Historical Binance Prediction Market Data: What Exists and Since When

Why the archive has two start dates

Predict.fun matches orders in an off-chain central limit order book and settles executions on BNB Smart Chain. That split determines what can be recovered later. Market routes and settlement records preserve identity and outcomes; they do not preserve every order that rested, moved or was cancelled.

Calling both datasets 'history' without separating them would overstate the available evidence. DepthFeed therefore exposes an earlier catalogue floor and a later capture floor.

Coverage by dataset

What can be backtested before the L2 start

The earlier catalogue supports studies of market availability, duration, resolution, final volume, final liquidity and outcome frequency. It can also provide the stable identifiers needed to join later reconstructed fills as that dataset becomes available.

It cannot support a realistic spread, queue or slippage backtest because those questions require the resting book at the moment a strategy would have traded.

What the forward archive adds

From July 17, each observed book contains full bid and ask ladders rather than a single probability or midpoint. Exchange and receive timestamps let a pipeline order the snapshots and observe gaps instead of assuming a perfectly uniform stream.

The same capture window contains separate chance, displayed-price and latest-trade series. That separation makes it possible to compare interface signals with executable liquidity without treating them as interchangeable.

What cannot be reconstructed

On-chain settlement can reveal executions, but it cannot reveal an order that rested and was cancelled without filling. It also cannot recreate the exact priority or complete queue between two observed events. Any pre-capture full-L2 series would therefore be synthetic rather than historical.