diff --git a/AGENTS.md b/AGENTS.md
index fc591ea5..f5fd70b1 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -4,7 +4,7 @@
 
 - `n64decomp/007`: WIP decompilation of GoldenEye 007 (N64), byte-matches US/EU/JP ROMs.
 - Active work: **PC port** modelled on the Perfect Dark PC port (same Rare "Indy" engine family).
-- **Read `docs/PCPortResearch.md` first** — architecture, GE-specific RSP deltas, phased plan, and review findings (A1–D32 in §11; current handoff + plan in §H).
+- **Read `docs/PCPortResearch.md` first** — architecture, GE-specific RSP deltas, phased plan, and review findings (A1–D33 in §11; current handoff + plan in §H).
 - **Starting a fresh session?** Read `docs/HANDOFF.md` — the paste-ready brief with the immediate task, standing procedure, and environment gotchas.
 
 ## Non-negotiables
@@ -47,7 +47,7 @@ Run `/linkcheck` for this sweep. Record new findings in `docs/PCPortResearch.md`
 - **Phase 0 (scaffolding):** done, committed.
 - **Compile+link milestone:** all ~235 TUs compile and `ge007.x86_64.exe` links clean (`ninja ge007 -k 0`, 236/236). Findings D7–D17 in `docs/PCPortResearch.md` §11; asset stubs in `port/src/assetstubs.c`, fast3d no-ops in `port/src/gfxstub.c` (delete when real fast3d lands).
 - **Phase 1 (boot to window):** done — ROM loads/validates/maps at cart base 0x10000000 (`romdata.c`), SDL2 window + GL clear loop (`video.c`), absolute asset symbols from `scripts/gen_romassets.py` → `port/src/romassets_<r>.s` (findings D18–D23).
-- **Phase 1.5 (boot to first frame):** IN PROGRESS — mainThread runs through `bossInitMainthreadData()`; D31 (langInit file-load SIGSEGV) fixed via real-zlib decompression (`port/src/rzdecomp.c`, replaces the excluded `src/game/decompress.c`+`zlib.c`). Current blocker is **D32** (ROM-struct pointer-width): Part A (u32 layout in `ModelAnimation`) done, Part B (load-time rebase in `expand_ani_table_entries`) open. Recurring sub-task: ABI/layout fixes per the D32 procedure (§H) as more asset loading is reached.
+- **Phase 1.5 (boot to first frame):** IN PROGRESS — mainThread runs through `bossInitMainthreadData()`; D31 (langInit file-load SIGSEGV) fixed via real-zlib decompression (`port/src/rzdecomp.c`, replaces the excluded `src/game/decompress.c`+`zlib.c`). Current blocker is **D32** (ROM-struct pointer-width): Part A (u32 layout in `ModelAnimation`) done; Part B **root-caused (D33)** — the ROM stores structured fields big-endian (per-field load-time fixup needed) + an x86-64 loop-stride bug in `expand_ani_table_entries`; fix designed, implementation next. Recurring sub-task: ABI/layout fixes per the D32 procedure (§H) as more asset loading is reached.
 - **Phase 2 (rendering):** bring in PD fast3d (replaces `port/src/gfxstub.c`), verify custom CC/RM modes against `gmain.s`, port the scheduler; replace asset stubs with ROM-backed data (D16). Assumes ABI-clean data loading from Phase 1.5.
 - **Phase 3 (audio + input):** libaudio → SDL; decide the ASP strategy (C2: `aspMain*` dummies in `ucode.c`).
 - **Phase 4 (saves + polish):** file-backed EEPROM; PFS/motor already no-op stubs.
diff --git a/docs/HANDOFF.md b/docs/HANDOFF.md
index c8366762..5d5bbaba 100644
--- a/docs/HANDOFF.md
+++ b/docs/HANDOFF.md
@@ -1,7 +1,7 @@
 # Handoff brief — GoldenEye 007 PC port (Phase 1.5: boot to first frame)
 
 _Paste-ready briefing for the next agent session. The authoritative detail lives in
-`AGENTS.md` and `docs/PCPortResearch.md` (§H handoff, §F/D31–D32 findings); this is
+`AGENTS.md` and `docs/PCPortResearch.md` (§H handoff, §F/D31–D33 findings); this is
 the summary + the immediate task._
 
 ## Your job
@@ -13,61 +13,99 @@ when the user is back: a booting game.
 
 ## Read first (authoritative, in order)
 1. `AGENTS.md` — non-negotiables, critical files, build, verification ritual, phase status.
-2. `docs/PCPortResearch.md` §H (handoff + plan), then §F/D31–D32 (findings).
+2. `docs/PCPortResearch.md` §H (handoff + plan), then **§F/D33** (root cause of the
+   current blocker + complete fix design), D31–D32 for background.
 
