Data SchemasPredict.fun
Markets catalog
A free Parquet file listing every Predict.fun market PMData has recorded, with the UTC dates it has L2 data for.
The Predict.fun markets catalog is one Parquet file that lists the Predict.fun markets PMData has recorded, and the UTC dates it has L2 data for each one. It needs no API key, uses no download quota and is updated daily.
Use it to find a market's market_id and dates, then download the data with the API reference.
GET
https://public-data.pmdata.dev/v1/predict_fun/markets.parquetSchema
Columns are listed in file order.
| # | Field | Type | Description |
|---|---|---|---|
| 1 | exchange | string | Exchange: predict_fun. |
| 2 | category_slug | string | null | Predict.fun category slug; the market's page is predict.fun/market/{category_slug}. One category can contain many markets. |
| 3 | market_id | int64 | Predict.fun numeric market id. Use it as {market_id} in the API. |
| 4 | record_start | timestamp (UTC) | First UTC date PMData has L2 data for the market. |
| 5 | record_end | timestamp (UTC) | Last UTC date PMData has L2 data for the market. Updated daily while the market is still being recorded. |
| 6 | market_created | timestamp (UTC) | null | Time Predict.fun created the market. |
| 7 | question | string | null | Market question. |
| 8 | title | string | null | Market title; in a category with many markets, the name of this market within it. |
| 9 | yes_outcome | string | null | Outcome whose order book the l2 files contain, e.g. Up. |
| 10 | no_outcome | string | null | The market's other outcome, e.g. Down. |
| 11 | series | string | null | Recurring Up/Down series, e.g. btc-updown-1h. Empty for other markets. |
Market information (columns 2 and 6–10) can be empty for a few markets.
Dates
record_startandrecord_endare UTC dates, the same dates as in the download URL/{YYYY}/{MM}/{DD}/.- If a market had no events at all on a day between
record_startandrecord_end, there is no file for that day. The API returns404 Not Found, which uses no download quota.
Example
Find a market by its question, then see which dates PMData recorded:
import pandas as pd
url = "https://public-data.pmdata.dev/v1/predict_fun/markets.parquet"
# Python's default User-Agent is blocked, so set one
catalog = pd.read_parquet(url, storage_options={"User-Agent": "pmdata-catalog"})
# By question text
matches = catalog[catalog["question"].str.contains("Bitcoin", case=False, na=False)]
# Or every market in one category
# matches = catalog[catalog["category_slug"] == "<category_slug>"]
print(matches[["market_id", "question", "title", "record_start", "record_end"]])Then download a date in that range:
https://api.pmdata.dev/v1/predict_fun/l2/{YYYY}/{MM}/{DD}/{market_id}.parquetThe catalog covers L2. lifecycle is one file per UTC day for all markets; see Predict.fun historical data details.