Status: DECODED scales,
INFERRED play volume (aurasnd; ambient
11-14)
Now: Send scales: a plain
request sets all eight to v = volume × sound
option × class volume (0x140240870), a
positional one takes them from the 3-D routine
(0x140241440); the 127 in s is still our
reading
Lead: who fills the
request’s +0x5c..+0x64 (the s multipliers)
Which
multipliers s takes
Status: INFERRED
(aurasnd)
Now: s = play volume × cue
volume (REAT 0x10) × track volume (TRAT 0x10)
>> 9
Where:pd_sound.gdtrack_gain()
Lead: the voice set-up at
0x140280121..0x14028015b reads them from the
play request (+0x5c, +0x60, +0x64); trace who
fills those
5.1 → stereo fold
Status: INFERRED
(aurasnd; ambient 11-14)
Now: Left = FL + 0.7071 (C
+ RL), right = FR + 0.7071 (C + RR), each over
1.7071; a track whose sides differ (every
positional request, one-sided sends) plays as a
left player on bus PdSendL and a
right partner on PdSendR (hard pan,
−6.02 dB); a stereo clip there is folded to mono
per side; equal sides play one stereo player.
(ambient r2 11-15) Mix levels: per
speaker the Palace bed (front send 125: −11.20
dB) and the music (volume and send 127: −4.67
dB) play exactly the decoded levels at the
default options, but the fold’s 1 / 1.7071 puts
both 4.65 dB lower on our two channels (bed
−13.35 dB, music −9.32 dB, test_ambient_scene
F): the balance is the decoded 6.53 dB, the
absolute stereo level is INFERRED with the
fold
Lead: the game hands
XAudio2 a 6-output matrix (per source channel:
not read); the device downmix is Windows’
Effect
send (id 11) and send id 12
Status: MISSING
(aurasnd)
Now: Not played (no effect
/ reverb voice)
Where:pd_sound.gd
Lead:(ambient
11-14) a positional request gives both
sends the linear distance factor (no direction);
the stage cues use them (highway 60–100 on id
11, Palace 60–100 on id 12). Open: the second
output voice (index at voice +0xf2) and its
effect chain; DSP_*.bin are not it
(the port never applies them, row
“DSP_*.bin files”)
Bank master
level
Status: MISSING
Now: Read into the index,
not applied
Where:pd_sound.gd
Lead: as above
Pan and 3D
distance falloff
Status: DECODED
(ambient build 11-14)
Now: The 3-D send routine:
per-cue near / far / curve / spread from
soundparam.dat, five speaker scales
around the local character (position and
facing), silent beyond far, recomputed every
frame (PdSoundMixer); the listener
is the Player (main.gd), the camera
only outside a battle. Track / cue pan bytes
(REAT / TRAT 0x14) still not applied
Lead: open: the Doppler
pitch step (0x140240e20), the listener’s second
point (C + 0x13f2c), the pan bytes’ reader
Positional
only with soundparam flag 0x10
Status: INFERRED
(ambient build 11-14)
Now: A play with a position
gets the 3-D sends only when the cue’s record
has flag 0x10 (540 of 650 common records);
others play as plain requests, a cue with no
record (an index without
param_groups) too
Where:pd_sound.gdrequest()
Lead: the start path of
0x140240a00 for a record without the flag
Voice bank
parameter group
Status: INFERRED
(ambient build 11-14)
Now: A voice bank uses
group 10 (player slot 0) for every caster
Where:pd_sound.gd VOICE_PARAM_GROUP
Lead: 0x140533708[slot] for
the voice bank slots (3 + player)
Stage
ambience: stage bank and sound emitters
Status: DECODED
(ambient build 11-14)
Now:pdsound.py converts every
MapSE bank and writes each stage
layout’s sound entries;
PdStageSound spawns them (held:
once; one-shot: every interval + 1 ticks,
repeat plays). Palace: bed, 10
water points, 6 drip emitters; st02: bed
Now: Stage cues 0 and 1
started once without a position; each tick both
take clamp(storm intensity, 0, 1) × 0x1000 (the
papers copy the storm’s value: 0x1402441b3 reads
storm + 0xec)
Where:pd_stage_sound.gd
Play volume
classes
Status: DECODED, fader
MISSING (ambient 11-14)
Now: Class = soundparam
bits 0–1, class volume 1.0 for all four
(0x14023fd40), multiplied into every
request
Where:pd_sound.gd CLASS_VOLUME,
request_volume()
Lead: the fader object
(0x1401baa40) that ramps a class mask: its
users
Movement