 ## Non-negotiables (the ones that bite)
 - N64 build files (`Makefile`, `tools/`, `rsp/`, `ld/`) — never touch.
 - **Game logic is unmodified.** Hardware deps go in `port/`. Narrow exception: mechanical,
-  semantics-preserving ABI/layout fixes forced by the 32→64-bit transition (embedded
-  pointers → `u32` + cast at use), each documented as a D3x finding. No behavior changes.
+  semantics-preserving ABI/layout fixes forced by the 32→64-bit transition, each documented
+  as a D3x finding. No behavior changes.
 - Region macros in `CMakeLists.txt` must mirror the N64 Makefile exactly.
 
-## Current state (verified)
+## Current state (verified this session)
 - **D31 FIXED:** langInit file-load crash gone (real-zlib `decompressdata` in
-  `port/src/rzdecomp.c`; excluded `src/game/decompress.c`+`zlib.c`). mainThread boots past
-  all 7 language files into `bossInitMainthreadData()`.
-- **D32 Part A DONE:** `ModelAnimation` pointer fields → `u32` (N64 layout, sizeof 64),
-  casts in `model.c`.
-- **D32 Part B = the current blocker.** Boot dies at `boss.c:233`
-  `init_weapon_animation_groups_maybe()` → … → `modelAnimReadRootMotionValue`
-  (`model.c:914`) reading `desc->bitCount`, because animation entries' `bitDescriptors` /
-  `bitStream` are not valid DRAM pointers after load.
+  `port/src/rzdecomp.c`). mainThread boots into `bossInitMainthreadData()`.
+- **D32 Part A DONE:** `ModelAnimation` pointer fields → `u32` (N64 layout, sizeof 64).
+- **D32 Part B = ROOT-CAUSED (finding D33). Two independent bugs, both confirmed:**
+  1. **ROM endianness.** The `.z64` file stores structured multi-byte fields in
+     **big-endian byte order** (bit-packed streams are raw). The animation blob's
+     20-byte record headers and descriptor arrays must be converted per-field at load
+     time: u32 → bswap32, u16 → bswap16, u8 → identity. No uniform word transform exists.
+     Validated in Python over all 173 ROM entries: 0 bad ranges, 0 non-sequential
+     descriptor arrays; simulated root-motion frames show smooth per-frame deltas.
+     The exact garbage seen under gdb (0xC82BD8C0) is reproduced by LE-reading the BE
+     file bytes + rebase. Evidence: GE-specific ROM header, `tools/utils.h` "v64→z64"
+     convention, build pipeline with no byte-swap step, BE toolchain defaults.
+  2. **x86-64 stride bug.** `expand_ani_table_entries()` iterates with `s32** var_v0;
+     var_v0++` — +4 bytes/iter on N64, but +8 on x86-64 (verified: `add $0x8,%rdx` in the
+     compiled binary). Only even-indexed entries get rebased; odd ones keep raw small
+     offsets (fire_standing is index 1 → would stay 0x144 → SIGSEGV on use). The earlier
+     "zeros at odd indices" gdb observation was a display artifact — the compiled `.data`
+     array is dense (objdump-verified). This pattern occurs nowhere else in the set.
 
-## Immediate task — D32 Part B
-Make `expand_ani_table_entries()` (`src/game/initanitable.c:233`) rebase **every** entry's
-`bitDescriptors` (u32 @0x08) and `bitStream` (u32 @0x10) to a valid V1 address (< 0x80000000).
+## Immediate task — implement the D33 fix (design complete, ~3 small edits)
+1. **`port/src/romdata.c` + `port/include/romdata.h`:** add
+   `void romdataFixupAnimationData(u8 *blob, u32 blobSize, const s32 *tableA, const s32 *tableB)`.
+   For each non-null (≠0 and ≠1 — sentinel is **1**) offset in both tables:
+   - record at `blob+off` (20 bytes): bswap32 fields at +0x00 (address), +0x08
+     (bitDescriptors), +0x10 (bitStream); bswap16 fields at +0x04 (frame count),
+     +0x0C (bitsPerFrame root motion), +0x0E (frame size bits); bytes +0x06/+0x07
+     (u8 angle width / loop flag) untouched.
+   - descriptor array at `[blob+bd, blob+bs)` step 6 (`ModelAnimBitField`): bswap16
+     words at +0 and +4 (bitOffset, valueOffset); bytes +2/+3 (bitCount, pad) untouched.
+     Guards: skip if bd==0 && bs==0; require `bd < bs <= blobSize` and `(bs-bd)%6==0`.
+   Full field semantics + layout rules in D33 (blob interleave: entry i's payload sits
+   at [PTR_ANIM_{i-1}+0x14, PTR_ANIM_i); all regions disjoint).
+2. **`src/game/initanitable.c`, `alloc_load_expand_ani_table()` (~line 265):** call the
+   fixup **between** `romCopy(...)` and `expand_ani_table_entries(...)` — it must run
+   *before* expand (expand reads/writes those fields as LE after fixup). Add
+   `#include "romdata.h"` (`port/include` is already on the include path). One-line
+   mechanical ABI edit — document with a D32/D33 comment.
+3. **`src/game/initanitable.c`, `expand_ani_table_entries()` (line 233):** change the loop
+   pointer from `s32** var_v0` to `s32 *var_v0` (cast at assignment: `var_v0 = (s32 *)arg0;`)
+   so every entry is visited on x86-64. Semantics-preserving on N64. Keep the rest of the
+   function body identical (`*var_v0 += base`, `((struct anim_entry *)(s32)*var_v0)->unk08/unk10`,
+   and loop 2's `*(s32 *)(s32)*var_v0 += entries_base`).
 
