Top-of-book quotes + game state across Polymarket and Kalshi sports markets for the 33-day window from Mar 28 → Apr 30 2026 (the depth-recorder was offline, but top-of-book + game state continued capturing). Kalshi side: full archive-grade schema (25 columns, parity with the Jan 2026 archive Kalshi side). Polymarket side: top-of-book only (11 columns) — no L2 depth. Known gap: the men's & women's college-basketball moneyline (KXNCAAMBGAME / KXNCAAWBGAME) ends ~Apr 3–4, so the early-April NCAA tournament championship team-to-win markets are not captured (their spread & total are). All other sports' moneyline/spread/total run the full window. Useful for backtesting score-reaction, trade-timing, and signal-generation strategies that don’t need depth metrics. If you need full L2 depth, see the Polymarket & Kalshi Orderbook Archive (Jan 2026) or the monthly drops starting May 2026.
Neither Polymarket nor Kalshi publishes historical orderbook data — their public APIs give you trades and last price, never the full book over time, and never joined to the game. That makes this data physically non-reconstructable retroactively: the score-sync has to be captured live, tick by tick, as each game happens. We have been recording it. A provider starting today cannot backfill what they did not capture. Every snapshot here carries the live score, period and clock at tick time — the one thing every other "tick data" seller is missing.
78M rows total · Kalshi: 34M (full schema, 60K tickers) · Poly: 44M top-of-book (19 sports, 27K tokens) · Mar 28 – Apr 30 2026
Join your win-probability output on event and clock, and measure how many seconds the book trailed a real probability shift after a score.
Full depth lets you model slippage and partial fills against what the market actually showed — not an assumption that you hit last price.
With Polymarket and Kalshi on one timeline, measure which book moves first after a scoring event — the basis for cross-venue arbitrage research.
Citable depth and trade data for prediction-market efficiency and price-discovery studies, with provenance metadata.
We publish real score-synced sample rows — not a marketing mockup. Load them in DuckDB, check the schema, confirm the score-join is there, then decide.