Polymarket Historical Data — Free Full L2 Order Book Archive

Free historical Polymarket order book data: a daily Parquet archive of the full L2 book recorded from the live CLOB WebSocket feed since 2026-07-25 — every book snapshot, price-change delta and trade the recorder received, plus a derived top-of-book stream, across the markets it records (crypto up/down, sports and politics). CC BY 4.0, one free API key, samples with no key at all.

Coverage: 3 streams · one Parquet partition per UTC day · CC BY 4.0 · sizes in the live manifest, since 2026-07-25 (the live manifest lists what exists right now). Free — API key — delivered via Archive.

Data summary

Getting started

# Every available date with per-stream sizes and row counts — no key required:
curl -O https://data.vike.io/archive/manifest.json

# Download one day of one stream (needs a free key; archive downloads do NOT
# count against your hourly limit). -C - resumes an interrupted download.
curl -H "X-API-Key: vk_..." -C - -O \
  "https://data.vike.io/archive/venue=polymarket/date=2026-10-11/trades.parquet"

# Free evaluation sample — one full up/down market, no key required:
curl -O https://data.vike.io/archive/samples/book_events.parquet

import io, requests
import polars as pl

H = {"X-API-Key": "vk_..."}
day = "2026-10-11"
url = f"https://data.vike.io/archive/venue=polymarket/date={day}/l1_quotes.parquet"
# Download the whole file, THEN parse it locally. A remote column reader
# re-fetches the footer and every column chunk as separate Range requests
# on each query — slower for you and heavier for the archive.
df = pl.read_parquet(io.BytesIO(requests.get(url, headers=H, timeout=600).content))
print(df.group_by("condition_id").len().sort("len", descending=True).head(10))

// The archive itself is bulk Parquet — use the Polymarket MCP tools to
// discover markets and resolve the condition_ids you then join the
// archive on:
{"jsonrpc": "2.0", "id": 1, "method": "tools/call",
 "params": {"name": "polymarket_markets",
            "arguments": {"limit": 20}}}

Schema

book_events

Full L2: snapshots + price_change deltas + trades + tick-size and status markers. `bids`/`asks` hold JSON level arrays.

trades

Taker tape only — one row per print.

l1_quotes

Derived top-of-book bid/ask, one row per L1 change.

Real rows from the free sample file

Real rows from the free evaluation sample — one full crypto up/down 5-minute market (both outcome tokens, open through resolution), regenerated daily; no key required. These rows were read from the 2026-09-01 sample. Source: https://data.vike.io/archive/samples/

trades

l1_quotes

book_events

Example applications

• Backtest an up/down market-making strategy

  If you are testing whether quoting inside the spread on 5-minute crypto up/down markets earns the maker share it appears to, replay book_events sequence-by-sequence and fill your quotes against the actual taker tape in trades — the archive carries both sides of that experiment at full depth.

• Measure resolution-window microstructure

  If your hypothesis is that spreads widen and depth vanishes in the final seconds before resolution, l1_quotes gives you top-of-book at every L1 change through a market's final minutes — no sampling, no conflation.

• Study cross-market information flow

  If you suspect politics and sports books move before correlated crypto up/down books, the whole universe is in one archive with a common clock (`ts` venue time plus `local_ts` capture time), so lead–lag is a join, not a data-engineering project.

Access

FAQ

Related datasets