Datasets
Prediction market data API (Polymarket · Kalshi)
Prediction markets produce something rare: a continuously updated probability with money behind it. They are also messy to use as data. Two venues describe similar questions with different contract terms, resolution criteria live in prose, markets are created and closed constantly, and a probability means nothing without the exact wording of what it resolves against.
Hindcast is building a normalized layer over Polymarket and Kalshi: contract metadata with resolution criteria kept verbatim, price and volume history stored as observations with their timestamps, and outcomes recorded when markets settle. Questions are linked to entities and topics where a defensible link exists, so a market about a company or a policy decision can be joined to the filings and central bank text about the same subject.
Point-in-time here means the obvious thing: the probability you query for a past date is the probability that was quoted then, not a number reconstructed with knowledge of how it resolved.
hindcast — session
previewhindcast> what did this market imply the week before it resolved?
→ tool: prediction_market { venue:"...", market_id:"...", as_of:"..." }
✓ quoted probability and volume at that timestamp
source: venue market data with resolution terms attached
What's covered
- Polymarket and Kalshi contracts normalized to one schema
- Resolution criteria kept verbatim alongside the structured fields
- Price and volume history as timestamped observations
- Settlement outcomes recorded when markets resolve
- Markets linked to entities and topics where the link is defensible
- No back-filled probabilities — only what was quoted at the time
Join the waitlist — this page's dataset is already selected.
Questions
Can prediction market prices be read as probabilities?
Approximately, and with care. Fees, capital costs and thin books all distort the mapping, and the resolution wording often decides more than the headline question. The data layer keeps the terms attached so the caveats stay visible rather than getting lost in a number.
Are similar markets across venues merged?
No. They are linked and left distinct, because two venues rarely resolve on identical criteria, and merging them would manufacture a spread that is really a definition mismatch.
Related
Join the waitlist. Vote the roadmap. First in, first served.