Prediction-Market Microstructure Dataset — Derived Event Analytics

Derived Kalshi microstructure tables for quote freezes, recovery curves, spread compression, leader-lag clusters, and price-discovery research.

Full schema & known gaps →· Sold as-is — but if a file is broken or won't open, email [email protected] and we'll send a fixed file or refund you. No hassle.

Inspect the free sample rows and schemas before buying.

Artifact vintage: the product page and shipped documentation state the exact coverage, row counts and known limitations. A purchase delivers the advertised frozen artifact; no future refresh is promised unless the checkout terms explicitly say so.

Season scope not verified Observed partial coverage

A competition season has not been verified for this archive. This is not a complete-season archive.

Book samples:
start unknown → end unknown

UTC timestamps retain their recorded precision; dates alone identify UTC days. Vintage, observation and sports event times are separate. Capture time does not establish exchange quote freshness.

Sports event starts: start unknown → end unknown. Event dates are not inferred from these quote capture windows.

Coverage qualifications
  • No season scope declared in the archive manifest
  • Complete-season fixture denominator has not been verified

Why the capture contract matters

Venue APIs expose useful live books, trades and price history, but they do not serve the same historical snapshot and game-context archive packaged here. Data that was not recorded during the live window cannot be recreated from a current-book endpoint later. The exact fields, cadence, venue coverage and limitations vary by product, so this page and the shipped documentation state the contract explicitly instead of calling every file complete or tick-level.

What's in the dataset

~587K derived event rows across 6 event libraries (largest 178K quote-freezes) + summary tables + 20+ charts, computed from 20M+ raw Kalshi ticks

Research uses for this dataset

Analyze quote freezes

Work from derived freeze events rather than rebuilding the detector before testing a research question.

Measure recovery curves

Compare how quickly prices and spreads normalized after the captured event classes.

Study price discovery

Use leader-lag and spread-compression tables with the published methodology.

Reproduce reported results

Load the Parquet tables and rerun the documented aggregations with your own filters.

See the real rows before you buy

We publish real sample rows — not a marketing mockup. Load them in DuckDB, inspect the schema and coverage contract, then decide.

← All data archives · Lead-lag research built on these events · Backtesting Polymarket strategies — tools & datasets · Data license