Datalake Schema
PolySimulator captures tick-level prediction market data and underlying oracle pricing into high-fidelity Apache Parquet datasets stored in Cloudflare R2 (polysim-datalake).
The archive format (version v3) is byte-compatible with the standard 16-column orderbook format, ensuring existing PyArrow, DuckDB, and Polars pipelines run unmodified across historical datasets.
1. Orderbook Dataset — v3/orderbook/polymarket/
Stores raw order book snapshots and tick events partitioned hourly: polymarket_orderbook_<YYYY-MM-DDTHH>.parquet.
Column Specification
Technical Details
- Compression: Zstandard (ZSTD) compression across all columns.
- Row Groups: 1,048,576 rows per row group.
- Sorting: Rows sorted by
market ascending for fast condition ID predicate pushdown.
2. Underlying Oracle Prices — v3/underlying/polymarket/
Captures reference asset spot prices (Binance spot, Chainlink oracles, and equity indices) used for settling binary Up/Down and scalar markets.
Column Specification
Reader Recipes
The archive is private. These recipes do not publish credentials or
raw objects. You must supply a local download/mount or an endpoint
configuration you already hold. Never commit BACKTEST_RO_*,
INGEST_RW_*, or R2 access keys.
PolySimulator does not ship a public download URL, a sample
parquet, or a credential. If you do not already have an authorized
local copy or an endpoint, these recipes will not run — that is
intentional.
Configure one of:
Reading with PyArrow
Reading with DuckDB