re(xbg7): characterise the residual 47 misses — 99.25% of resources decode

examples/why_missed.rs reports the furthest gate per undecoded resource. Disc-wide
that is 47 rows / 43 distinct names, and they are mostly not ship geometry:

  30  e_rou_/_rou_ pose & proxy composites (24-vertex marker boxes, extent 0.010)
   6  .DAT particle composites in ptc_pack
   8  damage/LOD variants (e101_bdy_02_d, e901_wing_05_*_m, ...)
   3  props/other (g005 extent 0.196, _rou_f001_wep_05, e_rou_e005)

By gate: extent 32, coverage 8, winding 7. The extent bucket is almost entirely the
pose-proxy boxes, and lowering that floor was measured and refuted earlier. The one
genuinely interesting residual is e901_wing_05_L/R at winding 0.587/0.570 against the
0.70 floor — the signature of a thin double-sided sheet, unproven without a capture
with the boss on screen (flying stage 16 puts the container in memory but the unit
never appeared).

So: 6 247 / 6 294 resources decode (99.25 %), and every real mesh drawn in three
captured missions decodes at the GPU's own offsets with index runs matching byte for
byte (93/93 and 128/128).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NptfmpjdpNCKEez6d2xvA9

docs/re/INDEX.md

@@ -19,7 +19,7 @@ Promote to a prose `structures/…md` file when a format needs behavioural notes
 | LSTA sprite list | ✅ | `sylpheed-formats/src/lsta.rs` | A display list of inline elements: **T8aD sprites and `PRMD` primitives**. The `count` at `0x04` is **exact and counts both** — `count == T8aD + PRMD` for **64/64** lists on the disc, which retires the old "a few entries disagree" note (it compared sprites against a total including primitives). **All 1 281 sprite frames decode** after the T8aD rectangle-list fix |
 | IXUD subtitle | 🟡/✅ | `sylpheed-formats/src/ixud.rs` + [movie link](movie-subtitle-link.md) | timed cues. **The movie↔subtitle↔voice link is solved — statically**, from the movie config record in `tables.pak` (schema `0x067025b9`), not from the running game as this row previously assumed: [101 movies mapped](captures/movie-subtitle-voice-map.csv), 94 with subtitles, 83 with voice, 21 with a telop overlay. 93 of 94 subtitle refs resolve in the language paks; **`SUBTITLE_S12B.tbl` is missing from all six languages** — a dangling reference on the disc. Naming is `SUBTITLE_<base>.tbl` / `VOICE_<base>` with six documented exceptions. The record's ~104 **script ids** are ❔ — positional pairing drifts by three because the IDXD pool dedupes repeated values |
 | Fonts (ttf/otf/ttc) | ✅ | `sylpheed-formats/src/font.rs` | standard OpenType, parsed via ttf-parser |
-| XBG7 mesh | ✅/🟡 | `sylpheed-formats/src/mesh.rs` + `tests/mesh_disc.rs` ([xbg7](structures/xbg7-mesh.md)) | **6 294 resources, 6 209 decode (98.7 %), 82 searched-and-missed** (2026-08-12, up from 5 480 / 87.1 %). Five evidence-driven fixes got there: **distinct anchor assignment** (no two resources may claim one buffer — proved by a capture showing the container holds both mirrored `e106` hull halves), the connectivity cap replaced by a **winding-consistency gate at 0.70**, **structural requirements on pre-pivot sub-meshes** (index range, then exact pool coverage), and **filtering after the assignment** so a subset query cannot differ from the full decode. Validated against a runtime capture that names the file offset of every buffer the engine drew: **46/46 drawn buffers claimed, 45 anchored exactly**. **No real mesh now decodes differently in different containers** — all 89 remaining cross-container disagreements are interchangeable 24-vertex bounding boxes, which no anchoring rule can pin (monotone order re-tested and refuted). Remaining misses attribute to the degeneracy/extent gate (42), winding (31) and coverage (9); the first was probed and its "obvious" fix refuted. Every decoded sub-mesh covers its own vertex pool. **The `[index buffer][vertex buffer]` layout is