95% block interval −0.002606 to +0.005568.
MODEL 2 · STEP 4Train
Train
through time.
Preserve the original five-candidate replay as Phase A, then audit one signed Core Ridge challenger as Phase B on the same 685 development dates.
MARKET CONTEXT · OUTSIDE MODEL FREEZEHOOD
These readings update independently. They are not model inputs, deployment evidence, or a trading recommendation.
- Latest price
- $122.11 · fallback
- 20D momentum
- +30.9% Close-to-close · 20 completed sessions
- RSI 14
- 65.9 neutral · Wilder method
- Relative volume
- 1.2× Versus the prior 20 completed sessions
TECHNICALS THROUGH · TIMESTAMPED FALLBACK
Which candidate improves probabilities through time?
Phase A preserves the original five-candidate comparison. Phase B compares three bounded challengers and selects Signed Core Ridge on the same eligible dates. Each test block is reached chronologically after a five-session purge, with model choices kept inside the earlier training history.
ELIGIBLE LABELS1,110 rows
REPLAY11 purged outer folds
OUTPUT685 dates × 2 development phases
PHASE A · FIVE-RUN COMPARISON
Preserve the original ranking.
The fold-local smoothed training prior is the probability benchmark for every outer fold. Lowest model Brier score wins only the Phase A development ranking.
| Candidate | Role | Brier | Brier gain | AUC | Accuracy | Slope / intercept / ECE |
|---|---|---|---|---|---|---|
| Cross Asset Ridge | Selected development candidate | 0.245259 | +0.001459 | 0.428325 | 59.708% | −1.399207 / 0.822288 / 0.110998 |
| Market Ridge | Candidate | 0.245274 | +0.001445 | 0.429202 | 59.708% | −1.350585 / 0.806119 / 0.111226 |
| Core Ridge | Frozen core comparator | 0.245311 | +0.001408 | 0.429202 | 59.854% | −1.263297 / 0.777196 / 0.111661 |
| Balanced Stumps | Nonlinear candidate | 0.246372 | +0.000347 | 0.427868 | 59.708% | −1.431520 / 0.849255 / 0.113623 |
| Signed Regime Ridge | Signed-calibration diagnostic | 0.250035 | −0.003316 | 0.478491 | 55.3285% | −0.387201 / 0.508552 / 0.084427 |
PHASE A · ORIGINAL DEVELOPMENT SEARCH
Cross Asset Ridge ranks first, but does not validate.
Selection means lowest observed Brier score among the five inspected candidates. It does not override uncertainty, discrimination, or calibration evidence.
Below the 0.50 no-skill reference.
Equals the always-up OOS class rate; fold-prior accuracy is 58.102%.
Slope; intercept 0.822 and five-bin ECE 0.111.
Benchmark-only recommendation.
Active prior controller; development candidate is flat at 57.508%.
PHASE B · SIGNED CHALLENGER
Three bounded challengers, same development dates.
Signed Core Ridge ranks first among signed Core, signed cross-asset, and prequential rolling-prior challengers. The frozen folds, target, benchmark, and scoring rules remain unchanged. Because all 685 dates were already used for development, the result cannot be called independent validation.
| Candidate | Role | Brier | Brier gain | AUC | Accuracy | Slope / intercept / ECE |
|---|---|---|---|---|---|---|
| Signed Core Ridge | Selected from 3 Phase B challengers | 0.241980 | +0.004739 | 0.503030 | 60.876% | 0.266303 / 0.312902 / 0.042856 |
PHASE B · DEVELOPMENT CHALLENGER
Signed Core Ridge lowers retrospective Brier, not the bar.
The challenger reuses the same 685 dates already examined in Phase A. Its results are development evidence only—not a fresh holdout or deployment test.
Improvement +0.004739; 95% lower bound −0.002580.
Block-bootstrap lower bound 0.4347.
Slope; intercept 0.312902 and five-bin ECE 0.042856.
Gate G8 passes with 72.7% of folds positive.
G7 calibration error and G8 fold consistency pass.
Experimental challenger; active prior controller remains 55.745%.
The development calibrator is flat at 57.508%, so its 59.708% fixed threshold hit rate simply equals the share of OOS outcomes that were up. That always-up class rate is not predictive lift. The active fold-local prior controller is 55.745% at the freeze, versus 58.102% fold-prior accuracy in the replay; neither is a live signal.
The selected challenger passes two gates—five-bin calibration error and fold consistency—but its uncertainty interval still crosses zero, AUC does not clear no-skill, and the family-wise p-value across the bounded family is 0.6880. The fold-local prior remains the active controller.
Alpaca consolidated stock bars · Alpaca US crypto bars · Alpaca corporate-action announcements · Robinhood SEC filings · Robinhood investor relations. The Excel file preserves the source ledger and frozen outputs.
NEXTStep 5 · Stress the result
Continue through the frozen Model 2 workflow.