polyorderbooks/polymarket-crypto-5min-orderbooks

Polymarket Crypto 5-Minute L2 Order Books

Full level-2 order book snapshots at 1-second resolution for 15 complete Polymarket crypto up/down markets — every resting bid and ask, not just top of book.

Seven assets: BTC, ETH, SOL, XRP, BNB, DOGE, HYPE. Each market is covered end to end across its entire five-minute life, so it can be replayed rather than sampled.

from datasets import load_dataset

ds = load_dataset("polyorderbooks/polymarket-crypto-5min-orderbooks", split="train")
row = ds[0]

print(row["market_slug"], row["timestamp"], row["outcome"])
print("best", row["best_bid"], "/", row["best_ask"])
print("depth", len(row["bids"]), "bid levels /", len(row["asks"]), "ask levels")

Coverage

Markets	15 complete five-minute markets
Assets	BTC, ETH, SOL, XRP, BNB, DOGE, HYPE
Rows	9,000 — one per (timestamp, outcome)
Price levels	846,219
Capture resolution	1 second
Window	2026-08-20 10:45:00Z – 10:59:59Z
Outcomes	Yes and No, both sides of each book

Why full depth

Midpoint and candle data tell you where a market was quoted. They do not tell you whether your size could trade there.

def cost_to_buy(row, size):
    """Average price actually achievable, walking the ask ladder."""
    filled = cost = 0.0
    for level in row["asks"]:
        take = min(level["size"], size - filled)
        cost += take * level["price"]
        filled += take
        if filled >= size:
            return cost / filled
    return None          # resting book cannot absorb this size

The gap between the quoted touch and the achievable average widens with order size and thin depth. A strategy evaluated on midpoints assumes the first and realises the second.

Schema

Field	Type	Description
timestamp	string	UTC, ISO-8601, one per second
market_slug	string	Polymarket market identifier
question	string	Market question
outcome	string	Yes or No
token_id	string	Polymarket CLOB token ID
bids	list of {price, size}	Full bid ladder, highest price first
asks	list of {price, size}	Full ask ladder, lowest price first
best_bid / best_ask	float	Top of each ladder
bid_levels / ask_levels	int	Price levels on each side
winner	string	Resolved outcome of the market

Prices are probabilities in [0, 1]. Sizes are fractional rather than integer share counts. Ladder depth varies between snapshots, so avoid assuming a fixed number of levels.

Data integrity

Every snapshot was checked before publication:

markets           : 15
snapshots         : 9,000
levels inspected  : 846,219
arbitrage bounds  : 4,500 outcome pairs checked, 0 violations
timestamp issues  : 0

Prices within [0, 1], ladders ordered best-first, no duplicate price levels, no crossed books, and the cross-outcome bounds respected throughout:

yes_ask + no_ask >= 1     buying both costs at least what the pair pays out
yes_bid + no_bid <= 1     selling both cannot raise more than the obligation

The checks are published as a runnable script in polymarket-orderbook-toolkit, alongside tools for replaying the book at any timestamp and measuring fill cost against the resting ladder.

Snapshots are recorded as observed, with no synthetic interpolation. A ladder repeated across consecutive seconds means the book did not change during that second.

Wider coverage

This sample covers a fifteen-minute window. Continuous historical coverage across Polymarket crypto markets is available through the API:

pip install polyorderbooks

• Documentation
• Working with L2 order books
• PolyOrderbooks

License

Creative Commons Attribution 4.0 (CC BY 4.0).