and landing cues of the character code
Status: MISSING
(ambient 11-14, sites from docs 11-10 /
audio)
Now: Not played
Where:player.gd,
combatant.gd
Lead: sites 0x1402528d0
(cue 0x24 or 0x40 by character), 0x140254780
(0x26), 0x1402549e0 (0x43), 0x140254b30 (0x4C),
0x140254c80 (0x52): which movement each belongs
to is not decoded
Now: 84 play requests (a
request = all tracks of one cue); a new request
with every slot busy is dropped
(test_ambient_scene G). The sends are recomputed
once per game tick (30 Hz, 0x1402400f0 is the
tick pass; H)
Lead: the event that steps
over them looks like a pitch or pan sweep
Event length
(0x1C)
Status: DECODED
(ambient build 11-14)
Now: Exported per track
(length_ms, end_mode);
the track stops at start + length (checked each
frame against the pool clock: up to one frame
late)
Where:pdsound.py,
pd_sound.gdrequest()
Lead: — (the 12 at 0x60 is
the pitch range, “Sound banks”, Pitch)
DSP_*.bin
files
Status: DECODED (docs
11-09): nothing to do
Now: Not read
Where: converter
Lead: Xbox audio DSP effect
images; the PC port loads the area’s file and
frees it without applying it (FORMATS “Sound
banks”), so the port has no reverb from them.
Only relevant to a later Xbox-faithful audio
option
Looping
cast sound stopped at release
Status: INFERRED
Now: Stopped at release,
and when the cast is cut short (caster killed or
freed)
Where:pd_sound.gdend_cast(),
data_skill.gd_cast
Lead: the game stops it at
the next cast; whether also at release is not
traced
Voice line on
every cast
Status: INFERRED
Now: Always on
Where:pd_sound.gdplay_cast()
Lead: the game does this
when its voice option is on; no option yet
Voice line pick
Status: INFERRED
Now: Deterministic per tick
+ skill id (useful once client hooks pass the
event’s tick); the game draws at random
Where:pd_sound.gdvoice_pick()
Lead: the game’s RNG call;
random is enough unless replays need it
Netplay client
audio
Status: MISSING, narrowed
(audio)
Now: A client plays the
stage music (stage and seed from the hello) and
its own lock-on and refused cues (from its own
body); it still hears no skill sounds or
voices
Where:data_skill.gd,
projectile.gd
Lead: hook NetGhost
spawn/gone and the cast/hit events
Voice language
Status: PLACEHOLDER
Now: English banks only
(converter --lang picks one
language per index)
Where:pdsound.py
Lead: an in-game language
setting
Boss sound banks
Status: MISSING
Now: Not converted; ids
above the boss base are silent
Where:pdsound.py,
pd_sound.gdresolve()
Lead: not used in
versus
Music
Status: DECODED
(audio)
Now: Converter
pdmusic.py writes one WAV per track
with its loop sample, the track table (flags,
group, file), the group overrides and the battle
scripts’ stage → track table;
PdMusic plays it as the play
routine does (0 stops, above 60 ignored, the
playing track returns, stop then start, the
group override), loops from the loop sample to
the end, at the decoded level (channel volume
and sends × music scale, option 60 on a fresh
save); the battle start plays table[stage]
(FORMATS “Music”, “Music volume, the sound
volume option and the battle UI cues”)
Where:pdmusic.py,
pd_music.gd,
main.gd
Lead: listen across a loop
point (row “Music loop seam by ear”); the rows
below
Music loop seam
by ear
Status: INFERRED
(audio)
Now: The loop starts at the
decoded block’s first sample; a continuity check
had it smoothest in 18 of 35 files (chance about
1 in 7), not listened to
Where:pdmusic.py
Lead: playtester listen
across the seam of a few tracks
Music
stage pick (tracks 16 / 19 + 0..2)
Status: INFERRED
(audio)
Now: The pick is rand() × 3
>> 15 from a PdRand seeded with the match
seed (both netplay peers pick the same); the
game draws from its shared rand() at gamestart,
whose state there is not traced
Where:pd_music.gd battle_start
Lead: the rand state at the
battle script’s gamestart
Music fade
(native 0x1014)
Status: INFERRED
(audio)
Now: A straight line per
tick to the target; a fade ending at 0 stops the
track
Where:pd_music.gd fade()
Lead: the step of the
stream fader modes 5 / 6 (0x1402a99f0..)
