my-NexusTradingForexBot/docs/forensics/FORENSIC_RECOVERY_REPORT.md

FORENSIC RECOVERY REPORT — 2026-09-04T08:30+0330

Operator: Nexus Main (recovery architect)
Scope: Every stash, worktree, branch, and uncommitted diff reachable from main @ 706269f1
Rule: No stash popped/dropped/deleted. Snapshots exported before any mutation. main verified buildable after every batch.

---

1. True Repository State (collected 08:40)

Signal	Value
Current branch	main
HEAD	706269f1 — fix release.yml CRLF corruption (backspace fix)
origin/main	706269f1 (in sync)
Working tree	clean (git status --short empty)
Staged	empty
Untracked (main wt)	none
Total local branches	~95 (mostly hermes-subagent/*)
Worktrees	16 live — 1 main + 1 .worktrees/subagent-sa-0-8c5a9a11 + 14 detached temp/verify/cache wts + 1 pinc-stash-rescue
Prunable wts	3 under scratch/rungate/ (781097ee, ddcbc9b7, eebefab4) — no filesystem backing
Stashes	6 — all created 2026-09-04 08:04–08:38 +0330
Recent tags	v9.0.9 @ 62cfc512, v9.0.8 @ 3cd6b6a7

Stash inventory ( git stash list --date=local )

#	Timestamp	Message	Base commit	Kind
stash@{0}	08:38:25	On hermes-subagent/subagent-sa-1-6eccae4e: hold-304-site-dirty-for-main	16e86d70 (flagship pro site)	2-parent merge stash, no untracked
stash@{1}	08:36:30	On hermes-subagent/subagent-sa-2-4e6dc39d: hold-303-docs-site-three_model for main pickup	959d7d90 (status v9.0.8 drift)	site _site + asset revert
stash@{2}	08:34:02	On hermes-subagent/subagent-sa-1-6eccae4e: hold-303-docs-site-three_model for main pickup	16e86d70	3 tracked files + 30 untracked probes
stash@{3}	08:29:31	On main: hold-302	66555ea7	docs + build_site 615-line cut + site regen
stash@{4}	08:24:16	On main: tmp-site	66555ea7	build_site 615-line cut + site regen
stash@{5}	08:04:53	On hermes-subagent/subagent-sa-2-4e6dc39d: css-js-old	16e86d70	site/assets/styles.css + site/assets/search.js only

Branch audit

• hermes-subagent/*: ~90. Most report 1177–1179 ahead because they were forked from a stale commit (16e86d70 / d6c0c1a3 / fbbac37d) long before the current main. Effective delta vs main is tiny — verified via git diff main..branch --stat | grep -v site/_site. Only two carry non-site weight:
  • hermes-subagent/subagent-sa-0-8c5a9a11 — docs premium pack (already merged to main via b50f85a6 / 7150e3de path; branch is now behind/ahead noise)
  • hermes-subagent/subagent-sa-2-4e6dc39d — 2 commits ahead of its base, docs merge + flagship status; no ML/execution change beyond site
• pinc-stash-rescue (also worktree C:/tmp/pinc-stash-wt): 1 merge commit 0c90725b over 797481f7. Diff vs main deletes installer/install.ps1 bulk, docs, scripts/dev, and reverts 4 capital-safety fixes — see §4. Classified REJECT.
• nse/checkpoint/mt5-pipeline-stash-20260903: a76b0a92 — foreign-wip stash checkpoint over 9ac6425c; bulk deletions, not for main.

Dirty worktrees (non-main)

Worktree	Dirty file	Nature
C:/Users/.../nse_bug223_failsbefore	CRLF warning only on test_audit_db_default_isolation_bug223.py; no content diff	noise
C:/Users/.../nse_qa_head_wt	tests/unit/test_release_system.py +119 lines — BUG-160 fail-before proofs (verify_checksums_resolve_from_installed_layout, tamper, staged root)	test-only, deferred
C:/Users/.../nse-relcert/scripts/cert/	untracked scripts/cert/	pending BUG-160 tooling — not for main until review

Agent processes

No live Hermes subagent processes detected at audit time. All hermes-subagent/* branches are quiescent commits; no PIDs holding the worktrees.

---

2. Stash Forensics (per-stash, patch-level)

All patches exported to forensic_recovery_20260904/stash-{0..5}.patch (tracked) and stash-{i}-tracked.patch.