-Gdb method (launch mode only — attach fails with error 87):
+**Do NOT transform the entries segment** (per-frame joint angles, file 0x124AC0,
+0x169EC0 bytes) — it is identity-encoded bit-packed data, not consumed at boot; verify
+visually in Phase 2 when Bond first animates.
+
+## Verify (gdb, launch mode only — attach fails with error 87)
 - Break at `boss.c:233` (right after `alloc_load_expand_ani_table` returns). base =
   `&ptr_animation_table->data`.
-- For each `animation_table_ptrs1[i]` / `animation_table_ptrs2[i]`, check
-  `*(u32*)(entry+8)` and `*(u32*)(entry+16)`.
-- Two anomalies to resolve first: (a) in-memory `animation_table_ptrs1[]` reads
-  `[ptr, 0, ptr, 0, …]` though the source is a dense `PTR_ANIM_*` list — reconcile that
-  against the `while (*var_v0 != 0)` fixup loop; (b) entry[0]'s `bitStream` high byte is
-  0x88, not 0x70. Likely a 32/64-bit or write-overlap issue in the two fixup loops.
-- Prefer understanding *why it works on N64* and matching that. If a game-code edit is
-  needed it's an ABI/layout change (allowed per non-neg #2) — document it as D3x.
+- For a sample of `animation_table_ptrs1/2[i]` entries (include odd indices!):
+  `*(u32*)(entry+8)` and `*(u32*)(entry+16)` must be in [0x70xxxxxx] (base + blob offset),
+  and the record's address field (+0x00) must be a cart address 0x10xxxxxx.
+- Confirm boot advances past `init_weapon_animation_groups_maybe()` into the next init
+  step of `bossInitMainthreadData`. Expect more D32-class faults further along (models,
+  textures, other tables) — apply the standing procedure below each time.
 
 ## Standing procedure (you will hit more of these)
 Every ROM-serialized struct with a pointer field faults the same way once you reach more
-asset loading (models, textures, other tables). The D32 fix procedure (doc §H): at the
-fault run `ptype /o <Struct>`; if a pointer field's offset/size diverges from its N64
-offset comment, change it to `u32` + cast at the use sites; verify the load-time rebase
-yields valid V1 addresses; rebuild; confirm the boot advances past this struct. Log each as
-D3x in §F and note it in AGENTS.md phase status.
+asset loading. The D32 fix procedure (doc §H): at the fault run `ptype /o <Struct>`; if a
+pointer field's offset/size diverges from its N64 offset comment, change it to `u32` + cast
+at the use sites; verify the load-time rebase yields valid V1 addresses; **also check the
+ROM bytes' endianness per-field (D33 rule: u32 bswap32 / u16 bswap16 / u8 identity) and any
+fixup loop's pointer stride on x86-64**; rebuild; confirm boot advances. Log each as D3x in
+§F and note it in AGENTS.md phase status.
 