Music centre /
rear sends
Status: INFERRED
(audio)
Now: Folded into the
channel’s side at −3 dB and normalised like the
sounds (row “5.1 → stereo fold”); sends 6 / 7 of
two tracks ignored
Where:pd_music.gd gain()
Lead: as “5.1 → stereo
fold”; what sends 6 / 7 are
Music
and sound volume options
Status: INFERRED
(audio)
Now: Fixed at the
fresh-save value 60 (decoded, 0x1401cb0e4); no
options screen sets them
Where:pd_music.gd option,
pd_sound.gd se_option
Lead: an options screen
(menus)
Menu music API
wiring
Status: INFERRED, narrowed
(menus r3)
Now: The menus’ natives
call PdMusic: 0x1010 / 0x1012 play (force 0 /
1), 0x1011 stop, 0x1013 volume, 0x1014 fade (the
title plays its track through 0x1010; the
battle-end script’s 0x1011, 0x1010(37), 0x1014
run too). main.gd_leave_to_chooser still stops the
battle track for the debug starts
Where:pd_menu.gd group 1,
pd_music.gd
Lead: the fade step (row
“Music fade”)
Stage
sound emitters in object lists
Status: INFERRED
(ambient build 11-14)
Now: The EF00 lists of
placed stage objects (definition state 0) are
placed with the placement (Rz·Ry·Rx, then the
position) and spawned with the stage; they stay
where they started (a moved or broken object
keeps its emitter). Counts match the docs’ table
(st08: 29 one-shots, st05: 5 points)
Now: None (storm bed and
papers: built, row above)
Where:pd_stagefx.gd
Lead: cue numbers and sites
listed (FORMATS “Stage sounds”); the stage banks
are converted now. (collapse 11-12) the
object break / support-lost / collapse handlers
start no cue of their own besides the sound
row’s break sound (M) (stagefx 11-15)
the collapse model’s cues 1 / 2 have no position
and belong to the st04 family and st08 only (M):
no stage cue on a Palace or Highway
collapse
Countdown
cues and sprites
Status: decoded
(menus)
Now: Menu sessions: cue 632
at ready ticks 0 / 30 / 60, cue 633 at 90;
number sprites 0x50 / 0x4F / 0x4E solid at (320,
166) with their banner-A glow (grey 40, step 2),
then sprite 0x54 as banner B (alpha 255, −8 per
tick after 50) and A (grey 50, step 1) at (320,
150) (FORMATS “battleend” → countdown
placement). Not shown for the debug / chooser /
net starts
Lead: decoded (M): cue
table by ground attribute × two columns + foot,
10-tick cooldown, at the foot joint.
