trades

All reconciled end-user trades. End-user maker and taker addresses are recovered from the CTF Exchange OrderFilled event stream — raw event addresses are typically smart-contract operators, and reconciliation attributes each fill to the actual wallet that initiated it.

Layout

Hive-partitioned by date — one file per day, coalesced from the underlying hourly cache:

trades/year=YYYY/month=MM/day=DD/data.parquet

Load

import polars as pl

# Scan the whole sample lazily
trades = pl.scan_parquet("trades/**/*.parquet", hive_partitioning=False)

# Or pull one specific day
day = pl.read_parquet("trades/year=2024/month=11/day=05/data.parquet")

from datasets import load_dataset
trades = load_dataset("vgregoire/polymarket-users", "trades")

Schema

Column	Type	Description
trade_id	str	Transaction hash on Polygon
timestamp	datetime[ns, UTC]	Block timestamp
market_id	str	Market identifier (matches markets.market_id, cast as string)
event_id	str	Parent event identifier (matches events.event_id)
prediction_id	str	The conditional token traded (matches predictions.prediction_id)
outcome	str	Outcome label (e.g., "Yes", "Trump", team name, …)
winner	bool	Whether this outcome resolved as the winner (null until resolution)
category	str	Canonical category, inherited from the parent event
category_original	str	Original Polymarket platform tag before mapping
price	float64	Trade price in USDC per conditional token share
quantity	float64	Trade quantity in conditional token shares
maker_address	str	Reconciled end-user maker wallet
taker_address	str	Reconciled end-user taker wallet
taker_bought	bool	True if taker bought conditional tokens, False if sold

Daily volumes

The full sample contains tens of millions of trades. Scanning lazily and projecting only the columns you need avoids materializing the whole dataset:

import polars as pl

# Daily volume per category, 2024-2026
volume = (
    pl.scan_parquet("trades/**/*.parquet", hive_partitioning=False)
    .with_columns(pl.col("timestamp").dt.date().alias("date"))
    .group_by(["date", "category"])
    .agg((pl.col("price") * pl.col("quantity")).sum().alias("volume_usd"))
    .collect()
)