 ## Environment (do not rediscover these)
 - MSYS2/MinGW tools are in `/c/msys64/mingw64/bin/` — **not** on PATH. Prefix every shell:
   `export PATH="/c/msys64/mingw64/bin:$PATH"`.
 - Build: `./build-pc.sh ntsc-final` (incremental; full rebuild only when `port/shim/`
-  changes). ROM is at `data/ge007.ntsc-final.z64`.
+  changes). ROM is at `data/ge007.ntsc-final.z64` (= `baserom.u.z64`, byte-identical).
 - gdb: **launch** mode only (`gdb -batch -ex "handle SIGSEGV stop" -ex run …`). Symbolicate
   offline with `addr2line -e build-pc/ge007.x86_64.exe -f -C <0x140000000+rel>`. Image base
   is `0x140000000` (re-verify with `info address` after a rebuild).
+- `objdump -d --disassemble=<fn>` / `objdump -s -j .data` on build-pc/ge007.x86_64.exe is a
+  fast way to check what the compiler actually emitted (used this session to confirm the
+  stride bug and the dense table array).
 - Many init functions use a fake RBP — compute stack offsets from the **entry RSP**, not RBP.
-- The D30 crash handler did **not** write `ge007.crash.log` for the game-thread fault this
-  session; attribute crashes via gdb. (Worth a separate look later.)
+- The D30 crash handler did **not** write `ge007.crash.log` for the game-thread fault;
+  attribute crashes via gdb. (Worth a separate look later.)
+- Python on this box: no f-strings with nested quotes (< 3.12) — use `%` formatting.
 
 ## Definition of done (this milestone)
 mainThread reaches `bossMainloop()` and a frame is actually drawn on the GL surface (not just
diff --git a/docs/PCPortResearch.md b/docs/PCPortResearch.md
index 17d96b26..0ae2698a 100644
--- a/docs/PCPortResearch.md
+++ b/docs/PCPortResearch.md
@@ -962,26 +962,96 @@ gameInit — and currently dies inside `langInit()` (D31).
   matches. `expand_ani_table_entries()` writes s32 to 0x08/0x10 via its own
   `struct anim_entry`, which now aligns exactly (no change needed there).
 
-  **Part B — pointer-rebase correctness: OPEN (next target).** After Part A the
-  *values* are still invalid for the crashing weapon-anim entry (gdb, break at
-  boss.c:233 after `alloc_load_expand_ani_table`): anim = 0x702adad4 = base
-  0x702ad8c0 + 0x214; `bitDescriptors` = 0xf714b067, `bitStream` = 0x000c9100 —
-  neither a V1 address. Two concrete anomalies to debug in
-  `expand_ani_table_entries()` (initanitable.c:233):
-  * `animation_table_ptrs1[]` is a **dense** array of `PTR_ANIM_*` offsets in
-    source, but in memory reads `[0x702ad8dc, 0, 0x702adad4, 0, …]` (zeros at
-    odd indices). The fixup loop `while (*var_v0 != 0)` would stop at the first
-    zero — yet even-index entries show rebased values. Reconcile the array's true
-    post-fixup shape and confirm every entry's unk08/unk10 is actually rebased
-    (the crashing anim is `animation_table_ptrs1[2]`).
-  * Even entry[0]'s `bitStream` comes out wrong (0x882ad8c0 — low 24 bits match
-    base, high byte 0x88≠0x70) while its `bitDescriptors` is correct
-    (0x702ad8c0). Inspect the second fixup loop (`**var_v0 +=
-    &_animation_entriesSegmentRomStart`) for a 32/64-bit or overlap bug.
+  **Part B — pointer-rebase correctness: ROOT-CAUSED (see D33), fix designed,
+  implementation pending.** The garbage values are *not* a bug in
+  `expand_ani_table_entries`'s arithmetic — the rebase is correct N64 code fed
+  wrong bytes, plus one genuine x86-64 ABI stride bug. Both root causes and the
+  full fix design are in **D33** below.
 
