About
The data nobody archives
DepthFeed is an independent research project. We capture and keep the one thing prediction-market traders can't get anywhere else: the full historical order book.
Why we built it
Backtesting a prediction-market strategy is only honest if you replay it against the book it would have actually traded against. But order-book depth is expensive to record in real time and impossible to backfill after the fact — so almost nobody keeps it. Free archives give you an hourly last price, which tells you nothing about spread, size, or slippage.
We run collectors against the Polymarket CLOB websocket and Kalshi's public orderbook, recording the full bid/ask book on both venues into a compact, durable archive. The result is a dataset you can replay tick by tick and size a real strategy on.
How we work
The stack is deliberately small and durable: a Rust collector and API, ClickHouse for the hot window, and a compact columnar cold archive. We document what we capture and how often, and we are precise about the difference between event-driven capture (Polymarket) and continuous polling (Kalshi) rather than rounding it up in marketing copy.
What buyers can inspect before subscribing
Coverage is documented by venue, asset, market series, capture method, and available history. The free Explorer path lets a buyer inspect the response shape before choosing a paid history window. REST serves stored observations, while the WebSocket surface publishes normalized current-book updates; the contracts are documented separately so an integration does not have to guess which timestamps or fields it received.
We also publish reproducible research built from the archive, including liquidity and execution examples. Those examples are meant to show what full-depth data changes in a backtest—not to promise a trading outcome. Start with the historical-data overview, review the API documentation, or compare the data surfaces a buyer should verify.
Limits we state explicitly
A recorded book describes the displayed ladder at its observation time. It does not prove queue position, hidden liquidity, or activity between stored observations. Coverage starts at different times by venue and series, and a missing reference join remains missing rather than being filled with an invented value. These boundaries are part of the product contract, because a buyer should be able to decide whether the archive fits a study before paying.
Not affiliated
DepthFeed is independent and is not affiliated with, endorsed by, or sponsored by Polymarket, Kalshi, or any exchange. All product names are the property of their respective owners.
Questions or data requests? Get in touch.
The archive, in numbers
Every figure is what the collectors have stored, not a projection. Coverage is documented by venue, asset and capture method, and the archive deepens every day.
640M
order books captured across Polymarket, Kalshi and Limitless
90+
days of history, all venues
7
assets: BTC, ETH, SOL, XRP, DOGE, BNB, HYPE
REST + WS
delivery: JSON snapshots and a live stream
How we work
The DepthFeed difference
Full depth, both sides
Every stored observation keeps the resting levels and sizes on both sides of the book, not a last price or a top-of-book quote.
Capture stated per venue
Event-driven on the Polymarket CLOB websocket, continuous polling on Kalshi. We say which is which instead of rounding it up.
Inspect before you subscribe
The free Explorer path shows the response shape, and coverage is documented by venue, series, method and available history.
Limits written down
A recorded book is the displayed ladder at its observation time. Missing joins stay missing; nothing is filled with an invented value.
What we publishProducts
Order-book depth, price bars and the tools that replay them, on one clock.
Common questions
Simple answers