Stash	Parents	Tracked diff stat	Overlap	Generated artifacts?	Conflicts with main?	Completeness	Provenance
stash@{0}	16e86d70 + index 1e2dcb1b	0 lines — git diff stash@{0}^1..stash@{0} empty	—	no	n/a	empty working stash (index had no content)	hermes SA-1 hold-304, appears to be a clean-stash of an already-clean wt
stash@{1}	959d7d90	339 files, +344 −3,352 — almost entirely site/_site/*.html (rev footer d267fbd → 959d7d9) + site/assets/search.js deletion of 1,504-line FLAGSHIP 9000 JS PREMIUM PACK (1500× void 0; // flag-js-####)	overlaps stash@{3}, stash@{4} site footer updates	yes — site/_site/ is generated	parent is 117 Commits behind main (959d7d90 lacks 3cd6b6a7 int-label fix, 8ba90681 era fix, 62cfc51 BUG-239, 706269f1 release.yml fix)	partial, stale	SA-2 docs-site hold — stale base makes it a regression if applied
stash@{2}	16e86d70	3 files, +16 −4 — docs/project/status.md (6→8), scripts/docs/build_site.py (+docs-enhance wiring), src/nexus_scalp/model_generation/three_model.py (+compute_70d_frame_fast for 70d) ; plus 30 untracked: .hermes/plans/..., _splice.py, scratch/ns_* probes, scratch/ci760/runtime_gate.json, etc.	three_model overlaps stash@{3} identically; docs-enhance/build_site overlaps stash@{3}/stash@{4}	site/_site not in tracked diff; untracked are scratch/probe — not to commit	parent 16e86d70 is 8 commits behind main; tracked patch applies cleanly except status bump now stale (main already at 9.0.9)	valid but mixed — one clean ML perf fix, one contested docs wiring, one obsolete status bump, plus scratch noise	SA-1 hold-303
stash@{3}	66555ea7	334 files, +~6349 −11715 (filtered non-site: docs/project/status.md, scripts/docs/build_site.py (615-line FLAG_BUILD_INDEX_0000…0140 deletion + docs-enhance wiring), site/assets/search.js FLAG-JS deletion) + site _site regen + src/nexus_scalp/model_generation/three_model.py same fast fix	three_model identical to stash@{2}; build_site FLAG removal overlaps stash@{4}; site footer overlaps stash@{1}/stash@{2}	site/_site generated; FLAG_BUILD_INDEX tail after raise SystemExit is dead code	parent 66555ea7 is behind main (missing 62cfc51, 706269f1); three_model patch still applies; FLAG removal applies cleanly but needs separate review	valid subset	main hold-302
stash@{4}	66555ea7	site/assets/search.js FLAG removal + scripts/docs/build_site.py 615-line FLAG deletion + site _site regen — no src/ change	build_site/site overlaps stash@{3}; site/assets overlaps stash@{1}	site _site + dead FLAG code	same parent as stash@{3}	docs-only	main tmp-site
stash@{5}	16e86d70	2 files +302 −1 — site/assets/styles.css (+288 FLAGSHIP expansion) + site/assets/search.js (+14 module inlines)	site assets overlap stashes 1/3/4 with opposite polarity (adds vs removes)	asset files are source for site build, but content is FLAGSHIP cosmetic	no src/	docs-only, superseded by later flagship merges on main	SA-2 css-js-old (oldest)

Key overlap map

stash@{2} three_model == stash@{3} three_model   (byte-identical patch)
stash@{1}/@{3}/@{4} build_site FLAG removal      (same 615-line cut, parent-dependent)
stash@{1}/@{3}/@{4} site/_site footer            (same rev bump, different base rev)
stash@{5} styles/search                          (+FLAG) vs stash@{1}/@{3} (-FLAG)  — CONTRADICTION

No stash contains a production ML artifact, key, or secret.

---

3. pinc-stash-rescue — High-Risk Revert Analysis

git diff main..pinc-stash-rescue --stat shows deletions across installer/, docs/, scripts/dev/, site/, plus 4 reverted safety gates:

Revert	File	What was removed	Why it must stay blocked
ScalerBundle.is_ready	src/nexus_scalp/application/live_engine.py:159	Zero/negative/non-finite std check — reverts to mean is not None and std is not None	Silent ÷0 → ±inf → nan_to_num → -1.0 poison, no-silent-fallback violated
LiveEngine._rebind_live_temporal_contract	live_engine.py:440-560	120-line temporal-contract rebind + gap invalidation	Trades on wrong L/max_gap_us, sequence/tensor mismatch, degrades BLOCKED→AVAILABLE
CHECK-MDL-03 era fix	src/nexus_scalp/forensics/checks_features.py:372	Artifact-own feature_schema_id dimension truth → reverts to lagging ACTIVE_SCHEMA_ID	Every canonical 70D 70d_liquidity/scalp_v3 deployment flagged CRITICAL false positive
ARCH-SEQ-UNIFY SSoT	src/nexus_scalp/model_generation/sequence.py	Re-exports from temporal_contract → hard-coded literals (L=32, 70)	Breaks single-source-of-truth; drift between train/live/replay

