marketlens/polymarket-resolved-markets

Polymarket Resolved Markets

Every Polymarket market that resolved between 2026-02-28 and 2026-08-10, with the outcome it settled to. 634,554 markets across 621 series and 9 categories, in one parquet file.

Polymarket's public API gives you markets that are open now. Once a market resolves it drops out of the feeds you would normally page through, and the resolution itself is emitted as a WebSocket event rather than a queryable endpoint, so a settled-outcome history is something you have to have been recording. This is that recording, published as a flat table.

Each row also carries the CLOB token_ids, so a market here can be joined straight back to py-clob-client or any other Polymarket tooling.

Coverage

Category	Markets
Crypto	525,273
Sports	74,762
Weather	18,105
Equities	9,644
Commodities	1,561
Indices	928
Economy	255
Other	250
Politics	11
(uncategorised)	3,765

Resolution dates run 2026-02-28 to 2026-08-10.

Columns

Column	Type	Notes
market_id	string	Polymarket condition id (0x...)
question	string	Market question as listed
series_slug, series_title	string	Parent series, null for one-off markets
series_recurrence	string	5m, 15m, hourly, daily, weekly, ...
category, subtype, market_type	string	subtype is e.g. up_or_down, hit_price, moneyline
outcome_names	list[string]	In platform order
token_ids	list[string]	CLOB token id per outcome, same order
winning_outcome	string	The outcome that settled true
winning_outcome_index	int32	Index into outcome_names
open_time_ms, close_time_ms	int64	Epoch milliseconds, UTC
resolved_at_ms, resolved_date	int64, date	When resolution was observed
volume_usd, liquidity_usd	float64	As reported by the platform
strike, strike_upper, strike_direction	float64, string	Populated for strike-based markets only
tick_size	float64	Minimum price increment

Loading

from datasets import load_dataset

ds = load_dataset("marketlens/polymarket-resolved-markets", split="train")

Or straight into pandas:

import pandas as pd

df = pd.read_parquet(
    "hf://datasets/marketlens/polymarket-resolved-markets/polymarket_resolved_markets.parquet"
)
base_rate = df.groupby("series_slug")["winning_outcome"].value_counts(normalize=True)

What is not in here

No prices, no order books, no trades. This is the index of what resolved and how it settled, which is the part that is awkward to reconstruct after the fact and easy to give away.

If you need the book itself, we keep tick-level L2 history for the same markets (full snapshot roughly every 60 seconds plus every price_change delta at millisecond resolution) along with trades and candles on the same time base: github.com/marketlenstrade/marketlens-python and marketlens.trade. The free tier is enough to pull a market and check the data.

Provenance

Collected by Marketlens with a dedicated WebSocket connection per market against Polymarket's public CLOB and Gamma APIs. Resolutions come from the platform's market_resolved event, not inferred from final price. Derived from publicly available Polymarket data.

Citation

@misc{marketlens_polymarket_resolved_markets_2026,
  title  = {Polymarket Resolved Markets},
  author = {Marketlens},
  year   = {2026},
  url    = {https://huggingface.co/datasets/marketlens/polymarket-resolved-markets}
}