Skip to content
This repository was archived by the owner on Oct 4, 2026. It is now read-only.

MHP3RD_FILE_MAP.md

Latest commit

 

History

History
142 lines (113 loc) · 8.59 KB

MHP3RD_FILE_MAP.md

File metadata and controls

142 lines (113 loc) · 8.59 KB
Raw

MHP3rd File Map — T1 findings (extraction + monster packaging)

Branch mhp3rd-monster-port. Produced by orchestrator pass 2026-06-18. This is the starting map for the asset-pipeline agents — it has confirmed facts AND open questions flagged [OPEN]. Resolve the OPENs as job #1 of the asset track.

Extraction (DONE — reproducible)

# ISO already extracted:
workspace/extracted_mhp3/PSP_GAME/USRDIR/DATA.BIN        # encrypted, 1.09 GB
workspace/extracted_mhp3/DATA_decrypted.BIN              # decrypted
workspace/extracted_mhp3/data_files/file_NNNNN.bin       # 5954 extracted files

# Re-run (venv active):
PYTHONPATH=tools/mhef:tools python -c "import mhef.psp,extract_iso; \
  mhef.psp.DataCipher(mhef.psp.MHP3_JP).decrypt_file(\
  'workspace/extracted_mhp3/PSP_GAME/USRDIR/DATA.BIN','workspace/extracted_mhp3/DATA_decrypted.BIN'); \
  extract_iso.extract_data_bin('workspace/extracted_mhp3/DATA_decrypted.BIN','workspace/extracted_mhp3/data_files/')"

Key API delta vs MHFU: MHP3rd DATA.BIN is LBA-keyed → use DataCipher.decrypt_file() (NOT decrypt(buff) — that needs an lba per block). DataCipher.MHP3 attr does NOT exist; use the module constant mhef.psp.MHP3_JP (=6).

PAC container (same family as MHFU)

u32 count + count×(u32 off,u32 size) table + data. Scanner: tools/mhp3rd/scan_monster_pacs.py (<data_dir> for histogram, --detail <file> for one PAC's sub layout).

Format deltas MHP3rd vs MHFU (CONFIRMED)

Resource MHFU (gen-2) MHP3rd (gen-3)
PMO geometry magic pmo\0 ver 1.0\0 magic pmo\0 ver 102\0 (per-mesh scale, diff normals)
Skeleton sub-0 magic 0xC0000000, bone_count@+4 magic 0x80000000, bone_count@+4 (same blob family, diff top-bit)
Texture .TMH0.14 .TMH0.14 (same)
Animation sub-3, magic 0x64/hdr 0x18 [RESOLVED 2026-06-22] in-quest movesets are separate raw .anim files (file_03997–04016 small/medium; file_05142–05424 big monsters; Brute = file_05250, 77 clips). Same recursive data as MHFU, compact u16 headers, variable header size. Parser anim.parse_p3rd; see docs/ANIMATION_FORMAT.md.
Container [skel,pmo,tmh,anim] 4-sub multi-sub (12–13) w/ gen-3 FourCC tags NSP\0(0x0050534e) ASB\0 ARV\0, dual PMO + dual TMH

Monster model PACs (located)

The 16 high-bone-count PACs (bones>30, FourCC-tagged) are the monster models — two mirrored groups (likely LO/HI or two villages' display set):

group 1: file_04868(73b) 04870(82b) 04872(67b) 04880(86b) 04898(104b) 04900(88b)
group 2: file_05032(73b) 05034(82b) 05036(67b) 05044(74b) 05062(108b) 05064(91b)
extra:   file_05094(124b) 05106(36b) 05297(47b) 05412(47b)

The 22-bone files (e.g. file_00107 371 KB PMO) are PLAYER/armor rigs, NOT monsters.

[RESOLVED] Geometry location (OPEN-1)

Geometry is in a companion file: file_NNNN.bin = PAC (tables + skeleton + TMH), file_NNNN+1.bin = raw GE display list (starts with 0x14000000 ORIGIN_ADDR command). The PMO header field ge_base (vals[12] of struct.unpack_from('<I4f2H8I', blob, 8)) equals len(pmo_blob) for companion-file PACs → redirect GE reads to geo_blob. Lobby-style PACs (e.g. file_05248 sub[5]) have ge_base < len(blob) = self-contained. Confirmed: file_04898 (104 bones, companion=file_04899): 95 groups, 34701 verts, 24901 faces. Parser: tools/mhfu_model/pmo_p3rd.py::parse(blob, geo_blob=None). Tests: 9/9 pass.

[RESOLVED] Identify port target (OPEN-2)

Ground truth from user (domain expert) — overrides all prior heuristic labels:

  • file_05248 = BRUTE TIGREX ← PORT TARGET
  • file_05229 = DIABLOS (was incorrectly labeled "Black Tigrex em023" by color heuristic)
  • Prior em058/em023 labels for these files were WRONG; texture-distance and sub7 heuristics are unreliable for species identification. User visual ID is authoritative.