(ambient build) per-foot step routine
0x140231010 on the display object: foot index at
+0x128, records of 0x58 bytes; a step fires when
record +0xb0 is set and +0xb4 is clear and the
cooldown (+0xbc) ≤ 0; position = the joint
matrix at record +0x80; column 1 for character
state 7..11 off a network match (0 otherwise),
in a network match from the remote object’s
+0x6d0. Open: the writer of record +0xb0 / +0xb4
(the foot-contact signal) and the caller passing
the ground attribute
Game
menus (title, main menu, versus set-up, pause,
results flow)
Status: INFERRED
(menus)
Now: The user’s own menu
scripts run on our interpreter
(PdMenu, decoded engine: table, go
to / back / history, tick, items, texture sets,
fader, event tasks, name stack, the group 0/1/2
built-ins the reached screens use). Reached:
profile screen with the title (7), main menu
(8), versus submenu, vsmode (5, player 2 joins
on pad 2), vsmenu (10: rule, stage, characters,
arsenals, lobby), the battle loading screen
(72), pause (49), results (66), lobby (70 →
rematch). Ours: player 2’s side is played by the
CPU (no local split-screen in our build); online
/ PC-port groups 9 and 10 read 0; the save-slot
layer is skipped (row “Menu save data”); the
pause opens with scene argument 0 = 0 (offline
rows, our reading of the pause object’s +0x12),
exit code 129 is read as “retry the battle”
(INFERRED) and our host clears flag 0x100000
when the pause closes (INFERRED); battleend
draws its banner and prompt (group 8 decoded)
and closes on the confirm; the split-screen
versus has no results list in this build
(decoded: only the online / system-link scenes
set it up). The options screen (4) is driven by
a test (controller rows)
Now: The game’s menus are
off by default: the debug
chooser stays the default start. They start with
--pd-menus (or
--menus[=<id>]), or with the
chooser’s “Start screen” row (saved in
user://menus.cfg, read at launch;
headless runs read only
--pd-settings=<file>). The
title drawing is built now (row “Title screen
look”)
Lead: playtester: make the
menus the default start now that the title
draws
Menu window
frame look
Status: PLACEHOLDER
(menus)
Now: Decoded two-colour
fills, list / text-box / help-bar colours and
text positions; our 1 px border instead of the
border sprites; the list scrollbar is a flat bar
in the decoded colours
Now: The scripts’ own
0x102B cues play through the common bank
(PdSound.play_flat), list cursor
moves play cue 3 as decoded
Where:pd_menu.gd_cue
Lead: the cue volume / pan
of 0x102B (played flat here)
Menu
save data (profile slots, arsenal names)
Status: PLACEHOLDER
(menus)
Now: The PC port’s
save-slot layer (0x208A–0x208C, 0xA00A / 0xA002,
records of 0x388 bytes at 0x14059ad50) is not
emulated: on the profile screen the slot
enumeration jumps to the main menu as after
loading a profile; in vsmode player 2 sees one
unnamed slot; the arsenal list has one entry per
player (our chosen arsenal) named “ARSENAL
n”
Where:pd_menu.gd (0x208A–0x208C, 0x2095,
0xA002)
Lead: the slot record
layout and the profile name / arsenal name
fields
Menu
data queries (characters, stages, unlocks)
Status: INFERRED
(menus)
Now: 0x2096 reads the
user’s character table
(menu/charaset.json from
charaset_.dat), 0x2097 the stage table
(stages.json); “unlocked” =
converted on our disk (the game reads the save’s
unlock bits); the character’s colour picks fall
back to colour 0 when that colour isn’t
converted
Lead: the save’s unlock
bits (0x1404616a4, bit 27 + i)
Menu
$-keyed strings
Status: PLACEHOLDER
(menus)
Now: Captions starting with
$ look up the PC port’s run-time
string table [0x1405b0918] (loaded by
0x140022820); the key is not in the game files
we have, so these draw nothing
Where:pd_menu.gd_text
Lead: the table’s source
(0x140281b6e)
Title screen
look
Status: INFERRED (menus
r3: built from docs 11-12 title)
Now: Built from FORMATS
“Title screen drawing” (PdTitle):
FDLogo sprites 0, 1, 3–11 from the user’s
texture-set slot 0; cloud plane (projection, UV
corners and per-draw steps, vertex colours,
pixel formula), base logo (A), glare chain (five
levels 256 × 192 … 16 × 12, 8 + 16 four-tap blur
passes, multiplied in at 0.8, additive tint
0.75ⁿ), glow logo, prompt with pulse, nine flame
sheets (frame offsets, masks, mirroring, flame
6’s u quirk), final logo (B), legal line (u to
1.09524), gold colour set, the decoded levels
and timing (intro 126 ticks, pulse 40, leave 50;
frame counter F steps once per menu tick).