-  The crash site is unchanged (still model.c:914) — Part A is a necessary
-  prerequisite, not a regression. Next session: gdb `expand_ani_table_entries`
-  step-by-step until every entry's bitDescriptors/bitStream is a valid V1 address.
+* **D33 — ROM file stores structured fields big-endian; animation load needs a per-field endianness fixup + an x86-64 loop-stride fix (root cause of D32 Part B).**
+
+  **Discovery.** The `.z64` ROM file is not a plain image: multi-byte
+  *structured* fields (record headers, tables) are stored in **big-endian byte
+  order**, while bit-packed data streams are raw. Evidence chain:
+  1. `baserom.u.z64` is byte-for-byte identical to `data/ge007.ntsc-final.z64`;
+     the ROM header is GE-specific (magic `80 37 12 40` from `src/rom_header.s`,
+     BE u32 CRC at 0x10/0x14 written by `tools/n64cksum.c`) — not the standard
+     N64 header layout.
+  2. Build pipeline: `$(LD)` → ELF → `objcopy -O binary --gap-fill=0xff` →
+     RareZip cdata compression (`data_compress.sh`) → `n64cksum`. No byte-swap
+     step; the mips64-elf-gcc toolchain (SGI-style BE default) emits BE bytes.
+  3. `tools/utils.h` documents the convention: `swap_bytes()` = "convert from
+     v64 to z64 ordering", `reverse_endian()` = "convert from n64 to z64
+     ordering" — the project's ".z64" is a word-swapped cart image (community
+     ".n64" format). Real cart = bswap32(file).
+  4. Animation-blob analysis: reading headers as BE u32/u16 yields self-
+     consistent values (frame counts, bit widths 0–31, offsets that tile the
+     blob exactly); LE reads yield garbage.
+
+  **Per-field transform rule** (applied at load time to the `animation_data`
+  segment): u32 fields → bswap32; u16 fields → bswap16; u8 fields → identity;
+  bit-packed streams → identity. No single uniform word transform exists (mixed
+  field widths).
+
+  **Blob layout** (fully mapped; all 173 entries verified): each C array
+  (`animation_table_ptrs1/2` in `src/game/initanitable.c`) is a sequence of
+  20-byte records at the `PTR_ANIM_*` offsets. Record fields: +0x00 address
+  (entries-segment offset, u32), +0x04 frame count (u16), +0x06 angle bit width
+  (u8, used by `modelAnimReadBitsAsU16Angle`), +0x07 loop flag (u8), +0x08
+  bitDescriptors blob offset (u32), +0x0C bitsPerFrame root motion (u16, used by
+  `sub_GAME_7F06D2E4`: `scaled = unk0C * frame`), +0x0E frame size in bits
+  (u16, used by `loadAnimationFrame`: `frameSize = unk0E >> 3`), +0x10
+  bitStream blob offset (u32). Tail (+0x14..+0x3C) is unused by any code — the
+  effective record size is 20 bytes (= `struct anim_entry`, 5×s32). Interleave
+  rule: entry *i*'s payload [descriptors][stream] sits at
+  [PTR_ANIM_{i-1}+0x14, PTR_ANIM_i); all regions are disjoint; stream size =
+  ceil(frameSizeBits × frames / 8) — exact fit on every entry (fire_standing:
+  544 bits × 106 frames → 252 bytes). Descriptor array = `ModelAnimBitField`
+  {u16 bitOffset, u8 bitCount, u8 pad, u16 valueOffset} ×4 (6 bytes each); after
+  the transform all 692 descriptor arrays pack bitOffsets sequentially from 0.
+  Null entries use sentinel **1** (not 0) — `expand_ani_table_entries` skips
+  `*var_v0 == 1`.
+
+  **Validation.** Simulated the full transform in Python over all 173 entries:
+  0 bad descriptor ranges, 0 non-sequential arrays. Root-motion simulation for
+  fire_standing frames 0–3 (widths 6/7/6/0, bitsPerFrame=19) gives smooth
+  per-frame deltas (x = 7, 10, 12, 11; y ≈ 1086–1088) — proving the stream
+  bytes are identity-correct (no transform). The previously observed garbage is
+  reproduced exactly: file bytes `[00 00 01 58]` at fire_standing_fast's bd →
+  LE read 0x58010000 + base 0x702ad8c0 = **0xC82BD8C0**, the exact value seen
+  under gdb. The "zeros at odd indices" observation was a gdb display artifact —
+  the compiled `.data` array is dense (verified by objdump of
+  build-pc/ge007.x86_64.exe).
+
+  **Second root cause: x86-64 stride bug in `expand_ani_table_entries`.** The
+  loops iterate with `s32** var_v0; var_v0++` — +4 bytes/iter on N64, but
+  `s32*` is 8 bytes on x86-64. Verified in the compiled binary (`add $0x8,%rdx`)
+  : only even-indexed entries are rebased; odd entries keep raw small offsets
+  (e.g. fire_standing, index 1, would stay 0x144 → instant SIGSEGV on use).
+  This pattern occurs nowhere else in the compiled set.
+
+  **Fix design (implementation pending — this is the immediate task):**
+  1. New port function `romdataFixupAnimationData(u8 *blob, u32 blobSize,
+     const s32 *tableA, const s32 *tableB)` in `port/src/romdata.c` (+ decl in
+     `port/include/romdata.h`): for each non-null (≠0, ≠1) offset in both
+     tables — bswap the 20-byte record header per-field (u32: +0x00/+0x08/
+     +0x10; u16: +0x04/+0x0C/+0x0E; u8: +0x06/+0x07 untouched); then transform
+     the descriptor array at [bd, bs) step 6 (bswap16 on words 0 and 4 only).
+     Guards: skip if bd==0 && bs==0; require bd < bs ≤ blobSize and
+     (bs−bd) % 6 == 0.
+  2. Call it in `alloc_load_expand_ani_table()` (initanitable.c ~line 265)
+     **between** the `romCopy` and `expand_ani_table_entries` — one line +
+     `#include "romdata.h"` (`port/include` is already on the include path).
+     Mechanical ABI edit per non-negotiable #2; must run *before* expand
+     (expand reads/writes the fields as LE after fixup).
+  3. Fix the stride in `expand_ani_table_entries()` (initanitable.c:233):
+     iterate with `s32 *` instead of `s32**` — mechanical ABI fix, semantics-
+     preserving on N64 where it was already correct.
+
+  **Open item.** The entries segment (per-frame joint angles; file offset
+  0x124AC0, 0x169EC0 bytes) is left as identity for Phase 1.5 — not consumed at
+  boot (only during rendering). Verify visually in Phase 2 when Bond first
+  animates.
 