Port target — confirmed by user:

File Geo Bones Meshes TMH RGB Identity
file_05248 file_05249 46 88 (152,159,154) BRUTE TIGREX — port target

Render: tmp/brute_tigrex_candidates/target_file05248_CONFIRMED.png — tan/grey body, red markings, 88 mesh groups, fully textured.

Other lobby PACs identified in this session (renders in tmp/brute_tigrex_candidates/):

File Geo Monster Render result
file_05229 file_05230 Diablos dark-olive body — user confirmed
file_05412 file_05413 Unknown Tigrex-rig variant dark black/charcoal with red accents
file_05409 file_05410 Unknown Tigrex-rig variant orange/rust-striped
file_05297 file_05298 Unknown (NOT Tigrex) green wyvern

Note: sub7 species index and skeleton fingerprint heuristics produced multiple wrong labels in this session. Do not use them for species ID without user confirmation.

[RESOLVED] Animation location (OPEN-3)

MHP3rd monster animation is in the PAC sub[3] (in-quest PACs like file_04016/04900), NOT in a separate em/animation/emNNN.bin file (file_05413 turned out to be a GE companion geometry list, not an animation pack). The anim.parse_p3rd parser is DONE (Task 5).

  • Tigrex (em058, file_04900) sub[3]: anim_count=87 (magic), slot_count=60, 10+ valid slots.
  • Black Tigrex (em023, file_04898) sub[3]: similar layout.
  • file_05413 = GE display list companion (0x14000000 GE opcodes), NOT animation.
  • Parser: tools/mhfu_model/anim.py::parse_p3rd(blob) -> AnimationPack. Tests: 16/16.
  • Full keyframe decode (within each slot block) is v2 — block layout differs from MHFU's P3rd pack.

Skeleton format (0x80000000) — confirmed details

Same bone-section layout as MHFU 0xC0000000 (stride 0x5C=92, fields at identical offsets). Differences vs MHFU:

  • Header: may have extra 4-byte word at +0x1C (lobby PACs); probe both 0x1C and 0x20 for the first 0x40000001/2 section magic to find real start.
  • Section magic: both 0x40000001 and 0x40000002 appear (in-quest PACs alternate freely).
  • Bone count in header may be off-by-1 if last section is truncated.

tools/mhfu_model/skeleton_p3rd.py::parse(blob) handles all variants; returns same Skeleton dataclass.

Hitzone / part data (Task 6) — confirmed location

Verdict: MHP3rd hitzone RESISTANCE values are NOT in the per-monster PAC. They live in a separate game data table (same pattern as MHFU). The PAC carries geometry / physics / AI-hint data, not the hit-resistance numbers themselves.

For the Brute Tigrex / Tigrex port, use MHFU's existing Tigrex damage tables — no conversion needed (the hosted model rides MHFU Tigrex's combat rules).

Sub roles decoded (file_04900 = likely Tigrex, 88 bones)

Sub Magic / size Contents
[0] pmo\0 v102 Main PMO (hi-quality companion-file geometry)
[1] 0x80000000 Skeleton (88 bones, 0x5C/bone sections)
[2] 0x57=87 u32, 1652B Part table — count = bone_count − 1; likely bone→part-ID mapping + some per-part flags; stride ~19B; NOT resistances
[3] .TMH0.14 Main texture package
[4] NSP\0 ver 325, 200B Node Spring Physics — f32 spring/damp params + 3×3 matrix (cloth/tail simulation)
[5] ASB\0 ver 330, 80B Attack/Shield Boundary — bounding center (3246, −6506, −3740), 2 u32 counts, then u8 values (0x0C=12, 0x4B=75, 0x4B=75); the 75s are likely part-break thresholds or attack multipliers, NOT standard resistances
[6] ARV\0 ver 321, 16B Attack Rate Variation — empty (all zeros); subspecies variant flag
[7] 2B blob Tiny flags blob (01 05)
[8] 0x00000102, 4096B CLUT / palette — indexed-texture look-up table (u16 pairs 0x4E20 = 20000, repeating = palette entries)
[9] pmo\0 v102 LOD / small PMO
[10] .TMH0.14 LOD texture
[11] 0x00000003, 5792B Hitbox geometry — sub-offset table (9 blobs): f32 radii, spring constants, collision sphere centers; NOT resistance values

Animation sub[3] format (confirmed, Task 5)

u32 anim_count, u32 hsize=0x18, u32 slot_count, pad; slot table at hsize-4. Per slot: u32 bone_count, u32 block_size, u32 loop_flag. Keyframe decode = v2 (deeper RE needed). Parser: tools/mhfu_model/anim.py::parse_p3rd(blob). Tests: 16/16 pass.

Sub-resource magics seen on a monster PAC (file_04868, reference)

[0] pmo  v102  | [1] skeleton 0x80000000 (73 bones) | [2] 0x00000048 (?)
[3] tmh        | [4] NSP\0 | [5] ASB\0 | [6] ARV\0 | [7] tiny(4B)
[8] 0x00000102 (4096B, palette/clut?) | [9] pmo v102 (small) | [10] tmh (small) | [11] 0x00000003 (?)