kachoio/polymarket-5-minute-crypto-up-down-markets

Polymarket 5-Minute Crypto Up/Down Markets

Second-by-second top-of-book order-book recordings of Polymarket's recurring 5-minute crypto up/down markets — bet on whether a coin closes a fixed 5-minute window higher or lower — for BTC, ETH, SOL, XRP, DOGE, HYPE and BNB.

• ~89,000 markets (one resolved 5-minute window each)
• 26,858,579 per-second observations (≈300 ticks per market)
• Span: BTC 24 Mar 2026 → 18 May 2026; the other six 5 Apr 2026 → 18 May 2026 (UTC)
• Coverage 99.8%+, no duplicates. A fixed historical window, not a live feed.

There is no freely available history of these markets anywhere else — Polymarket exposes the live book but not the past. I recorded this to backtest a trading bot and I'm releasing it as-is.

Full write-up, with the story and how I used it: https://kacho.io/polymarket-5min-crypto-dataset

Dataset structure

Two tables per coin, joined on condition_id:

• markets — one row per resolved 5-minute market (identifiers, timing, outcome, tick count).
• ticks — one row per second (the order book), linked back to markets by condition_id.

Files are <coin>_markets.parquet and <coin>_ticks.parquet. Two configs are exposed (markets and ticks), each with one split per coin.

markets columns

Column	Type	Meaning
condition_id	string	Polymarket on-chain condition ID. Unique — the join key to ticks.
event_id	string	Polymarket Gamma event ID.
slug	string	Market slug, e.g. btc-updown-5m-1774745100.
market_start	timestamp[UTC]	5-minute window open (boundary-aligned to :00/:05/:10…).
market_end	timestamp[UTC]	Resolution time = market_start + 5 min.
recorded_at	timestamp[UTC]	When the row was written, just after market_end.
token_up	string	ERC-1155 token ID for the "Up" outcome.
token_down	string	ERC-1155 token ID for the "Down" outcome.
volume	float64	Gamma-reported volume (USDC) at discovery time.
liquidity	float64	Gamma-reported liquidity (USDC) at discovery time.
outcome	string	Up / Down / null. Inferred from the final tick's bid, not on-chain resolution.
n_ticks	int64	Number of per-second rows for this market in ticks (≈300).

ticks columns (one 1-second sample of both books)

Column	Type	Meaning
condition_id	string	Join key back to markets.
t	int64	Sample time, unix epoch seconds (UTC).
ts_utc	timestamp[UTC]	Same instant as t, as an ISO timestamp.
bu / au	float64	Best bid / ask, Up token (USDC 0–1).
bd / ad	float64	Best bid / ask, Down token (USDC 0–1).
su / sd	float64	Size at best bid, Up / Down (shares). NaN if that side was empty.
sau / sad	float64	Size at best ask, Up / Down (shares). NaN if that side was empty.
du / dd	float64	Bid-side depth within 5¢ of best bid, Up / Down — Σ(size×price), USDC.

Each outcome's price is its market-implied probability. A point estimate for P(Up) is the mid, (bu + au) / 2. Up and Down are independent books, so bu + bd won't equal 1 — the gap is the spread. The only NaNs in the data are in the resting-size columns (su/sd/sau/sad) when a side of the book was empty.

Usage

import pandas as pd

markets = pd.read_parquet("btc_markets.parquet")   # one row per market
ticks   = pd.read_parquet("btc_ticks.parquet")     # one row per second

ticks["mid_up"] = (ticks["bu"] + ticks["au"]) / 2       # implied P(Up)

# the per-second history of a single market
first = markets.iloc[0]["condition_id"]
one = ticks[ticks["condition_id"] == first].sort_values("t")
print(one[["ts_utc", "bu", "au", "mid_up", "du"]].head())

Or via datasets (config = table, split = coin):

from datasets import load_dataset
repo = "kachoio/polymarket-5-minute-crypto-up-down-markets"
ticks   = load_dataset(repo, "ticks",   split="btc")
markets = load_dataset(repo, "markets", split="btc")

Coverage & known limitations

• Fixed window, ends 18 May 2026 — treat as history, not live.
• BTC has ~12 extra days than the other six coins.
• Coverage 99.84–99.88% per coin. The 19–20 missing windows per coin are shared across all coins (collector outages, not coin-specific): the big one is 18 Apr 10:35→11:55 UTC (1.2h), plus a few 1–2 window blips on 15–16 Apr.
• ~99.97% of markets have the full ~300 ticks; no row has internal 1-second gaps; no duplicates (condition_id is unique). n_ticks tells you a market's exact count.
• outcome is inferred (best-effort, can be null near edges).
• volume/liquidity are point-in-time at discovery, not end-of-window.
• Ask-side depth not recorded — du/dd are bid-side only.
• Sampling is best-effort 1 Hz from a cache; a tick is written only when at least one side had book data (why a few markets fall short of 300, and why resting-size columns are NaN where a side was empty).

Provenance

A custom recorder subscribed to Polymarket's public CLOB WebSocket order-book feed, kept an in-memory book per active market's Up/Down tokens, discovered markets via the Gamma API every 30s, sampled cached top-of-book once per second, and wrote each market's ticks when its window closed. All timestamps UTC.

Licence & disclaimer

CC0 1.0 (public domain) — no attribution required, no restrictions. A link back to kacho.io is appreciated but never required.

Derived from public Polymarket market data (5-minute crypto up/down order books) via Polymarket's public CLOB WebSocket and Gamma API. An independent, transformed recording (per-second top-of-book aggregates). Not affiliated with or endorsed by Polymarket. Provided "as is" for research/educational use, no warranty.

Citation

Kacho. (2026). Polymarket 5-Minute Crypto Up/Down Markets [Data set]. kacho.io.
https://kacho.io/polymarket-5min-crypto-dataset