QA Ticket Resolution Log #3 14 / 14 RESOLVED

The .gam data authors rotations in degrees — across all 35 levels the values are clean editor angles (0, 90, 170, 180, 220, 270). But gam_loader.c stored them raw into e->rx/ry/rz, which the player heading (sinf(e->ry)), the draw paths, and the camera all consume as radians. So 90° became 90 rad ≡ 116.6°, 270° became ≡ 350.1°, and 180° became ≡ 167.3° — close enough to "right" in some places that it survived three QA passes. A second, subtler half: the engine mirrors the original's left-handed Z at the load boundary (z = −z). This ticket initially treated the mirror term alone as the sign rule; ticket #4 later showed that the original's left-handed rotation convention adds the second negation, leaving current X/Y rotation un-negated.

Dumping the .gam instances settled it: every flagged pickup authors SpriteDatabase=sprites.omt, SpriteIndex=106 — and chunk 106 in sprites.omt is a canvas literally named "hidden" (the editor's ?-placeholder, below). These are invisible trigger volumes sitting on top of their visible referent — the hydrant mesh is level geometry, the level2 rescue cat is the separate 3KIT entity. The engine had been "helpfully" binding referent meshes per tag (drawing a second, untextured hydrant). The same dump showed pad/pad2 author 106 too (rows removed), while godphone authors chunk 184 — a real phone sprite — so it now renders as the original does.

Level1.gam 3PIC nest1   sprites.omt idx=106   ("hidden")
Level1.gam 3PIC nest2   sprites.omt idx=106   ("hidden")
Level1.gam 3PIC boatl   sprites.omt idx=106   ("hidden")
Level1.gam 3PIC hydrant sprites.omt idx=106   ("hidden")
Level1.gam 3PIC pad     sprites.omt idx=106   ("hidden")
Level1.gam 3PIC godphone sprites.omt idx=184  ("phone" — a real sprite)

The decomp settles the class's nature: C3DArrow's base chain is C3DSpriteType → C3DSprite — it's a billboard sprite, not a mesh class. All eight instances author sprites.omt chunk 33, which is the canvas named "arrow" (size 200). The engine's per-FourCC table bound 3Darrow.ASE — a leftover from before the sprite tier existed — which shadowed the authored sprite reference.

The reporter named the exact container, and the decompiled C3DPhoneBooth::InitObject agrees verbatim: it resolves "phone.omt" and binds shape 0 from it. phone.omt holds one 3DSh chunk — id 0, named "phone" (the booth, 482 units tall) — plus canvases Phone2 / phone / PhoneFloor. phone.ASE, what the engine had been drawing, is the handheld walkie-talkie prop.

// C3DPhoneBooth::vfunc_01_007 (InitObject), Ghidra @ 00435780
puVar1 = FUN_0046a910(s_phone_omt_004f0314);   // resolve "phone.omt"
uVar2  = FUN_00477ba0(this[0x12a].vftable, 0); // bind shape id 0 = the booth

A fun one: the "soda" is real. The instances author SpriteDatabase=icons.omt, SpriteIndex=4 — but icons.omt is the level editor's 12-icon set (trigger, dispatch, load, question…), and chunk 4 is the generic "sprite" placeholder icon, which happens to be drawn as a little green bottle. The original class swaps in its real visual at runtime; the engine was faithfully rendering the editor's stand-in. The actual leaves canvas lives in sprites.omt as chunk 45, "leave0000".

The Phase-5 tree sizes were measured per-tree from captured 4-vertex billboard drawcalls, matched to placements by position clustering. Re-auditing the table against the source meshes exposed the two flagged trees as mis-associations: every tree* placement shares the identical instanced trunk mesh, and the eleven siblings measure 450–700 units — except tree07 at 50, with a one-off 64×64 texture no other tree uses. Same story for treebranch04 (200 vs its family's 500–600, on a 652-unit-tall trunk). Those two drawcalls had been clustered onto the wrong quads.

tree01  450   tree06  700   tree07   50 ←  tree29 650   tree30 550   tree31/32 600
treebranch02 500   treebranch03 600   treebranch04 200 ←   treebranch07 500

The hardest item of the ticket — because every previous answer (including two from this ticket) was a texture for a mesh that should never draw. BLOCK_Rocket03 has no material links and bare full-canvas UVs because it is a collision volume: the original's BLOCK*-prefixed meshes are invisible blockers, each with a separate visible twin (BLOCKbench ↔ bench05, BLOCKING_road ↔ the road). Ticket #2 baked it white, this ticket first bound a stale captured tex_id, then a stripe tile — three guesses painting an object the original never shows, on top of the real rocket. sandmanfan settled it: "all Block meshes should be invisible" — the visible playground rocket, fins included, is Rocketa, which was already textured correctly (canvas "Rocket2", the red/white/blue striped sheet).

The QA report's own asset field gave it away: firedoor.ase. C3DSchoolDoor authors a per-instance ASEFile/PNGFile pair exactly like the swing door class the engine already handles — level1c says firedoor, level3 says doorretro (with exit.png / retrodoor.png variants), level2a says doorfowl. The engine's one-size-fits-all 3SCD → DOOR.glb row ignored all of it. One wrinkle: the install's firedoor.ASE is a degenerate 8-vertex export stub, but the textured glb/ase/firedoor.glb twin already existed from the door-family migration.

The sprite-pickup draw path "helpfully" lifted every billboard by half its size (y + size/2) on the assumption that authored pickup positions sit at ground level. The fishbowls disproved the assumption with numbers: they author Y=−472.8 on a floor at −500 — the authors placed the center, just like the original's OMediaCanvasElement places its canvas AT the element position. The lift pushed every sprite pickup (fishbowls +50, jetpack +75) half a quad too high.

Straight from the decomp: C3DCindy's asset registration loads cindy.png once and registers HISTOP → cindstop.ase as the default stop/idle animation. cindycheer.ASE is the cheering anim the class swaps to at the race finish — not the default pose for a captive on the Yokian ship.

| texture   | cindy.png                  | 00414fa0 | loaded once at registration |
| animation | HISTOP -> cindstop.ase     | 00414fa0 | default stop/idle           |