Estimates with probabilities frozen in the ledger.
01 / Public track record
The calibration record.
Every forecast is logged with its probability frozen at issuance. When the named source resolves the question, the outcome is scored once and stays visible.
Outcomes recorded once and scored once.
Lower is better. A coin-flip baseline scores 0.25.
Calibration by probability bucket
The pale bar is the mean forecast. The amber bar is how often the event occurred. With a meaningful sample, calibrated forecasts bring the two into alignment.
0.5–0.6
n=—
0.6–0.7
n=—
Resolved forecasts
Rows appear from the verified calibration feed. Until then, no synthetic data stands in for a public record.
| Forecast | Frozen p | Resolved | Outcome | Brier | Log loss |
|---|---|---|---|---|---|
| No resolved forecasts are published yet. | |||||
Honest reading
A forecast ledger is useful only when its limitations are reported with the same precision as its scores.
The record is young, and it says so. Forecasts are logged with frozen probabilities now; scores appear as real outcomes arrive. Until the resolved sample passes roughly 30, bucket statistics are noise. Signal Atlas does not backfill, edit a frozen probability, rescore an outcome, or present small-n results as evidence of skill.
How scoring works
Two proper scoring rules and two baselines keep a plausible story from masquerading as a measurable record.
Brier score
(p − outcome)². A perfect forecast scores 0. Always forecasting 0.5 scores 0.25.
Log loss
Confident wrong forecasts are penalized sharply. Cautious uncertainty is treated differently from false certainty.
Two baselines
Results are compared with a coin flip and the observed base rate. If the record beats neither, it says so.
Boundary
This page is track-record accounting for research packets. It is not evidence of trading skill, a solicitation, or financial or betting advice. Past calibration cannot guarantee future calibration.