Datasets¶

The Polymarket Users dataset is split into several related tables. Each serves a different access pattern; they share the same user_address, prediction_id, market_id, and event_id identifiers and can be joined freely.

Map¶

metadata         features          PnL                    trade data
─────────        ─────────         ──────                 ──────────
markets          user_features     user_pnl_summary       trades
events                             pnl_daily              ohlcv_1d
                                   pnl_category_daily     ohlcv_1h
                                                          ohlcv_5m

Pick the right PnL table¶

Three tables expose PnL at different granularities. Pick the smallest one that satisfies your access pattern:

If you want…	Use
A single number per user (terminal PnL across variants, ± portfolio/cash split for base)	`user_pnl_summary`
Per-user time-series PnL	`pnl_daily` — needs forward-fill
Per-(user, category) time-series PnL	`pnl_category_daily`

Identifiers¶

Column	Type	Source	Joins to
`user_address`	`str`	Polygon wallet address (end user, after reconciliation)	All user-level tables
`prediction_id`	`str`	CTF Exchange conditional token id	`trades.prediction_id`, `ohlcv_*.prediction_id`, `predictions.prediction_id`
`market_id`	`int64` in `markets` / `predictions`, `str` in `trades` / `ohlcv_*`	Polymarket internal market id	`markets.market_id`, `predictions.market_id`, `trades.market_id`, `ohlcv_*.market_id` — cast trades' string id to `int64` when joining back to `markets`
`event_id`	`str`	Polymarket internal event id	`events.event_id`, `markets.event_id`, `trades.event_id`
`category`	`str`	Canonical category, tag-based (Sports, Crypto, Finance, Politics, Tech, Culture, Weather; or `Untagged`)	`markets`, `events`, `trades`, `pnl_category_daily*`

See the Methodology page for how categories are assigned and how end-user addresses are reconciled from the CTF Exchange event stream.