 ### G. Phase 2 status (current)
 
@@ -1028,9 +1098,14 @@ shim expansion in compiled code.
 
 **State.** D31 (langInit file-load SIGSEGV) is fixed & verified — mainThread
 boots past all language-file loading into `bossInitMainthreadData()`. The boot
-now dies in `init_weapon_animation_groups_maybe()` on the **D32** ROM-struct
-pointer-width issue. D32 Part A (struct layout) is done; Part B (load-time
-rebase correctness) is the next target. No frames rendered yet.
+dies in `init_weapon_animation_groups_maybe()` on the **D32** ROM-struct issue.
+D32 Part A (struct layout) is done. **D32 Part B is now fully root-caused
+(D33):** (1) the ROM file stores the animation segment's structured fields in
+big-endian byte order — a per-field endianness fixup is required at load time;
+(2) `expand_ani_table_entries` has an x86-64 loop-stride bug (`s32**` advances
+8 bytes/iter, rebasing only even-indexed entries). The fix design is complete
+and validated against all 173 ROM entries; **implementation is the immediate
+task** (three small edits — see D33 "Fix design"). No frames rendered yet.
 
 **D32 repeatable fix procedure** (apply to any ROM-serialized struct that faults
 on a pointer-field read):
@@ -1043,18 +1118,23 @@ on a pointer-field read):
    into those u32 fields; if not, debug the fixup (see Part B below).
 4. Rebuild and confirm the boot advances past this struct to the next init step.
 
-**Next concrete target — D32 Part B.** Make `expand_ani_table_entries()`
-(initanitable.c:233) rebase *every* entry's `bitDescriptors`/`bitStream` to a
-valid V1 address. Gdb plan: break at boss.c:233 (after `alloc_load_expand_ani_table`),
-step through the two fixup loops, and for each `animation_table_ptrs1[i]` /
-`animation_table_ptrs2[i]` entry verify `*(u32*)(entry+8)` and `*(u32*)(entry+16)`
-are in [0x70xxxxxx]. Resolve these two anomalies first:
-* In-memory `animation_table_ptrs1[]` reads `[ptr, 0, ptr, 0, …]` (zeros at odd
-  indices) though the source is a dense `PTR_ANIM_*` list — reconcile the array's
-  true shape vs the `while (*var_v0 != 0)` loop.
-* entry[0]'s `bitStream` = 0x882ad8c0 (high byte 0x88≠0x70) while its
-  `bitDescriptors` = 0x702ad8c0 is correct — inspect the second fixup loop
-  (`**var_v0 += &_animation_entriesSegmentRomStart`) for a 32/64-bit or overlap bug.
+**Next concrete target — implement the D33 fix.** Three edits (full design in
+D33):
+1. `port/src/romdata.c` + `port/include/romdata.h`: add
+   `romdataFixupAnimationData(u8 *blob, u32 blobSize, const s32 *tableA,
+   const s32 *tableB)` — per-field BE→LE transform of the 20-byte records and
+   the [bd, bs) descriptor arrays (u32 → bswap32, u16 → bswap16, u8 identity).
+2. `src/game/initanitable.c` `alloc_load_expand_ani_table()`: call the fixup
+   between `romCopy` and `expand_ani_table_entries` (+ `#include "romdata.h"`).
+3. `src/game/initanitable.c` `expand_ani_table_entries()`: change the loop
+   pointer from `s32**` to `s32 *` so every entry is visited on x86-64.
+
+Then: build (`./build-pc.sh ntsc-final`), verify under gdb (launch mode):
+break at boss.c:233 after `alloc_load_expand_ani_table`; for a sample of
+`animation_table_ptrs1/2[i]` entries confirm `*(u32*)(entry+8)` and
+`*(u32*)(entry+16)` are in [0x70xxxxxx] (blob base + offset) and the record's
+address field is a cart address (0x10xxxxxx); confirm boot advances past
+`init_weapon_animation_groups_maybe`. Commit at the milestone.
 