Verdict: Branch is an inverted stash (hard git stash push --include-untracked + selective commit attempt). It captures a valid WIP moment but, merged as-is, would re-introduce 4 P0/P1 regressions. Do not merge. No cherry-pick without per-hunk triage. Preserved as pinc-stash-rescue for manual hunk recovery if needed.

---

4. Special Deep Audit Areas — Result

• Model/ML pipeline: Only real ML code in stashes is three_model.py fast-path (see §5). No scaler, tensor shape, dataset hash, label, champion, torch.load, or retrain change in any stash requires action. pinc reverts above are the only pipeline risk — blocked.
• Live trading/execution: No execution, sizing, risk-limit, or order-routing change in any stash. pinc's gap/tensor revert would have permitted trading on degraded model — blocked.
• Persistence/state: No SQLite/Postgres, flush, or candidate/champion write in any stash. No migration. Test nse_bug223_failsbefore shows only CRLF noise. No corruption risk.
• Security: No model-swap endpoint, pickle/torch.load, API auth, or file-load change in stashes. No path traversal.
• Data/broker: No spread, OHLCV, or state-machine change in stashes.

---

5. The Only Verified Production Fix — three_model.py 70d Fast Path

Signal	Detail
Source	stash@{2} and stash@{3} identically: build_feature_frame(..., variant!=50d_main) → compute_70d_frame → compute_70d_frame_fast
Semantics	compute_70d_frame_fast docstring: "byte-identical to compute_70d_frame, same contract, same columns, same values — O(n·window) vs O(n²)"
Verification (isolated)	200 synthetic M1 bars → slow vs fast: cols equal 80/80, rows 146/146 each, max │feat_0 diff│ = 0.0, max │feat_0..4 diff│ = 0.0
Lint	Raw patch left dead from schema_v2 import compute_70d_frame → ruff F401 — repaired by removing dead import; final ruff check clean
Tests	test_three_model_pipeline 5/5, test_70d_bug106_incremental_phase19 2/2, test_70d_incremental slow suite (opt-in) — all green; py_compile clean
Risk	LOW — perf only, preserves OOS parity, no new branch

---

6. What Was NOT Recovered and Why

• stash@{0} empty — nothing to recover (duplicate of a clean stash).
• stash@{1} — stale base + site-only + would revert 3cd6b6a7 int-label fix if forced; rejected as obsolete/duplicate.
• stash@{2} status bump (6→8) — superseded (main now at 9.0.9). Not applied.
• stash@{2} docs-enhance wiring + stash@{2}/@{3} 615-line FLAG tail — deferred P3 (see blockers). main currently has 600 FLAG_BUILD_INDEX lines after raise SystemExit — dead code, ruff clean, py_compile clean, but intentionally not cleaned in this recovery batch to keep scope minimal. Wiring for docs-enhance.css/js was removed at d267fbd4 to fix F841 (unused locals) — the variables were used in the prior shell() template but the late revert removed the template refs. Re-wiring needs a single coherent PR (template + copy loop + assets) with a ruff/build_site end-to-end build test — not a drive-by stash pop.
• stash@{2} untracked scratch/ns_* probes, .hermes/plans, _splice.py — rejected (tracked scratch/ probes are per-memory rule noise; scratch/ is tracked but probes are ephemeral).
• stash@{4} / stash@{5} — docs-only, overlapping and contradictory (add vs remove FLAGSHIP), no production code — deferred.
• pinc-stash-rescue — rejected per §3.
• nse_qa_head_wt BUG-160 +119 — sound fail-before proofs; deferred to BUG-160 branch (needs release.yml pre-stage + ISCC contract first — see that branch's logic).

---

Appendices

• Forensic snapshots: forensic_recovery_20260904/stash-{0..5}.patch, stash-{i}-tracked.patch, stash-{i}-parent.txt
• Post-merge verification: ruff check clean on three_model.py; pytest test_three_model_pipeline -q 5 passed
• No stash was popped/dropped. pinc-stash-rescue worktree left intact at C:/tmp/pinc-stash-wt.