now runtime-verified** (2026-08-13): with the F10 capture extended to log each draw's index buffer, all **42** drawn `Stage_S02` buffers match our decoded index count exactly, all 42 have their index union cover the pool exactly, and the 30 single-block cases all sit at `pad ≤ 3` — so `e106_eng_02_l`'s old rejection was the connectivity gate, not a misplaced index buffer. The `indices=` mystery was the capture keeping only the **first of several index batches** per buffer. **And comparing index VALUES found the biggest silent defect yet**: the anchor took the first `pad` that validated, so a block whose index data sits at pad 2 was read **one element late** — 76/93 captured runs matched, all 17 differences a one-element shift. Scoring pads by degenerate triangles + winding fixes it: **93/93** captured runs now match byte for byte, disc-wide degenerate runs **582 → 1** (the grouped path had the same bug; and two resources were anchored on a degenerate lookalike earlier in file order), **590 of 8 850** sub-meshes re-wired with 10 vertex anchors moved, resources decoded unchanged at 6 209. Cross-container minority decodes 89 → 96 — *because* the decoder improved: `_rou_f402_dead` now has a majority (32×25×8) so its seven wrong copies are named instead of hidden. One dirty run remains, blocked by distinct assignment on a 24-vertex box |
+| XBG7 mesh | ✅/🟡 | `sylpheed-formats/src/mesh.rs` + `tests/mesh_disc.rs` ([xbg7](structures/xbg7-mesh.md)) | **6 294 resources, 6 209 decode (98.7 %), 82 searched-and-missed** (2026-08-12, up from 5 480 / 87.1 %). Five evidence-driven fixes got there: **distinct anchor assignment** (no two resources may claim one buffer — proved by a capture showing the container holds both mirrored `e106` hull halves), the connectivity cap replaced by a **winding-consistency gate at 0.70**, **structural requirements on pre-pivot sub-meshes** (index range, then exact pool coverage), and **filtering after the assignment** so a subset query cannot differ from the full decode. Validated against a runtime capture that names the file offset of every buffer the engine drew: **46/46 drawn buffers claimed, 45 anchored exactly**. **No real mesh now decodes differently in different containers** — all 89 remaining cross-container disagreements are interchangeable 24-vertex bounding boxes, which no anchoring rule can pin (monotone order re-tested and refuted). Remaining misses attribute to the degeneracy/extent gate (42), winding (31) and coverage (9); the first was probed and its "obvious" fix refuted. Every decoded sub-mesh covers its own vertex pool. **The `[index buffer][vertex buffer]` layout is now runtime-verified** (2026-08-13): with the F10 capture extended to log each draw's index buffer, all **42** drawn `Stage_S02` buffers match our decoded index count exactly, all 42 have their index union cover the pool exactly, and the 30 single-block cases all sit at `pad ≤ 3` — so `e106_eng_02_l`'s old rejection was the connectivity gate, not a misplaced index buffer. The `indices=` mystery was the capture keeping only the **first of several index batches** per buffer. **And comparing index VALUES found the biggest silent defect yet**: the anchor took the first `pad` that validated, so a block whose index data sits at pad 2 was read **one element late** — 76/93 captured runs matched, all 17 differences a one-element shift. Scoring pads by degenerate triangles + winding fixes it: **93/93** captured runs now match byte for byte, disc-wide degenerate runs **582 → 1** (the grouped path had the same bug; and two resources were anchored on a degenerate lookalike earlier in file order), **590 of 8 850** sub-meshes re-wired with 10 vertex anchors moved, resources decoded unchanged at 6 209. Cross-container minority decodes 89 → 96 — *because* the decoder improved: `_rou_f402_dead` now has a majority (32×25×8) so its seven wrong copies are