Buyer's guide
The best Kalshi order-book data, and how to choose it
Most Kalshi data you can find is a last price sampled once an hour — fine for a chart, useless for a backtest. Here is what actually separates a Kalshi source you can trade on from one that only looks like data.
Buyer's guide
Five criteria that decide whether data is backtestable
1. Resolution: event-driven, not interval-sampled
A snapshot taken on a fixed clock — hourly, every minute, every few hundred milliseconds — misses everything that happens between ticks. Kalshi's short-dated markets settle in 5 to 60 minutes, so an interval sample captures only a handful of frames from a market's entire life. DepthFeed Polymarket capture uses full-book seeds and CLOB price-change deltas; normalized observations are persisted at the configured cadence. Kalshi uses paced full-depth REST polling under the upstream quota. Inspect returned timestamps, gaps and documented series coverage; neither method guarantees that every intermediate state is retained.
2. Depth, not the last price
A last traded price (or a single mid) hides the spread and the size resting at each level. Without the full ladder you cannot measure slippage, and a backtest that assumes you filled at the mid is a backtest that lies. DepthFeed DepthFeed serves the complete bid/ask book, both sides, every level — up to 100 levels per side on Kalshi — so fills size against the liquidity that was genuinely there.
3. Coverage: one schema, every venue and asset
Single-venue datasets force you to stitch together a different format for each market and re-write your loader every time you add one. Partial asset coverage quietly caps which strategies you can even test. DepthFeed DepthFeed serves Polymarket, Kalshi, and Limitless in one stable columnar schema across seven assets — BTC, ETH, SOL, XRP, DOGE, BNB, HYPE — so the same code reads every venue.
4. Delivery: choose files, historical requests or live updates
Daily files suit reproducible batch research; bounded API requests suit interactive historical queries; live updates answer current-state questions. Check schemas, availability, price and licensing for the delivery method you need. A historical file remains historical evidence rather than becoming incorrect merely because the current market changes. DepthFeed DepthFeed provides metered REST history, normalized live WebSocket frames and listed daily Polymarket and Kalshi Parquet files. Their availability and response contracts are documented separately.
5. Realistic fills you can trust
The whole point of backtesting is to know whether a strategy would have filled, and at what price. That answer only exists if you replay the actual order book the strategy would have traded against, timestamped finely enough to line up with the move that drove it. DepthFeed DepthFeed exposes a millisecond observation time and ASOF-aligns the nearest preceding Binance reference when available. Source timestamps remain venue-specific.
Where the usual options fall short
When people look for Kalshi data, they reach for one of four things. Each is useful for something — and none of them is the order book you backtest against.
- The exchange's own API
- Check the exchange documentation for the exact historical objects, retention and depth offered. Current books, trade history and candles answer different questions; none alone proves that a full past ladder was retained.
- Free hourly archives
- Give you a last price sampled once an hour. That is one frame from a market that may have lived five minutes — no spread, no depth, nothing to size a fill against.
- Last-price & trade-tape APIs
- Tell you what printed, not what was resting. You see executed trades but never the liquidity behind them, so slippage and fill probability stay invisible.
- Single-venue file dumps
- Often carry real depth, but for one venue, in one bespoke format, as a static download — no live stream, no second venue, and stale the moment it lands.
Why DepthFeed
DepthFeed combines stored Kalshi ladders with REST history, normalized live WebSocket frames, explicit plan windows, and documented venue-specific cadence.
