Formats - Legend of Legaia RE

How this section is organised

The pages are grouped the way the disc is layered. Disc + container is the outermost shell: raw sectors, the single big archive (PROT.DAT) almost everything lives in, and the developer-left name map for its entries. Compression + dispatch covers how the game unpacks and routes what it loads. Per-asset formats are the individual asset types themselves. Battle / stat tables are the data tables baked into the game's executable. Streaming + scene containers are the bundles a scene loads as a unit, and runtime overlay carriers are the archive entries that hold code rather than data. If you're new, start with disc geometry and PROT.DAT - every other page assumes those two.

Each format carries a badge for how solid its decode is. The badge is a claim about evidence, not about how complete the page looks - trust it accordingly.

Confirmed - verified end-to-end against real on-disc data, with passing tests.
Inferred - deduced empirically from byte patterns; structurally consistent but not yet exhaustively validated.

A format can mix the two, and the better pages rate it per-field rather than per-page - the encounter record's shape and reader are Confirmed while the encoding within scripts stays Inferred. Where a page has its own confidence section, that section is authoritative over the badge here.

Disc + container layer

What the disc image looks like before any in-game format gets involved: the sector layout, the one archive file that holds nearly everything, and the text file that names its contents.

PSX disc geometry Confirmed

How a PlayStation CD is physically laid out: 2352-byte Mode2 sectors of which 2048 bytes are file content, plus the ISO9660 filesystem on top. The iso crate's RawDisc::read_sector hides the sector layer and returns clean 2048-byte user data.

PROT.DAT TOC Confirmed

The disc's single big asset archive - 1233 numbered entries holding nearly every texture, model, script, sound bank, and code overlay in the game - and the table of contents that indexes it: start_lba = toc[p+2], size = toc[p+5] - toc[p+3] + 4. Walked at boot by FUN_8003E4E8.

CDNAME.TXT name map Confirmed

A developer-left text file naming the archive's entries via #define name N lines; each define tags a block start and names inherit forward. The numbers are the game's own raw TOC indices, offset 2 from this project's extraction numbering - a shift that explains a whole family of historical "mislabeled block" findings.

DMY.DAT (dev fixtures) Confirmed

The archive's sibling file, which turns out to contain no real game content: a memory-bus test pattern plus paired random blobs, shipped on the retail disc but never loaded.

Compression + dispatch

How the game turns archive bytes into usable assets: the compression scheme, the type-byte dispatcher that routes each blob to its handler, and the several near-identical "many assets in one file" pack layouts.

Legaia LZS Confirmed

The dictionary compression wrapped around most bulk assets - a custom LZSS variant reverse-engineered from FUN_8001A55C. 4096-byte ring buffer, zero-init - which means "decompresses without error" is not a validity signal; always magic-check the decoded output.

Asset type dispatcher Confirmed

How the game knows what a loaded blob is: FUN_8001F05C dispatches on an 8-bit type byte (0=TIM, 2=TMD, 4=MES, 6=ANM, 7=VDF, 8=SIN, 9=TMD2, B=MOVE2), deciding LZS-vs-raw per asset. The type names come from the devs' own error strings left in the executable.

Asset descriptor format Confirmed

The (type_byte << 24) | size calling convention used by FUN_8001F05C, and the descriptor-pair walker that no code in the executable calls - it's reached only from a runtime overlay.

Pack format Confirmed

The simplest multi-asset bundle: header u32 count + u32 word_offsets[count], then assets back-to-back. Used inside DATA_FIELD streaming chunks.

Standalone TIM-pack Confirmed

A near-identical multi-texture bundle used by some standalone PROT entries, with different header math: (magic_lo, magic_hi, count<16, marker=0x01) + offsets; byte_offset = word_index*4 + 4. Includes the hidden boot-UI bundle at the head of the archive.

Field-pack format Confirmed

A magic-prefixed (0x01059B84) scene bundle whose 97-slot indexed block turns out to be a byte-identical shared template across all 4 carrier entries - and whose magic nothing in the game ever checks. Consumer is FUN_801D6704 (overlay 0897); slots are read at hard-coded offsets.

Player battle files Confirmed

The per-character battle-model files data\battle\PLAYER1..4 (retail battle_data, extraction 0863..0866): header + LZS palette chain + 12-byte descriptor table + per-slot LZS streams that decompress to [32-byte header + Legaia TMD + texture pool]. The old “16 MB container at 0865” was an over-read window into the monster archive (0867).

Row-479 NPC CLUTs Confirmed

NPC colour palettes (CLUTs - palettes stored as rows of framebuffer pixels): plain PSX TIMs in scene PROT entries with CLUT block at (fb_x=0, fb_y=479, w=256, h=1). The engine uploads them with merge-zeros semantics so multiple scene TIMs targeting the same row coexist (full slots 0..14 + partial slots 0..7).

Per-asset formats

The individual asset types - what a single texture, model, sound bank, or dialog file looks like once unpacked.

PSX TIM Confirmed

The PlayStation's standard texture format. 4 / 8 / 15 bpp; palettes (CLUTs) scatter across PROT entries. crates/tim exports to PNG.

