SniperGold_ML/ml/p3/p3_s2211_debt_closure/README.md

# P3-S22.1 VERIFICATION DEBT-CLOSURE — DISPOSABLE / VERIFICATION-ONLY

**Status**: disposable research/audit namespace. NOT production, NOT ML
training, NOT contract changes.

## Scope

Closes the four P3-S22 verification-debt items:

| Debt item | Script | Output |
|---|---|---|
| 1. F2 numeric zone-boundary parity | `s2211_s1_f2_boundary.py` | `output/s2211_s1_f2_boundary.json` |
| 2. Historical M30/H4 boundary-timestamp parity | `s2211_s2_mtf_boundary.py` | `output/s2211_s2_mtf_boundary.json` |
| 3. Population-level terminal replay | `s2211_s3_population_feasibility.py` | `output/s2211_s3_population_feasibility.json` |
| 4. P3-S21.1 coefficient-level re-audit | `s2211_s4_coef_reaudit.py` | `output/s2211_s4_reaudit.json`, `s2211_s4_coefficients.csv`, `s2211_s4_repro.json` |

## Governing documents

- `docs/CODE_VERIFICATION_AND_SILENT_BUG_POLICY_v1.md` (authoritative).
- `docs/P3_S22_TERMINAL_INDICATOR_VERIFICATION.md` (debt list §X).
- `docs/P3_S12_ZONE_CONTRACT_REPAIR.md`, `docs/SMC_FVG_SPEC_v1.md`,
  `docs/SMC_ORDER_BLOCK_SPEC_v1.md` (F2 contract).
- `docs/SNIPERGOLD_CANONICAL_SETUP_CONTRACT_v1.md` (§L MTF as-of; §C/§N gates).

## Reproducibility

Every script is deterministic; expect byte-identical outputs on repeat runs
except the `generated_utc` metadata field. Run twice and compare:

```
python s2211_s1_f2_boundary.py
python s2211_s2_mtf_boundary.py
python s2211_s3_population_feasibility.py
python s2211_s4_coef_reaudit.py
```

## Findings recorded (see the debt-closure report)

- **DISCREPANCY-ADJ-1** (Subphase 1): MQL5 production `AF_FVGZoneState` /
  `AF_OBZoneState` never assign `mit_state = AF_ZONE_PARTIALLY_FILLED`; a
  partially-filled zone is labeled `UNMITIGATED(0)` + `partial_filled` flag,
  while the frozen F2 contract and the committed research/reference chain
  encode it as `PARTIALLY_FILLED(1)`. Consumer availability is identical
  (IsActive). Severity LOW; no consumer-visible or historical-result impact
  found. NOT repaired in this phase.
- **DISCREPANCY-ADJ-2** (Subphase 2): the committed M30 gate is constant only
  within index-pair M15 segments (research resample construction), not within
  fixed-UTC 30-min buckets; it drifts across session gaps. H4 is constant
  within the frozen as-of grid (0 violations) except 9 US-market-holiday
  artifacts. Severity LOW-MODERATE; no setup-level impact demonstrated.
- **Subphase 3**: population replay classified
  `UNVERIFIED - ENVIRONMENT/DATA REPRODUCTION LIMITATION` (proven, not PASS).

## Rules

- No production MQL5 file, no F1-F4 contract, no FEATURE_CONTRACT, no label
  contract, no dataset, no model is modified by anything in this folder.
- All mutation testing is performed on disposable in-memory copies only.
- This folder is committed only as audit evidence; it is not a runtime
  component.