Polymarket. Pull any market, any days. · the catalog, landed on one event
the unitA market: a whole series (MLB: every game) or a standalone event (Presidential Election Winner 2028). Never a single question.
a market·dayOne market on one UTC day: L1 quotes, L2 depth and trades as parquet.
freeWith a free account: one sample day, 2026-07-15, any market, up to 1 GB a pull. The cyan dot marks it on the calendar. Every other day is paid.
paidÀ la carte from $2 a market·day (top of book and trades), $4 with the 25-level book, a range in one checkout, or a Polymarket plan that roams the venue. One paid pull covers up to 31 days / 10 GB and is one download.
event · tech · polymarket · ⚡ instant full depth
Which Millennium Prize Problem will AI solve next?
⚡ instanttechevent
Tick-by-tick historical market data for the Polymarket event Which Millennium Prize Problem will AI solve next?: a standalone event, sold as its own market. Captured live from the venue's own websocket feed. Pick a day or a range of days on the calendar below and pull it as parquet: one free sample day (2026-07-15) with a free account, any range à la carte from $2 a market·day in one checkout, or every day on a plan.
market·days
71
coverage
2026-09-09 → 2026-11-18
71 day span
outcomes
6
on disk
685 MB
pullable (l1 + l2 + trades) · +1.3 GB raw, Custom
rows on disk
52.6M
Days on disk · click a day; a second click makes a range
2026-09-09OctNov2026-11-18
click a day; a second click makes a range
September 2026
MTWTFSS
12345678
October 2026
MTWTFSS
November 2026
MTWTFSS
192021222324252627282930
full depth, instantfull depthtop of bookno datafree sample day (2026-07-15): not on this one
coverage as a list
2026-09-09 → 2026-11-18 · 71 days with files
no data on the sample day 2026-07-15
What each file covers · measured
L1 quotes28 MB · every plan
L2 depth655 MB · 10 levels on Analyst · 25 on Quant
Trade tape2 MB · every plan
Raw atoms1.3 GB · Custom
Pull Which Millennium Prize Problem will AI solve next?· ~10 MB/day pullable
Or Starter, $29 a month: top of book and trades for every market, and the 25-level book for one market you choose: Which Millennium Prize Problem will AI solve next?. Analyst ($49) has 10 levels on every market; Quant ($99) has 25.
A monthly plan includes the 7 days before you join and every day after, so these days are in it. First payment: $29.
A free account pulls 2026-07-15, any market, up to 1.0 GB, at no cost. Every other day is $2 a market·day (top of book and trades) or $4 with the 25-level book, a range in one checkout, or on a plan bought for Polymarket: see plans.
What is on disk for this event
Measured from the partition catalog, not estimated. A day appears here only if the file exists.
L1 quotes
71 days3.3M rows29.3 MBevery plan
Best bid, best ask and mid for every outcome token, written on every change — not sampled, not bucketed into candles.
2026-09-09 → 2026-11-18
L2 order-book depth
71 days32.1M rows687 MB10 levels on Analyst · 25 on Quant
The limit-order-book ladder on every book update — up to 25 levels per side. Resting size, spread regimes, sweeps and refills.
2026-09-09 → 2026-11-18
Trade tape
67 days4k rows1.8 MBevery plan
Every fill with price, size and aggressor side, captured from the venue's own socket feed as it happened.
2026-09-09 → 2026-11-18
Raw websocket atoms
71 days17.2M rows1.4 GBCustom
The untouched websocket messages our collectors received, compressed and hash-audited, for byte-for-byte replay.
2026-09-09 → 2026-11-18
What a row looks like
Top-of-book (l1) rows from the free sample day, verbatim. Every event in the archive ships this schema; the sample bundle has the full day with L2 and trades.
Every row here came off Polymarket's own websocket feed as it happened, recorded by dual-region sequence-audited collectors and written to a normalized, Hive-partitioned archive. Quotes are written on every change rather than sampled on a timer, so the price history replays in true arrival order with the collector's nanosecond receive timestamp on each row. Nothing here is reconstructed from a public REST endpoint after the fact, which is the difference that matters for microstructure work: resting size, spread regimes and sweeps cannot be recovered once they are gone.
Bundles arrive as parquet with a CSV twin and a reference dictionary mapping every id back to the real question, so a single read_parquet gives you a dataframe. À la carte, any range of days is one checkout at $2 a market·day for quotes and trades or $4 with full L2 depth, or a plan roams the whole venue. The schema is documented in full.
Related Polymarket markets
More tech markets in the archive. Every one of these has data on disk.