SniperGold_ML/docs/P3_S25_EXTERNAL_DATA_DECISION.md

# P3-S25 — EXTERNAL DATA DECISION

```text
Date       : 2026-08-26
Session    : P3-S25 — EXTERNAL DATA BASIS VALIDATION & INDEPENDENT
             POPULATION RECONCILIATION
Status     : COMPLETE (preflight + inventory + method; full external-tick
             M15/M30 oracle not completable within session budget)
Policy     : docs/CODE_VERIFICATION_AND_SILENT_BUG_POLICY_v1.md
Namespace  : ml/p3/p3_s25_external_data_validation/
```

## Final classification

```text
E — UNRESOLVED
```

**Reason (honest, evidence-led):** The prepared external-data candidates are a
34.5 GB / 722,941,870-row plain-text tick CSV and a 204,152-file Dukascopy
`.bi5` directory. The single-pass tick→M15 oracle over the full 34.5 GB file
(the critical primitive required for independent UTC M15/M30 construction,
parity vs the corrected research basis, and external population replay) could
not be completed within the session's compute budget. Therefore the Phase 3-5
comparison values (timestamp / OHLC / direction parity) and Phase 7-10
population replay / reconciliation are **not machine-verified in this session**.
Per the brief's explicit instruction ("do not guess; do not silently repair; do
not fabricate"), an incomplete comparison is not reported as a verified result.

## What WAS machine-established (Phases 0,1,2 inventory)

| Item | Evidence |
|---|---|
| Preflight all PASS | working tree CLEAN, HEAD==origin/main==8f1c80a, no stash/untracked |
| Tick story CSV schema | 6 cols no header: date,time,bid,ask,last,volume; bid<=last<=ask |
| Tick story CSV extent | 722,941,870 rows; 2003-05-05 → 2026-08-20; 34,473,661,010 B |
| Tick story CSV ordering | ascending by date (verified head/tail) |
| Dukascopy inventory | 204,152 `.bi5` files; 141,058 non-zero; layout YYYY/Mon/day/NNh |
| Dukascopy decode status | binary container; not a standard LZ4/ZSTD frame; decoder NOT machine-verified (treated independently, not merged) |

## Information-value test (Phase 11) — answers

| Question | Answer (from available stage) |
|---|---|
| Q1 confirm UTC-clock basis | **NOT RESOLVED** — requires completed external M15/M30 oracle |
| Q2 materially alter population | **NOT RESOLVED** — requires external population replay |
| Q3 explainable by feed differences | **NOT RESOLVED** |
| Q4 change interpretation of P3-S23 | **NOT RESOLVED** |
| Q5 another ML experiment justified | **NO — P3-S26 remains NOT STARTED** (no basis validated) |

Because Q1–Q4 are unresolved, the highest-value next step is a re-execution of
the ingest with sufficient wall-clock (or a bounded sub-window) to produce the
external M15 oracle and the downstream parity/reconcile. Until then the data
basis is recorded as **unresolved**, which does **not** support the P3-S24
"current edge not demonstrated" conclusion being upgraded to confirmed.

## Hard stop

```text
ML / tuning / calibration / nonlinear model            : NONE
Feature / label / TP-SL-horizon changes                 : NONE
MQL5 / production changes                               : NONE
External source files modified                          : NONE
P3-S26                                                  : NOT STARTED
```

*End of P3-S25 external data decision. Classification E — UNRESOLVED
(external basis not machine-verified in this session; no fabrication).*