Calibration evidence
Every priced scan record is precommitted before the market resolves; losses and non-actionable forecasts are kept. Confidence is earned only through independent, resolved events — never asserted.
Report generated 2026-08-27T09:58:19.784891+00:00 Evidence ledger verified (1264858 records)
Overall reliability
| Band | n | Mean predicted | Actual hit rate |
|---|---|---|---|
| 0.0-0.1 | 8637 | 3.0% | 4.4% |
| 0.1-0.2 | 4155 | 15.1% | 15.9% |
| 0.2-0.3 | 4415 | 24.9% | 27.1% |
| 0.3-0.4 | 2447 | 32.9% | 34.2% |
| 0.4-0.5 | 418 | 44.7% | 25.6% |
| 0.5-0.6 | 292 | 54.4% | 34.6% |
| 0.6-0.7 | 174 | 64.5% | 44.3% |
| 0.7-0.8 | 111 | 74.9% | 45.9% |
| 0.8-0.9 | 58 | 84.3% | 60.3% |
| 0.9-1.0 | 52 | 95.6% | 76.9% |
Prospective evidence by exact model version
Only rows produced under the exact deployed version can unlock that version's execution interlock.
| Family | Model version | Precommitted | Resolved rows | Independent events | Brier |
|---|---|---|---|---|---|
| economics.core_cpi | core-cpi-official-ensemble-v2 |
1960 | 0 | 0 | — |
| economics.gdp | gdp-official-ensemble-v2 |
406 | 0 | 0 | — |
| economics.headline_cpi | headline-cpi-official-ensemble-v2 |
4028 | 141 | 3 | 0.0448 |
| economics.labor | labor-official-history-v1 |
3920 | 0 | 0 | — |
| economics.recession | spf-recession-v1 |
58 | 9 | 2 | 0.0319 |
| energy.henry_hub_spot | henry-hub-post-issuance-annual-max-v3 |
315 | 0 | 0 | — |
| energy.henry_hub_spot | henry-hub-seasonal-annual-max-v2 |
7 | 0 | 0 | — |
| sports.world_cup | world-cup-live-bracket-elo-v2 |
374 | 293 | 44 | 0.121 |
| weather.hurricane_season | noaa-seasonal-count-v2-update-aware |
1125 | 0 | 0 | — |
| weather.temperature | weather-ensemble-v2 |
1920 | 1920 | 4 | 0.1303 |
| weather.temperature | weather-ensemble-v3-power-calibrated |
21996 | 18396 | 39 | 0.1266 |
Retrospective model-development validation
weather.temperature: weather-ensemble-v3-power-calibrated was tested by applying its
power_transform (gamma 0.65) to 240 independent events from
weather-ensemble-v2.
Brier: 0.1303 → 0.1256. Walk-forward improved: yes. This does not count toward prospective promotion.
Promotion readiness by family
| Family | Independent events | Next checkpoint | Concordance | Closing coverage | Eligible |
|---|---|---|---|---|---|
| economics.headline_cpi | 3 | 30 | no checks yet | 100.0% (3 groups) | accumulating |
| economics.recession | 2 | 30 | no checks yet | 100.0% (2 groups) | accumulating |
| sports.world_cup | 44 | 100 | no checks yet | 50.0% (22 groups) | accumulating |
| weather.temperature | 39 | 100 | 100.0% (1561 checks) | 100.0% (1561 groups) | accumulating |
all priced scan records are precommitted; losses and non-actionable forecasts are retained
Machine-readable: https://api.trueodd.xyz/v1/calibration. Each resolved forecast is receipted — verify receipts.