95% block interval -0.003414 to +0.005436.
MODEL 3 · STEP 5Stress
Stress
the result.
Audit discrimination, calibration, flatness, five-run stability, fold consistency, null tests, and regimes against 9 frozen gates.
MARKET CONTEXT · OUTSIDE MODEL FREEZENKE
These readings update independently. They are not model inputs, deployment evidence, or a trading recommendation.
- Latest price
- $38.40 · fallback
- 20D momentum
- -7.9% Close-to-close · 20 completed sessions
- RSI 14
- 40.1 neutral · Wilder method
- Relative volume
- 0.7× Versus the prior 20 completed sessions
TECHNICALS THROUGH · TIMESTAMPED FALLBACK
2 of 9 gates pass. Deployment stays off.
The evidence contract covers improvement uncertainty, discrimination, log loss, calibration, fold and run stability, forecast flatness, a circular-shift null, and regime sensitivity. The corrected protocol uses a development-defined volatility threshold and enumerates every eligible circular shift. It was not externally preregistered.
TERMINAL HOLDOUT · AUDIT-CORRECTED RERUN
A small point estimate does not survive uncertainty.
Candidate selection, tuning, scaling, and calibration exclude these terminal labels. The corrected protocol was frozen before this rerun, but an earlier provisional run inspected the same period; this is not independent confirmation or external preregistration.
Bootstrap lower bound 0.448.
Holdout loss comparison versus the frozen prior.
Five-bin ECE; slope 2.868 and intercept -0.668.
Probability standard deviation; P95–P05 span 6.4%.
NO-GO · research only.
| Gate | Rule | Observed | Result |
|---|---|---|---|
| G1 | Holdout Brier improvement > 0 and 95% block-bootstrap lower bound > 0 | +0.000977; 95% CI -0.003414 to +0.005436 | FAIL |
| G2 | Holdout ROC AUC >= 0.52 and 95% block-bootstrap lower bound > 0.50 | AUC 0.564293; 95% lower 0.447774 | FAIL |
| G3 | Holdout log-loss improvement versus the frozen prior > 0 | +0.001946 | PASS |
| G4 | Calibration slope 0.70-1.30 and absolute intercept <= 0.10 | slope 2.8683; intercept -0.6677 | FAIL |
| G5 | Five-bin equal-frequency expected calibration error < 5% | 0.106663 (10.67%) | FAIL |
| G6 | Positive development-fold Brier improvement in at least 60% of folds | 6 / 10 (60.0%) | PASS |
| G7 | At least 4 of 5 terminal-holdout runs improve Brier; run probability MAD <= 1% | 4 / 5 positive; mean absolute run dispersion 1.1001% | FAIL |
| G8 | Forecasts are non-flat: holdout probability std >= 2% and P95-P05 span >= 8% | std 2.0205%; span 6.4235% | FAIL |
| G9 | Exact circular-shift p < 0.05 and no development-defined holdout regime loses more than 2 Brier points | shift p 0.1116; worst regime -0.001783 | FAIL |
| Run | Seed | Brier improvement | AUC | ECE | Probability std. |
|---|---|---|---|---|---|
| 1 | 20260801 | +0.000782 | 0.555 | 10.4% | 2.5% |
| 2 | 20260802 | +0.001776 | 0.608 | 11.2% | 2.3% |
| 3 | 20260803 | +0.000513 | 0.568 | 11.2% | 2.2% |
| 4 | 20260804 | -0.000624 | 0.513 | 10.1% | 2.7% |
| 5 | 20260805 | +0.001236 | 0.568 | 10.8% | 2.8% |
NEXTStep 6 · Summary and recommendation
Continue through the audit-corrected Model 3 workflow.