TWAP-Settled Crypto Order Books

Polymarket crypto Up/Down order books, full depth at one row per second, across the 5-minute and 15-minute cycles on BTC, ETH, SOL, XRP, BNB and DOGE. Anything dated before 2026-08-13 is simulated: a book we captured live, with every quote re-priced under the 60-second TWAP that Polymarket settles on today. From 2026-08-13 onward the 5-minute cycle settled that way on Polymarket itself, so those books are straight captures — one-second snapshots of the real thing, untouched. Every cycle carries both winner labels and the exact shift applied, so the two eras stay directly comparable.

Build your download

Choose the market cycle, then how much history you want. Weeks can be combined: pick as many as you like, minimum one. Every period is marked Simulated, Part real or Real, and the running total below tells you which you have in the basket before you check out.

What you are buying

Two kinds of book ship under one product, split on a single date per cycle. Nothing is blended, and no file mixes the two without saying so.

• Real, from 2026-08-13 onward. Polymarket moved these markets onto the 60-second TWAP on 2026-08-07 for 15-minute cycles and 2026-08-13 for 5-minute cycles, which ran a 30-second window in between. From its own date, each cycle is a straight capture: the live Polymarket book, one row per second, full depth, resolving on the same average the exchange paid out on. Nothing is modelled or adjusted.
• Simulated, before that date. Those markets settled on a closing snapshot, so a TWAP-settled book for them did not exist and we reconstruct it. The starting point is still a book we captured live — every order, every size, every second of it real. What we change is price: each quote is moved by the difference between the snapshot value and the 60-second average. price_shift ships in every row, so subtracting it returns the untouched captured book exactly.
• You always know which you are holding. Each period is tagged in the picker above, the running total names the mix in your basket, and every download carries a manifest recording the simulated/real split by date, the cycle it covers and the window used. A period straddling the boundary is labelled part real rather than rounded to the flattering half.
• Both winner labels, on every cycle. Each cycle resolves twice — once under the old closing-snapshot rule, once under the 60s TWAP — so the cycles the rule change actually flips can be isolated instead of guessed at. Across the periods on offer here that is 7.9%–9.6% of cycles.
• What the simulation does not claim. Spreads and depth are held at their observed values: they do not tighten under TWAP, because nobody was trading under TWAP at the time. No behavioural response is modelled. And Chainlink does not publish the feed’s sampling boundaries, weighting or rounding, so the window edge is our assumption — the last 60 seconds strictly before close, recorded as such in every manifest. Neither caveat touches the captured books; both apply only to the simulated span.
• Coverage. 5-minute and 15-minute Up/Down cycles on BTC, ETH, SOL, XRP, BNB and DOGE. The archive opens per coin, so the earliest periods genuinely ship four coins rather than six — each period lists what it actually contains.