 **Non-negotiable #2 refinement (applied to AGENTS.md).** The original "game code
 compiles unmodified / fix belongs in port/" is too absolute: pointer-width layout
diff --git a/tools_pc/disasm.py b/tools_pc/disasm.py
new file mode 100644
index 00000000..f9e5e307
--- /dev/null
+++ b/tools_pc/disasm.py
@@ -0,0 +1,130 @@
+#!/usr/bin/env python3
+"""Minimal MIPS disassembler for the BE-encoded GE ROM.
+
+Usage: disasm.py <ram_addr_hex> [n_instructions]
+Maps game-segment RAM (base 0x7F000000) to file offset 0x34B28.
+Also handles lib segment (RAM base = _startSegmentEnd - 0x10000000, ROM 0x1050).
+"""
+import struct, sys
+
+ROM = open('data/ge007.ntsc-final.z64', 'rb').read()
+GAME_RAM_BASE = 0x7F000000
+GAME_ROM_OFF = 0x34B28
+LIB_ROM_OFF = 0x1050
+LIB_RAM_BASE = 0x70000450  # from boot.s comment: code at ROM 0x1050 -> RAM 0x70000450
+
+REGS = ["$zero","$at","$v0","$v1","$a0","$a1","$a2","$a3",
+        "$t0","$t1","$t2","$t3","$t4","$t5","$t6","$t7",
+        "$s0","$s1","$s2","$s3","$s4","$s5","$s6","$s7",
+        "$t8","$t9","$t10","$t11","$t12","$t13","$gp","$sp",
+        "$fp","$ra"]
+
+RTYPE = {0x00:"sll",0x02:"srl",0x04:"sra",0x06:"sllv",0x08:"srav",0x0A:"sralv",
+         0x0C:"and",0x0E:"or",0x0F:"xor",0x10:"nor",0x12:"slt",0x14:"sltu",
+         0x18:"add",0x1A:"addu",0x1C:"sub",0x1E:"subu",
+         0x20:"mult",0x21:"multu",0x22:"div",0x23:"divu",
+         0x24:"madd",0x25:"maddu",0x26:"msub",0x27:"msubu"}
+
+COP0 = {0x00:"mfc0",0x01:"mfc1",0x02:"mfc2",0x10:"mtc0",0x11:"mtc1",0x12:"mtc2",
+        0x40:"bc0t",0x44:"bc0f"}
+
+def disasm_word(w):
+    op = w >> 26
+    rs = (w >> 21) & 31
+    rt = (w >> 16) & 31
+    rd = (w >> 11) & 31
+    sa = (w >> 6) & 31
+    fn = w & 63
+    imm = w & 0xFFFF
+    def s16(v): return v - 0x10000 if v >= 0x8000 else v
+
+    if op == 0x00:
+        if fn in RTYPE:
+            if fn in (0x00,0x02,0x04,0x06,0x08,0x0A):
+                return "%s %s,%s,%d" % (RTYPE[fn], REGS[rd], REGS[rt], sa)
+            if fn in (0x20,0x21,0x22,0x23,0x24,0x25,0x26,0x27):
+                return "%s %s,%s" % (RTYPE[fn], REGS[rs], REGS[rt])
+            return "%s %s,%s,%s" % (RTYPE[fn], REGS[rd], REGS[rs], REGS[rt])
+        if fn == 0x19:  # tgei etc rare
+            return "tgeiu? %s,%s,0x%04x" % (REGS[rd], REGS[rs], imm)
+        return ".word 0x%08x" % w
+    if op == 0x01:
+        c = (w >> 21) & 3
+        if c == 0:
+            if fn in COP0:
+                if fn in (0x00,0x10):
+                    return "%s %s,0x%02x" % (COP0[fn], REGS[rt], rs)
+                return "%s %s,0x%02x" % (COP0[fn], REGS[rs], rt)
+            if fn == 0x00: return "mfc0? %s" % REGS[rt]
+            return ".word 0x%08x" % w
+        if c == 1:  # cop1
+            if fn in (0x00,0x01,0x02,0x03):
+                names = ["mfc1","cfc1","mtc1","ctc1"]
+                return "%s %s,%s" % (names[fn], REGS[rt], REGS[rs])
+            if fn == 0x46: return "bc1t +0x%04x" % s16(imm)
+            if fn == 0x47: return "bc1f +0x%04x" % s16(imm)
+            if fn == 0x4E: return "bc1t? (cond)"
+            if fn == 0x4F: return "bc1f? (cond)"
+            # fp ops
+            fops = {0x10:"abs.s",0x11:"add.s",0x12:"div.s",0x13:"cmp.s",0x14:"cvt.s.w",