named instead of hidden. One dirty run remains, blocked by distinct assignment on a 24-vertex box. **Then the descriptor gave up its last structural secret**: it declares a vertex layout **per sub-mesh** (`n201_01` → strides 24/24/24/**28**, capture-confirmed), and grouped selection must prefer the candidate explaining the **whole** pool rather than the first whose pivot validates — together they take never-decoding resources **85 → 47** (**6 247 / 6 294 = 99.25 %** decode), put `n201_01` on all four capture-proven offsets and raise the stage-05 capture oracle to **128/128**. The residual 47 is 30 pose/proxy composites (0.010-unit marker boxes), 6 `.DAT` particle composites, 8 damage/LOD variants and 3 props — not a threshold away |
 | Capital-ship part placement | ✅ | `sylpheed-formats/src/ship.rs` (static) + [runtime capture](ship-placement-runtime-capture.md) | Placement is **sound** (hull static-exact against the `e106` capture; cross-id mounting genuinely narrow, 2 pairs across 335 ships). The XBG7 mis-decode this row used to blame for "ships assemble wrong" — a shared turret ~100× too large in some containers — is **fixed** (2026-08-12, the exact-coverage requirement): `e303_wep_01` now decodes 49×23×42 everywhere and places at ±179 on the `e106` hull, and no real mesh disagrees across containers. A composite-node audit confirmed the assembler itself never applied a bad scale (all nodes scale 1.0, orthonormal). Still open: `static_assembly_matches_runtime_capture` walks capture parts only, so **extra** static placements cannot fail it |
 | Weapon fields defaulted on disc | ✅ | [runtime struct](structures/weapon-struct-runtime.md) · [DATA SHEET route](weapon-datasheet-runtime.md) | **Solved.** Canary maps guest RAM into `/dev/shm`, so the parsed `Weapon`/`Shell` objects are readable live; their layout is solved against disc ground truth (zero contradictions over 100+ records). All 126 weapons, exact numbers, no story progress needed — [4 393 values](captures/weapon-runtime-fields.csv) the disc does not carry. Supersedes the letter-bucket limit of the DATA SHEET route, which now serves as the independent cross-check |
 | Unit (craft/vessel) fields defaulted on disc | ✅/🟡 | [runtime struct](structures/unit-struct-runtime.md) | The parsed `unit\UN_*.tbl` definition object, vtable `0x820af844`, ≥`0x380` bytes, one per unit — **discovered, not assumed** (`unit_discover.py`), and distinguished from the spawned-entity class `0x820af030` by being one-per-ID and byte-constant within a run. Across runs only pointer words move — `--crosscheck` proves **no reported field offset is run-dependent** (two words, `+0x2c8`/`+0x2d0`, are stage-dependent and remain unidentified). 27 fields ✅ (21 units, 7 runs); the `Maneuver` block is **schema declaration order, 4 bytes/field, base `0x9c` with a two-slot gap after `AA_Roll_Min`** (29 anchors, 0 conflicts), which also pins 5 fields *no* disc record ever values. Angles are **radians at runtime, degrees on disc**. **Re-derived independently 2026-08-13 from the loader's own key strings** (`sub_82341A20`; the field name for each store is a string in the image): **159 fields**, agreeing with this solver on **25 of 25 shared offsets**, verified at **406 values matching the disc and 0 disagreeing** over 11 live objects spanning UNIT and VESSEL — landed as `data/unit_definition_layout.txt` + `sylpheed_formats::unit_layout` + a no-emulator test, with **121 defaulted fields** read out ([live-unit-definitions](live-unit-definitions.md)). Unlike weapons, unit definitions are instantiated **per stage**, so coverage (21/110) grows by visiting missions — but a defaulted field is **not** a global constant: `Size_Y` provably inherits `Size_X` (7 independent units, 6 distinct values), and three more sibling rules are recorded ❔, recovering 65 values in units never visited — [values](captures/unit-runtime-fields.csv) |