Legaia TMD Confirmed

The 3D mesh format - a custom variant of the standard PSX TMD, magic 0x80000002. Object pointers are byte offsets relative to header end; primitives use a custom 8-byte group header + per-mode descriptor table at DAT_8007326C.

VAB sound bank Confirmed

PSX-standard instrument bank (program/tone/SPU-ADPCM sample). crates/vab extracts and decodes to WAV. Banks live across the sound blocks; the retail vab_01 block (extraction 1070..1192) is wall-to-wall VAB - the old “vab_01 is misleading” was the +2 filename shift.

PsyQ SEQ Confirmed

The music sequence format - PsyQ's MIDI-derived pQES container. 13-byte header (PPQN, tempo, time-sig) + delta-time / MIDI events with running status. Drives SsSeqOpen / SsSeqPlay. crates/seq parses + walks.

XA-ADPCM Confirmed

CD-XA Mode 2 Form 2 streaming audio (voice shouts, ambience). crates/xa demuxes per (file_no, ch_no) channel and emits one WAV per stream.

MES dialog Confirmed

The dialog-text container. Two variants: Compact (0x404 magic) and Records (0x44 0x78). Bytecode tokens (Glyph / Op65 / Op4c / Op26 / End). Renderer is overlay-resident.

Dialog font Confirmed

The proportional Latin font the dialog renderer draws with (FUN_80036888). 256-byte width table at 0x80073F1C; 38-entry escape table at 0x80074050; glyph bitmaps in VRAM at (896, 0), 4bpp, 16x16 cells (drawn 14x15).

ANM animation Confirmed

Skeletal animation container (asset-type 6). (count + byte_offsets + records); marker_1 = 0x080C across all observed records. Keyframe pose decoder ported in legaia_anm::AnimPlayer.

Player-character meshes Confirmed

Where Vahn, Noa, and Gala's models actually live. Field form = PROT 0874 §0 (low-poly pack, parser legaia_asset::character_pack); battle form = assembled per character from the player battle files' equipment-id sections (battle_char_assembly), with PROT 1204 other5 as the Baka Fighter / default-equipment sibling pack. Object-local vertices; rest poses come from the ANM / record[0] idle streams.

Monster animation Confirmed

Enemy battle animation inside the monster archive (extraction entry 0867). Per-action transform-keyframe stream at entry +0x8c: [u8 parts][u8 frames] + nine-byte TRS records (one part per TMD object). Action 0 is the idle loop; decoder FUN_8004998c.

MDT - move tables Confirmed

Tactical Arts move records (the fighting-combo system's data). Both layouts parse cleanly; the consumer is FUN_800204F8. The runtime MOVE base is sourced per-scene from a CDNAME slot, not from a single boot-time PROT entry.

Move-power table Confirmed

Battle-action per-move power + behaviour records (26-byte stride, VA 0x801F4F5C, PROT 0898). Damage roll, homing, hit reaction, sound cue and spawned-effect lists; resolved via the id→index map at 0x801F4E63 (the same id space as the spell table).

Art data - Tactical Arts records Inferred

Per-character art records (Vahn / Noa / Gala; PROT entry 0x05C4). 40-field schema: command sequence, action constant, power bytes, hit timing, special / hit effect cues, status effects, repeat-frame replay.

Per-character save record Confirmed

The in-RAM record that holds one party member's whole state: 0x414 bytes at 0x80084708 + slot * 0x414. Cheat-database-pinned offset table for stats, level, magic rank, spells, summons, equipment.

Battle / stat tables

The game's numbers - item names, spell stats, equipment bonuses, steal chances - live as static data tables inside SCUS_942.54 (the game's main executable), not in the asset archive. Each table below is byte-pinned against the executable; DAT_80xxxxxx names are the tables' RAM addresses as labeled in Ghidra.

Spell table Confirmed

Static spell stats (DAT_800754C8) + name pointers (DAT_800754D0), 12-byte stride. Player Seru-magic block 0x81..=0x8B; named monster attacks at 0x25... Parser legaia_asset::spell_names.

Item-name table Confirmed

Static item-name table PTR_DAT_8007436C[id*3] (256 ids, 12-byte stride). The id space monster drops, steals, and equipment all index. Parser legaia_asset::item_names.

Item-effect table Confirmed

What each usable item does: 130 4-byte effect descriptors at DAT_800752C0 - effect class, tier, all-party / field / battle usability flags. Parser legaia_asset::item_effect.

Equipment stat-bonus table Confirmed

8-byte records at DAT_80074F68 per equippable item: attack / defense / agility / speed bonuses, equip-character mask, slot type, Ra-Seru flag. Parser legaia_asset::equip_stats.

Accessory passive table Confirmed

64-slot passive-index space mapping accessories ("Goods") to per-character ability bits (char+0xF4, aggregator FUN_80042558), with stat-% magnitudes inline. Parser legaia_asset::accessory_passive.

