Validation layer
Proof & Backtests
Runs
2
Proof state
fixture
Archive status
100/100
computed from 3125 settled/graded canonical picks vs the 100-pick platform readiness floor (MIN_SETTLED_PICKS_FOR_LEARNING) — calibration report: collecting (below the performance-stats gate)
Next milestone
store predictions
One prediction stored = proof starts
Proof Readiness
computed from 3125 settled/graded canonical picks vs the 100-pick platform readiness floor (MIN_SETTLED_PICKS_FOR_LEARNING) — calibration report: collecting (below the performance-stats gate)
Proof Readiness · 100 / 100
Predictions are logged, settled, and checked once live data flows
Proof Readiness is computed from real settled/graded pick counts against the platform's own readiness floor, never hand-typed. It reads 'unmeasured' instead of a number whenever the database isn't reachable at render time, and it rises automatically as real picks settle. Every pick is archived with the model version, input data snapshot, and outcome, making the calibration score auditable.
2 runs recorded
Backtest Archive
Fixture, not a production record. These rows demonstrate the proof-archive UI and scoring math; they are not real settled predictions.
| run id | type | status | mae | calibration | proven |
|---|---|---|---|---|---|
| bt_fantasy_fixture | fantasy_projection | fixture_backtest | 5.8 | limited | UI and scoring math can display proof … |
| bt_volatility_fixture | volatility | fixture_backtest | — | — | — |