top: the September BTC sample shows median trade size of 42,254 shares with a p90 of 65,081, while top-of-book quotes move in single-tick increments (0.5 cents, 1 cent on the ladder).

One-sidedness is the rule, not the exception: at least 90% of sampled 250ms frames had size on only one side, and 98.6% of frames priced at 0.99 or 0.995 — the market is a ladder with a habitually missing leg.

Depth is the first casualty of a repricing. On the August 23 event, ETH bid depth fell 96% and ask depth rose 491% across the same window that lower bounds on ETH, BNB, DOGE, and SOL collapsed within 26 seconds — a maker's inventory and quotes are re-valued in the same second the reference moves.

• Auction the spread: with the book often single-tick wide but one-sided, a resting quote that becomes the only two-sided frame gets filled on both sides at different moments — favorable flow pays the spread.
• Repricing flow is directional and fast: capturing the 65-second ETH repricing means quoting the ladder, not re-quoting the mid, because the mid is a summary and the ladder is what trades.
• Settlement flows concentrate in the last minute: with ask presence in only 3 of 18 one-second frames in the documented sample, a maker's true final-minute inventory risk is ladder exposure, not book depth.
• Fill-flow cross-checks: 4 tracked fills versus market prices in the measured study slipped the wide end of the quoted spread at reprice spikes — quote sizes, not just prices, are the variable under your control.

Baseline from the measured book: assume ≥90% of frames are one-sided, quote in 0.5c ticks, size at a fraction of the median trade, and re-quote on a reprice trigger rather than a timer.

Use the archive to replay your quoting logic against history before capital moves: the replay study reproduced 36.6% of the on-chain reconstruction from 1,498 captured frames in the 2026-05-02 15:00 hour — enough to tune size and re-quote thresholds, not enough to trust a black box.

Instrument differentiation is free data: sports books show consistent 35–36 levels both sides; Up/Down books vacillate between a full ladder and a single level. Quoting policy should differ by series, and the archive records both in the same schema.

The maker's constraint is not latency to the exchange, it is the shape of the book: you cannot rest a two-sided quote when the market you are matching is one-sided 90% of the time. Size must be sized to the other resting side, which is often a single level of small depth.

Inventory risk concentrates at edges: near 0/1 the ladder is compressed to one-tick increments and the price is all settlement terms; near unresolved strikes the reprice can move the ladder a cent-and-a-quarter in a second (0.044 to 0.996 observed once).

Every edge above degrades, not grows, with age: the same traits that make a market tight make it fragile, so quotes scale out of tight books into longer instruments rather than into size.

Market making on Polymarket is quoting a fragment: a ladder that is one-sided most of the time, top-of-book depth measured in thousands of shares against a median trade of 42,254, and a reference move that re-values the whole book in a second.

The strategies that fit the fragment are the ones that commodify it: size to the resting side, quote the ladder not the mid, re-quote on reprice triggers, and exit final-minute risk through the ladder with the ask often absent.

Replay before re-gaming. The 2026-05-02 reconstruction clips, the 26-second multi-coin collapse, and the 65-second ETH repricing each teach a different discipline; a maker who simulates on the archive is a maker who knows which reprice he is sized for.

• Implied probability
• Liquidity
• Order flow
• Settlement data