Signal
The model sees enough edge after costs and the risk gates are clear.
Methodology
How Forecast Alpha turns market prices into auditable signals.
Forecast Alpha is built around a simple standard: a probability gap is not enough. A useful prediction-market read must show the model estimate, market price, costs, risk gates, and proof trail before it earns a decision label.
Research only. Not financial advice. Prediction markets involve risk including total loss.
Decision pipeline
Evidence to probability to proof
Read the market
Pull active contracts, prices, order-book context, metadata, settlement timing, and category-specific features.
Estimate fair probability
Compare market-implied probability with a model estimate built from event context, crypto features, historical outcomes, and confidence signals.
Subtract friction
Convert raw disagreement into expected value after modeled fees, spread, slippage, timing, and execution constraints.
Apply refusal gates
Block weak reads when liquidity, ambiguity, stale data, risk, settlement wording, or confidence does not clear the threshold.
Record before outcome
Persist model reads before resolution so Brier score, calibration, and beat-rate calculations can be audited later.
Decision gates
The model produces three public decision states.
The labels are meant to reduce false urgency: some markets are actionable research signals, some are worth watching, and many should be refused.
Watch
The market is interesting, but the edge, confidence, or timing is not strong enough yet.
No trade
A blocking condition makes the market unsuitable despite any apparent probability gap.
Performance metrics
The model should be judged by calibration, not hype.
The public product emphasizes metrics that can be checked after resolution. Wins matter, but so do the probability quality, the cost-adjusted decision, and the cases where the system correctly refuses to act.
Brier
calibration
lower is better
Gates
decision quality
edge, risk, liquidity
Versioned
audit trail
before resolution
Brier score
Probability accuracy. Lower is better; 0 is perfect and 0.25 is chance for a 50/50 forecast.
Calibration
Whether events forecast at a given probability happen at roughly that frequency.
Beat rate
How often the model's probability scores better than the market probability on the same resolved row.
Profit factor
Paper or live P&L quality: average gain relative to average loss after costs.
Rejection rate
How often the system refuses to act because a gate blocks the read.
Model version
The exact model family used for a prediction so performance can be compared across versions.
Public brand promise
Forecast Alpha should be understood as a research terminal with verified signals and transparent probabilities. Any private/live execution system should stay separate from public marketing until audited proof and safety controls justify discussing it.
Verification
Why the track record is the trust anchor.
Forecasts are useful only if they are recorded before outcomes are known. Forecast Alpha persists model reads with timestamps, market probabilities, model versions, and outcomes once available. That lets users inspect not just whether a market was called correctly, but whether the stated probability was well calibrated over time.
FAQ
Why does Forecast Alpha show no-trade decisions?
No-trade decisions show where apparent probability gaps fail cost, liquidity, confidence, timing, ambiguity, or data-quality gates.
What is the main performance metric?
Brier score and calibration are the primary probability-quality metrics. Paper P&L and win rate are useful, but they are secondary to whether probabilities match outcomes over time.
Does Forecast Alpha provide financial advice?
No. Forecast Alpha is a research and decision-support terminal. It does not guarantee outcomes or provide financial advice.