+                    0x15:"cvt.w.s",0x16:"max.s",0x17:"min.s",0x18:"mov.s",0x19:"neg.s",
+                    0x1A:"recip.s",0x1B:"rsq.s",0x1C:"sqrt.s",0x20:"abs.d",0x21:"add.d",
+                    0x22:"div.d",0x23:"cmp.d",0x26:"max.d",0x27:"min.d",0x28:"mov.d",
+                    0x29:"neg.d",0x41:"and.b",0x42:"or.b",0x45:"cvt.s.w"}
+            if fn in fops:
+                return "%s %f%d,%f%d,%f%d" % (fops[fn], rd, rs, rt, 0)
+            return ".word 0x%08x" % w
+        if c == 2:
+            return "cop2? .word 0x%08x" % w
+        return ".word 0x%08x" % w
+    if op == 0x02: return "jalr %s%s" % (REGS[rs], ", "+REGS[rt] if rt else "")
+    if op == 0x03: return "syscall"
+    if op == 0x04: return "beq %s,%s,+0x%04x" % (REGS[rs], REGS[rt], s16(imm))
+    if op == 0x05: return "bne %s,%s,+0x%04x" % (REGS[rs], REGS[rt], s16(imm))
+    if op == 0x06: return "blez %s,+0x%04x" % (REGS[rs], s16(imm))
+    if op == 0x07: return "bgtz %s,+0x%04x" % (REGS[rs], s16(imm))
+    if op == 0x08: return "j 0x%08x" % ((w & 0x3FFFFFF) << 2)
+    if op == 0x09: return "jal 0x%08x" % ((w & 0x3FFFFFF) << 2)
+    if op == 0x0A: return "addi %s,%s,0x%04x" % (REGS[rt], REGS[rs], s16(imm))
+    if op == 0x0B: return "addiu %s,%s,0x%04x" % (REGS[rt], REGS[rs], s16(imm))
+    if op == 0x0C: return "andi %s,%s,0x%04x" % (REGS[rt], REGS[rs], imm)
+    if op == 0x0D: return "slti %s,%s,0x%04x" % (REGS[rt], REGS[rs], s16(imm))
+    if op == 0x0E: return "sltiu %s,%s,0x%04x" % (REGS[rt], REGS[rs], imm)
+    if op == 0x0F: return "ori %s,%s,0x%04x" % (REGS[rt], REGS[rs], imm)
+    if op == 0x10: return "xori %s,%s,0x%04x" % (REGS[rt], REGS[rs], imm)
+    if op == 0x11: return "lui %s,0x%04x" % (REGS[rt], imm)
+    if op in range(0x20, 0x3C):
+        name = {0x20:"lb",0x21:"lhu",0x22:"lw",0x23:"lbu",0x24:"lwl",0x25:"lwr",
+                0x28:"sb",0x29:"sh",0x2A:"sw",0x2B:"swl",0x2D:"swr",0x2F:"ld"}.get(op, ".word")
+        if op in (0x28,0x29,0x2A):
+            return "%s 0x%04x(%s)" % (name, s16(imm), REGS[rs])
+        if op == 0x2F:
+            return "ld %s,0x%04x(%s)" % (REGS[rt], s16(imm), REGS[rs])
+        return "%s %s,0x%04x(%s)" % (name, REGS[rt], s16(imm), REGS[rs])
+    if op == 0x3C: return "lui %s,0x%04x" % (REGS[rt], imm)
+    if op == 0x3D: return "ld %s,0x%04x(%s)" % (REGS[rt], s16(imm), REGS[rs])
+    return ".word 0x%08x" % w
+
+def ram_to_file(ram):
+    if GAME_RAM_BASE <= ram < GAME_RAM_BASE + 0x200000:
+        return GAME_ROM_OFF + (ram - GAME_RAM_BASE)
+    if LIB_RAM_BASE <= ram < LIB_RAM_BASE + 0x40000:
+        return LIB_ROM_OFF + (ram - LIB_RAM_BASE)
+    # bss/data in lib segment: RAM 0x8002xxxx etc — not in ROM; skip
+    return None
+
+def main():
+    if len(sys.argv) < 2:
+        print(__doc__); sys.exit(1)
+    ram = int(sys.argv[1], 16)
+    n = int(sys.argv[2]) if len(sys.argv) > 2 else 40
+    off = ram_to_file(ram)
+    if off is None or off < 0 or off + 4*n > len(ROM):
+        print("cannot map RAM %x" % ram); sys.exit(1)
+    for i in range(n):
+        w = struct.unpack_from('>I', ROM, off + i*4)[0]
+        addr = ram + i*4
+        print("  %08x:  %08x   %s" % (addr, w, disasm_word(w)))
+
+if __name__ == '__main__':
+    main()