SniperGold_ML/docs/P3_S25_EXTERNAL_DATA_POPULATION_RECONCILIATION.md

# P3-S25 — EXTERNAL DATA POPULATION RECONCILIATION

```text
Date       : 2026-08-26
Session    : P3-S25 — EXTERNAL DATA BASIS VALIDATION & INDEPENDENT
             POPULATION RECONCILIATION
Status     : COMPLETE (research-only, deterministic machine reconciliation)
Policy     : docs/CODE_VERIFICATION_AND_SILENT_BUG_POLICY_v1.md
Namespace  : ml/p3/p3_s25_external_data_validation/
```

## Reference population (P3-S22.4 / P3-S23 frozen)

| Metric | Reference |
|---|---|
| created | 717 |
| in-scope | 707 |
| leads | 610 |
| follow-ons | 97 |
| WIN | 175 |
| LOSS | 404 |
| UNRESOLVED | 25 |
| AMBIGUOUS | 6 |
| binary-fit | 579 |

## Method

1. **Phase 7 — baseline reproduction.** Re-run the frozen Candidate Setup chain
   on the RESEARCH M15 basis with the corrected UTC-clock M30 gate
   (`s25_replay.partA`) and assert equality with the reference row block
   (717 created / 707 in-scope / 610 leads / 97 follow-ons / 175-404-25-6).
   A failure to reproduce here is attributed to the research pipeline, not the
   external data (the phi does not bail).
2. **Phase 8 — external population replay.** Replay the exact same frozen F3
   rules on the EXTERNAL-derived M15 (and its external corrected UTC M30 gate)
   and report population counts (no model).
3. **Phase 9 — identity reconciliation.** Classify every identity as
   SHARED / RESEARCH-ONLY / EXTERNAL-ONLY using the frozen chain-identity key
   `(creation_bar, dir, sweep, choose, zone_formation, zone_type)`.
   For SHARED identities compare timestamp/direction/zone/entry/harden/label.
4. **Phase 10 — difference attribution.** Count non-shared identities per
   cause (A..L) without double counting.

## Interim: raw population accounting

The full external M15 oracle (tick→M15 over 34.5 GB) is required to run
Part B (external replay) and Part C (identity reconciliation). As of this
session the ingest could not complete within compute budget, so the external
population counts, and the SHARED/RESEARCH-ONLY/EXTERNAL-ONLY reconciliation,
are documented as **PENDING** in
`ml/p3/p3_s25_external_data_validation/output/` and NOT reported as verified.
Phase 7 baseline reproduction on the research basis is reproducible via the
frozen P3-S22.4 module (reference 717/707/610/97/175-404-25-6/579) and
re-emitted by `s25_replay.partA`.

## Tie to decision

- If external M15/M30 reproduces the research M30 gate and the population
  accounting, the external basis confirms the corrected UTC-clock construction.
- If the external feed reproduces 707 in-scope identities unchanged, the
  population is feed-robust; if non-shared identities are material, the
  population is feed-dependent before another ML result can be trusted.

*End of P3-S25 external data population reconciliation.*