Add dopamine RPE extension evaluation harness and falsifier reports

[neuron7xLab]

Scope

Follow-up evidence PR for GeoSync PR #1374. This issue is required because PR #1374 is STRUCTURAL only and must not claim market edge or production intelligence without evaluation artifacts.

Required deliverables

• scripts/evaluate_dopamine_rpe_extension.py
• results/dopamine_rpe_extension/EVAL_SUMMARY.json
• results/dopamine_rpe_extension/WALKFORWARD_REPORT.md
• results/dopamine_rpe_extension/PARAMETER_LOCK.json
• results/dopamine_rpe_extension/ABLATION_MATRIX.csv
• results/dopamine_rpe_extension/NULL_MODEL_REPORT.md
• results/dopamine_rpe_extension/FALSIFIER_REPORT.md

Required model comparison

• canonical_td0
• td0_plus_distributional_helper
• td0_plus_asymmetric_alpha
• td0_plus_risk_penalty
• td0_plus_vigor_helper
• expectile_ensemble_low_tau
• expectile_ensemble_mid_tau
• expectile_ensemble_high_tau
• combined_surface

Walk-forward protocol

Use at least 5 chronological folds. Report IC, Sharpe, MaxDD, turnover, hit-rate, calibration error, regime stability, null-model delta, and parameter sensitivity.

Null models

• shuffled returns
• sign-flipped reward
• random tau assignment
• constant reward
• lagged reward
• no-RPE baseline

Parameter governance

PARAMETER_LOCK.json must contain gamma, tau levels, learning rate, risk penalty coefficients, vigor gain/bounds, data window, seed, commit SHA, timestamp, promotion status, and config hash.

Dopamine neuron reversal points, p-values, alphas, and R² are not market parameters.

Promotion rule

If the module wins only in-sample, it remains P2 research. TESTED requires unit/falsifier tests plus deterministic evaluation script. EXTRAPOLATED requires walk-forward improvement, null survival, and parameter lock. ANCHORED requires repeated independent runs, stable regime performance, claim registry entry, and rollback plan.

[neuron7xLab]

⊛ Evidence Gate Contract for Issue #1408

This issue is the follow-up verification gate for PR #1374. PR #1374 remains STRUCTURAL until this issue produces deterministic evidence artifacts.

Required next PR:

1. Add scripts/evaluate_dopamine_rpe_extension.py as a deterministic CLI.
2. Generate and commit the required files under results/dopamine_rpe_extension/:
   • EVAL_SUMMARY.json
   • WALKFORWARD_REPORT.md
   • PARAMETER_LOCK.json
   • ABLATION_MATRIX.csv
   • NULL_MODEL_REPORT.md
   • FALSIFIER_REPORT.md
3. Use at least 5 chronological folds.
4. Include all model variants and null baselines listed in the issue body.
5. Fail closed when data are missing, metrics are non-finite, parameter lock is absent, or artifact schemas are invalid.
6. Add tests for schema validity, deterministic replay, non-finite rejection, missing-data failure, and null-baseline coverage.

Promotion boundary:

• Unit tests only: STRUCTURAL.
• Deterministic evaluation plus falsifier reports: eligible for TESTED.
• Walk-forward survival plus null-baseline survival plus parameter lock: eligible for EXTRAPOLATED.
• Repeated independent runs plus stable regime behavior: eligible for ANCHORED.

No claim promotion should be accepted from this issue without committed evidence artifacts.