Polymarket historical data —
every tick, every book update, every trade.
Tick-by-tick price history, the full 25-level L2 order book and the complete trade tape — captured live from the venue by dual-region, sequence-audited collectors and delivered as clean parquet/CSV bundles or over a REST API.
No websocket feed handler to write, no capture plant to babysit, no gaps the moment you blink. The market you didn't think to record last month is already on disk — we never stopped watching.
Running your own plant — capture, storage, redundancy, normalization — starts at $1,000+/month, and every month you spend building it is a month of history you cannot go back for.
- ✓Whole venue, continuously — every market on Polymarket, not the handful you thought to point a recorder at.
- ✓Analysis-ready — deduplicated, sequence-audited, stable schemas and join keys across every market and day.
- ✓Yours in minutes — pick a market·day, and the bundle stages while you finish your coffee.
BTC Up or Down 4h · 2026-08-19 · full day of L1 / L2 / trades + one hour of raw websocket.
Create free account →- ✓$0 — 5 market·days of your choosing, any market, any date on disk
- ✓Parquet with a CSV twin, plus the reference dictionary to join it
- ✓The whole catalog, browsable — see exactly what exists before you buy
Takes about a minute. No card, no sales call. Paid plans start at $89/mo and cancel any time.
Already know the market·day you need? Buy it straight from the catalog — checkout takes an email and a card, no account first.
Three steps, about five minutes.
There is no onboarding call, no data-request form and no quote to wait on. The whole thing is self-serve, and the free tier is the same delivery path a paying customer uses.
Email and password. No card, no sales call, no waiting on a quote — the catalog opens immediately. Or skip it: a single market·day can be bought as a guest, and the receipt sets your password afterwards.
Search the catalog of every market·day on disk — sports, politics, crypto, culture — and claim any 5 of them. The bundle stages in about a minute.
Parquet with a CSV twin, plus a reference dictionary that maps every id to the real question. One read_parquet and you have a dataframe.
import pandas as pd
l1 = pd.read_parquet("l1.parquet") # every quote change, ns timestamps
book = pd.read_parquet("l2.parquet") # 25 levels a side, every update
trades = pd.read_parquet("trades.parquet") # the print tape
l1["mid"] = (l1.best_bid + l1.best_ask) / 2Four layers of history, one contiguous archive.
Everything below comes from the same captured tape and joins on the same keys — outcome token, market, event, series — with nanosecond receive timestamps on a single clock. Full column-by-column detail is in the schema reference, and every change to it is dated in the changelog.
Price history — L1 quotes
every tickBest bid, best ask and mid for every outcome token, recorded on every change — not sampled, not bucketed into candles. Each row carries the collector's nanosecond receive timestamp, so the price history replays in true arrival order.
Order-book history — L2 depth
25 levelsThe full limit-order-book ladder, up to 25 levels per side, on every book update. Resting size, spread regimes, sweeps and refills — the microstructure detail that sampled price feeds throw away and that can't be reconstructed after the fact.
Trade history — the complete tape
since February 2026Every fill with price, size and aggressor side, captured tick-by-tick from the venue's own feed — the print tape as it happened, not a reconstruction. Older history is a separate layer: on-chain trade fills read from Polygon settlement, back to 2023 per market, sold on their own (Premium and up).
Raw feed — websocket atoms
ProfessionalThe untouched websocket messages our collectors received, compressed and hash-audited, plus the deterministic replay engine. For desks and auditors who need to rebuild any book at any nanosecond, in-house, byte-for-byte.
Download it as CSV or parquet. Or don't download it at all.
Every bundle is columnar parquet with a CSV twin — one read_parquet or read_csv away from a dataframe, in pandas, R or Excel. Take it whichever way suits the job.
Pick a market·day, watch it stage, pull the file. Interrupted transfers resume rather than restarting, so a multi-gigabyte pull survives a flaky connection.
List the catalog, request any market·day, poll the build and stream the bundle — key-authenticated end to end. Feed a backtest grid or a training pipeline without clicking anything.
Self-serve setup with your own key. Your bundles appear grouped by date, alongside a rolling window of the live normalized tree scoped to the markets you chose.
curl -H "Authorization: Bearer tf_live_…" \
-X POST https://tickfoundry.com/api/v1/downloads \
-d '{"unit_id": "…", "date": "2026-07-15"}'Full endpoint reference, auth and worked examples in the API docs.
The whole venue, continuously — not just the markets you thought to watch.
A DIY recorder only captures what it was pointed at. TickFoundry captures every market on Polymarket, all day, every day — sports, politics, crypto, culture — so when your research changes direction, the history is already there. Two redundant collectors in separate regions record the same feed; sequence numbers are audited mechanically and gaps are cross-filled between them.
Tick and book capture since 2026-05-11 · on-chain fills back to 2023-08, per market — thinner before 2026. We start clean and accumulate — one contiguous archive, no synthetic backfill. Where a day has a known gap, the bundle says so in writing. Our research notes are computed from the same bundles we sell, so you can rerun every figure yourself.
Start at zero. Scale to the whole venue.
No per-market pricing and no credits arithmetic — one flat plan reaches any market·day when you need it. You do not have to know in advance which markets you will want: every plan roams the whole venue, you download only the market·days you actually use, and keeping the rest on disk stays our problem, not yours. So does keeping the capture alive: the monitoring, and the pager when a feed drops at three in the morning. Founder pricing — 25% off for the life of the subscription — is live on Explorer and Premium while 41 seats remain.
Create a free account and pull any 5 market·days from the live venue, plus the full sample bundle. Validate the schema against your pipeline before you pay a cent.
Create free account →Personal-use access to every market on the venue — pull any series or event, any day, as much as you need within a fair-use cap. Built for the solo quant, PhD student and indie researcher.
Start Explorer →Everything in Explorer, uncapped on downloads — the full 25-level L2 ladder, a fair-use cap on the API, the Binance + Chainlink spot (RTDS) feeds for spot-versus-prediction-market correlation work, and SFTP delivery.
Start Premium →Everything in Premium plus the raw websocket atoms and replay source, hourly (or faster) updates, and delivery customized to your infrastructure — a dedicated S3 bucket with your own routes, SFTP, VPC peering or direct push — with an SLA, a named contact and a bespoke license.
Talk to sales →The history exists because we captured it.
Now it's a download away.
5 market·days and the full sample bundle, free, with no card — enough to run your idea against the real tape before you decide anything.
Or skip the account entirely: buy a single market·day from the catalog with just an email and a card.