Data and methodology
What a Riddle record can and cannot support.
How to read the observations, contracts and comparisons in Riddle, and the criteria an evaluation uses to decide whether the data fits a question. This describes how evidence is treated today. It does not claim that every historical field has been recovered or repaired.
Observations
Six things to know before using the data.
What an observation is
An observation is a value Riddle recorded for one contract, on one venue, at a recorded time. That time is when Riddle recorded or materialised the value. It is not a venue timestamp, and it is not a statement that the value was tradable at that moment.
Original and recovered evidence
Observations Riddle recorded at the time are original evidence. Data retrieved later from a venue or provider to fill a gap is recovered evidence. They answer different questions and should not be read as one continuous series, and an evaluation says which is which.
Timing
Prices are indicative and age from the moment they are recorded. Read every value together with its time, and check its age against your own threshold before you use it.
Missing fields
Quote fields such as yes and no bids, asks and sizes are kept distinct. When a venue does not provide a field, or Riddle did not record it, the field stays missing. It is not filled in, inferred or treated as zero.
Match limitations
Contracts with similar headlines are candidate comparisons. Two contracts are treated as equivalent only when their wording, resolution dates, settlement sources and edge cases support it, and Riddle does not certify equivalence on your behalf.
Ranking scores and calibrated forecasts
Scores that order or surface markets, including any signal weights you set, are ranking tools. They are not probabilities and not calibrated forecasts. Morpheus is an AI research assistant, and its conclusions are interpretations to check, not certified results.
Field-level definitions for the API, including which timestamps can be null, are in Data semantics.
Coverage
What the data covers today.
- Kalshi, Polymarket and Limitless, named for identification only.
- History is the partial set of observations Riddle holds, not a complete venue record.
- There is no verified complete cross-venue history, complete order-book archive or standardised USD volume across venues.
- Volume and other venue fields keep each venue's own meaning.
Forecast performance
No verified public results yet.
No verified public forecasting performance is available. Riddle's forecasting performance is unknown, and no estimate, sample or illustrative figure is published in its place. The method below is a proposal for how results would be recorded once they exist.
Proposed forecast methodology
Proposed, not in forceHow forecasts would be graded.
- What would be graded
- A probability between 0 and 1 published for one specific binary market, for the exact contract as that venue words it. Attention scores, feed rankings and quoted cross-venue differences are not forecasts and would not be graded.
- When it would lock
- One published forecast per market, timestamped at publication and locked from that point, with the lock time shown beside every scored row.
- Which markets would qualify
- Binary markets that resolve on a published rule set and were live on a venue Riddle covers at lock time. Markets that void, cancel or change settlement terms after lock would be listed as excluded, with the reason.
- How it would be scored
- Brier score at resolution: the squared difference between the locked probability and the outcome. Lower is better. 0.00 is a perfect forecast and 0.25 is equivalent to always saying fifty-fifty. A score computed on a handful of resolutions says very little, so any score would carry its resolved count beside it.
- The baseline
- The venue bid-ask midpoint observed at lock time, where available. Beating that baseline would be the test, and failing to beat it would appear in the same table.
- Publication and corrections
- An immutable published record: a scored row would not be deleted or rewritten, and any correction would be appended with what changed and when. Each row would carry the market, venue, locked probability and lock time, baseline, outcome and both scores.
This process is proposed and not implemented. Nothing on this page should be read as a published result.
Corrections and questions
Tell us when something looks wrong.
If a record, match or description here is wrong, write to rishi@getriddle.ai with the venue and market identifier. Desks evaluating the data can ask for the coverage, field and timestamp detail for their question.