A prediction-market price is a tradable crowd estimate constrained by its contract and liquidity. Read the exact resolution rule, timestamp and spread before treating price as probability. HyperFX compares like-for-like probabilities and keeps crowd price separate from its own model estimate.
What to remember
- Read the resolution source first.
- Price is informative, not infallible.
- Low liquidity can distort probability.
- Calibration needs many resolved questions.
The contract wording defines the event
Two similar headlines can settle differently because one uses a date boundary, named data source or precise threshold. Read the full YES condition, deadline, resolution source and void rules before interpreting the price.
HyperFX presents external market data as a distinct source. It does not imply affiliation, and the crowd estimate must not be blended invisibly into the AI probability. Keeping the two values separate makes disagreement inspectable.
- Question and deadline
- Resolution source
- YES and NO conditions
- Void or ambiguity rules
- Source timestamp
Liquidity, spread and price quality
A last traded price may be stale; the best bid and ask show what can actually trade. Wide spreads, shallow books and concentrated holders make the headline probability less reliable. Volume alone does not guarantee current depth.
Compare probabilities at the same moment and for the same outcome. For mutually exclusive outcomes, normalise when fees or market design create an overround. A visible difference is not usable edge unless it exceeds modelling error and execution cost.
A 60% price means the market trades near 0.60 under its rules; it does not mean the event is known to occur six times out of ten.
Measure forecasts after resolution
Calibration asks whether events forecast near 70% occur roughly 70% of the time across a sufficiently large sample. Accuracy alone rewards always choosing favourites and ignores confidence. Brier score and log loss penalise overconfidence more directly.
Keep unresolved and void contracts outside result denominators. Segment politics, sports, crypto and macro questions because information flow differs. Past crowd accuracy and model accuracy are evidence about a process, not a guarantee for the next event.
- Calibration buckets
- Brier or log-loss score
- Representative sample
- Category segmentation
- Immutable forecast history