Steal table Confirmed

Per-monster [steal_chance_pct, steal_item_id] pairs at DAT_80077828 (chance FIRST - the reverse of the drop field order). What the Evil God Icon steals. Parser legaia_asset::steal_table.

New-game starting party Confirmed

4-record template at 0x80078C4C (Vahn / Noa / Gala / Terra; 26-byte stride) that seeds the live 0x80084708 character records; opening scene town01. Parser legaia_asset::new_game.

summon.dat / readef.DAT Confirmed

Battle side-band streaming slots (extraction 893/894 = raw TOC 0x37F/0x380), 0x10800 bytes each: per-special-attack CLUTs, texture pages, summon-creature actor records, and the player art-anim "ME" archives. Parser legaia_asset::summon_readef.

Streaming + scene containers

The bundle formats a scene loads as a unit - room geometry, textures, scripts, and audio wrapped together - plus the records that drive encounters and scene transitions.

DATA_FIELD streaming Confirmed

The typed-chunk multi-asset container parsed by crates/asset::parse_streaming. Sister detector data_field_truncated catches the shape where the final chunk's declared size walks past EOF.

Scene bundles Confirmed

The family of per-scene container shapes - five sister detectors covering ~50% of PROT entries: scene_tmd_stream, scene_vab_stream, scene_asset_table, scene_scripted_asset_table, scene_event_scripts, plus tmd_size_prefix.

scene_v12_table Confirmed

Per-scene container: runtime-fixup header + inline record table + move-VM stager prescript at sector offset 0x800. 97 PROT entries (one per scene); strict header algebra N = 4*param + 22; the field VM installs prescript records via FUN_800252EC, the move VM (FUN_80023070) runs them.

World-map slot-4 records Inferred

Slot 4 of each kingdom bundle (PROT 0086 / 0245 / 0392, type byte 0x05): a per-kingdom runtime library of small object-local 3D meshes. Each 8-byte record is a GTE vertex (i16 x, y, z, attr); the renderer reads the pool in place. Container byte-verified vs live RAM. Not the bulk continent terrain source.

Effect bundles Confirmed

Battle/field visual effects. Magic 0x02018B0C (efect.dat). 2-pack wrapper: 14 sprite-anim entries + 33 effect scripts indexed by effect ID.

Per-scene primitive scratch buffer Inferred

A negative finding: the 1700-byte RAM cluster at 0x80108EA4..0x80109550 that earlier readings called a "navmesh" is actually a per-scene scratch buffer the renderer fills with assembled GPU primitive data on scene entry. No RAM cell anywhere holds a pointer into the window.

Encounter record Confirmed

Which monsters you can meet where: the on-disc record installed at actor[+0x94] by the script-VM and read by FUN_801DA51C. Layout: [3 reserved][count: u8][monster_ids: u8[count]]. Reader copies the ids into the global formation cell at 0x8007BD0C.

MAN relocation Confirmed

How to safely resize records inside a scene's script file (MAN). Scene-transition (0x3F door) destinations are partition-2 records reached via the partition-2 record-offset table (runtime-pinned); resizing a destination name fixes the partition tables, section-0 offset, intra-record jump deltas, and the external descriptor size. Powers the door randomizer.

STR FMV table Confirmed

How the game finds its movies: libcd's CdlFILE directory cache at 0x801CAE08 and the 23-slot FMV dispatch table at 0x801D0A6C (32-byte stride) the play loop reads - nine retail slots covering every movie on the disc; per-scene fmv_id triggers are disc-pinned as literal operands in the scene MAN scripts.

Runtime overlay carriers

Archive entries that hold code, not assets: overlays - chunks of MIPS machine code the game loads into RAM on demand, where most of the game's logic actually lives.

MIPS overlay code Inferred

PROT entries carrying MIPS code that loads into the overlay window at 0x801C0000+. The detector finds them by the function-prologue fingerprint at byte 0.

Overlay pointer-table code Inferred

Sister format: same overlay-code family, but the file starts with a table of function pointers instead of a prologue. 42 PROT entries - including the ones titled with battle special-attack names like "Hell's Music".

Auxiliary

Window widget scripts Confirmed

The bytecode programs behind the UI's window choreography - the shop's slide-in/slide-away, the menu's staged windows. Fixed 4-byte instructions [opcode][window id][operand], resident as a program table in the menu overlay's own data segment, interpreted by FUN_801D6628.

SFX descriptor table Confirmed

The sound-effect lookup table: static DAT_8006F198 (8-byte stride, 100 entries). Per cue: VAB program/tone, voice count, mixer channel. Byte-exact vs live save-state RAM.

Sound-driver path strings Confirmed

The dev filenames the sound driver loads by: 3 SCUS consumers (sound init / streaming-asset loader / mode-aware ext dispatcher), 8 extensions: .MAP, .PCH, .spk, .dpk, etc.

Pochi-filler placeholder slots Confirmed

266 PROT entries are unused placeholders padded with the ASCII word "pochi" (a Japanese dog name - dev fill). The trap: past the padding sits stale mastering scratch that often parses as a complete, valid texture.