Backup manifest: v1.0.6 fast defect wave archive row

BACKUP_MANIFEST.md

@@ -160,3 +160,11 @@ The archives are live snapshots of 2026-11-02 taken while DJ Lab Campaign Wave E
 **Canonical replacement (parent, 2026-11-04):** `djlab-app.exe` replaced with the v1.0.5 final build — 20,492,288 B, sha256 `9e8f2cf1d1f2db50bcc89d8402a8e4dae3d37e40b11d466cae7c4dae8e363194`, HF commit `7d55fa07070713ec9401ea5d77b5936d3215be76`. Parent independently re-verified before replacing: local stat + sha256 recomputed (exact match), own PE census = exactly the same 17 imported DLLs as v1.0.4, gate record read in proof/CHARTS_WAVE_105_REPORT.md (core 144/0; CA 31/0; full gate green; AA's SAA20 heartbeat dispositioned with two solo re-runs, the second 21/0 on the identical binary; CB live cross-checks named SKIPs — the sandbox webview cannot reach the chart hosts, lanes proven against captured live payloads, no fixture proof presented as live). Served-bytes proof follows in this file's verification note (canonical downloaded back and re-hashed by the parent).
 
 - Content files after the delivery: **42**, plus this manifest = 43. Sum of file sizes: **9,917,341,571 bytes** (the canonical replacement: −20,450,816 +20,492,288).
+
+## FAST DEFECT WAVE — v1.0.6 (2026-11-04)
+
+| File | Size | SHA-256 | HF commit | What it is |
+|---|---|---|---|---|
+| `djlab-app-v1.0.6-final.exe` | 20,463,616 | `c1b31d6ae7790b1b51426b7991c991d3a01bf33c63e0dcf3f113b7ad26c99e5e` | `4a31c236eac9fa8afc42d4d01e05ef6236f338e9` | DJ Lab **v1.0.6** — the fast defect wave (report: proof/DEFECT_WAVE_106_REPORT.md; spec: proof/REALRUN_103_DEMAND.md "REAL-RUN v1.0.4 INTAKE — D53…D56"). **D41 (escalated — reported twice by the user):** on the Statistics screen a rail playlist/folder click now changes the SCOPE and STAYS on Statistics (header names the scope, all cards re-render atomically) — the v1.0.1 live re-scope restored for this screen only; D34's yield-to-grid keeps applying to every other front screen and a modal still blocks. Proven by new phase AD (SAD0–SAD4: three in-place re-scopes + the Charts control) and by phases O/W re-cut to the ruling (the same precedent as v1.0.3's re-cut for D34). **D53 — Generate Cue Points is STRUCTURE-AWARE:** a new core engine (`generate_cues_structured`) decodes the track once and reads per-beat energies of the total signal + a low band; anchors — drop · breakdown (minimum 64 beats by default, Settings-configurable) · second drop · second breakdown · fade-out (low band only) — are detected by energy comparison, snapped to 16-beat phrases, and template cues (Start · 64 before the drop · Drop · Breakdown · Second drop · Second breakdown · 32 before the end · Fade out/End) are placed relative to the anchors under the track's real grid (ANLZ PQTZ wins, else the derived grid, labelled as such in the ledger). Skeleton rule: no drop found → ZERO placements and every entry skipped naming it; every other skip names its reason; a flat track gets no invented cues. Existing cues: generation refuses BY NAME on a cued track; the new Settings toggle "Replace DJ Lab's own generated cues" refreshes only DJ Lab's own `DJ Lab ·`-prefixed cues in the same spine batch — the user's own cues are never touched. Core proof cuegen_proof 6/6 (planted section maps at 120/128/140 BPM land exactly; flat → 8 named skips, 0 cues; min-breakdown and grid-offset cases); GUI phase AE proves the flat refusal-to-invent, the cued-track refusal, replace-own keeping the user's cue, exact planted milliseconds through the spine, the placed/skipped ledger, and Undo restoring the pre-generation state. **D54 — identify survives separator-less blob names:** the parser derives blob splits (comma artist lists, artist-suffix forms, '#'-prefixed titles) and, when the blob form's gates veto everything, the cascade retries the split forms (labelled parse-derived in the evidence) before declaring not-found; the retry's title gate accepts space-stripped equality (blob spacing is destroyed) while the ARTIST gate and the ≥2-lane agreement rule stay at full strength; D51 (duration ranks, never vetoes) and D46 (per-lane reporting + named rejections) untouched. Phase AD SAD5–SAD11: his exact fixture '#that Power will.i.am, Justin Bieber (intro) (clean).mp3' — the v1.0.4 veto reproduced under the blob parse, the wrong-artist candidate still rejected by name under the split parse, and the fixture identifies as "#thatPOWER" / will.i.am against stubbed lanes carrying the real recording. **D55 — preview-strip overflow: VERIFIED, NOT REPRODUCED, no code changed:** the playing row's mini-wave canvas measures inside its Preview cell (canvas 333..515 ⊆ cell 329..519 — exactly the CSS inset arithmetic) at natural layout and at 880/1900 px wrap widths (phase AE SAE9–SAE11); his shot's strip stops at the Title text and matches a full-cell mini-wave inside a wide Preview cell next to the D40-collapsed Title column; the full analysis + the sandbox-WebKit vs his WebView2 caveat are in the report. Release profile, version 1.0.6 in all five places + Release Notes head, **static CRT preserved** (PE census proof/pe_census_106.txt = exactly the same 17 system DLLs as v1.0.5, set-diff verified). Core battery **150/150 across 21 targets** (v1.0.5 baseline 144 + cuegen_proof's 6). Full GUI gate (gate_driver_106) green on the final source — every phase 0 failed, incl. AD 12/0 · AE 12/0 · AA 21/0 (SAA20 inside its line this run, no disposition needed) · CB 4/0 with its named live-check skips. Environment disclosures (sandbox WebKit stack rebuilt from the recorded 60-deb Azure set after a total loss + the glvnd EGL vendor registration restored; EDQUOT incidents answered by freeing regenerable build cache inside our own tree only; one 1920×1080 AE variant run stalled in the harness and is recorded as an environment limitation in the report, not as evidence) are in the report. Served bytes verified after upload (CDN LFS hash equals the local sha256). The repo's `djlab-app.exe` is NOT touched by this upload. |
+
+- Content files after this addition: **43**, plus this manifest = 44. Sum of file sizes: **9,937,805,187 bytes**.