INFERRED in our build: the alpha reference value
(0: only alpha 0 is discarded), the quad
triangulation for flame 6’s quirk (TL-TR-BR /
TL-BR-BL), the glare source cleared to
transparent black at 853 × 480 (the port samples
its window-size back buffer), the blur passes
folded into one kernel (borders clamp once), the
cloud steps per menu tick (30 Hz), the title
drawn under the script’s own sprites
Lead: open: alpha reference
value, colour channel order (M), the quad
helper’s triangle order; compare with
footage
Logos
movie and idle demo movies
Status: PLACEHOLDER
(menus r3)
Now: No movie playback:
0x312D (open) and 0x312B (finished) answer
“done” at once, 0x312F / 0x3127 do nothing, so
the profile script falls through to the title
(on the reached path the script opened no movie:
0 opens in test_title)
Where:pd_menu.gd
Lead: decoded: started by
the profile script through the event-movie
built-ins (open, skip mask, finished), demo
after 1800 idle ticks; not decoded: the open
result for a missing file
Title jingle
Status: MISSING (docs
11-12 title)
Now: None
Lead: the title sound bank
is loaded / released by 0x8046–0x8049; the cue
and 0x804A not decoded
Menu
blend modes, 3D backdrop and model preview
Status: PLACEHOLDER
(menus)
Now: 0x20A3 (blend state)
is kept but draws stay plain alpha;
0x2000–0x2007 (menu lights, backdrop object) and
0x20A4 (character model preview) draw
nothing
Where:pd_menu.gd
Lead: the engine’s blend
enum (render state 0x14028e4e0), 0x140208590
dispatch
Menu
message box and dialog look
Status: PLACEHOLDER
(menus)
Now: Type 5 message boxes
draw as a text box with our layout; A → event 2,
B → event 4
Where:pd_menu.gd_draw_dialog,
_dialog_input
Lead: 0x140169d60 (dialog
object)
Menu
font shadow / fade-in start
Status: INFERRED
(menus)
Now: Rich-text mode 2 / 4
shadow = black at +1,+1; a fade-in started at
alpha 0 starts from 255
Where:pd_font.gd, pd_menu.gd_fade
Lead: the font2 shader’s
shadow pass; the fader’s start value
Battle
UI cues (lock-on, card pick, KO)
Status: DECODED, narrowed
(audio)
Now: Played: the lock-on
cue 4 (the HUD panel’s 10-tick on-screen timer,
re-armed by every off-screen tick or switch,
also after the cue; the count comes before the
restart in a tick) and the refused-press cue 2
(a status refusing the pickup / the button, the
skill-allowed check, aura level below the cost;
an empty button and aura-short only are silent),
for this machine’s player only, decided locally
on a netplay client too. Not found: a KO cue;
battle end belongs to the menus’ battle-end
script
Where:pd_battle_cues.gd,
player.gd
Lead: KO sites; the card
pick (capsule take) cue is row “Capsule take /
move one-shot models and sounds”
Lock-on cue
on-screen test
Status: INFERRED
(audio)
Now: The target counts as
on screen when its reticle aim point projects
inside our camera’s viewport and in front of it;
the game reads its camera tick’s projection flag
for its fixed 853:480 view
Where:player.gd _on_screen
Lead: the projection flag’s
bounds (camera tick step 9, 0x140253c00)
Announcer
Status: — (docs 11-10
audio)
Now: None
Lead: decoded: the game has
no announcer voice; the countdown is two common
cues (row “Countdown cues and sprites”). Nothing
to build
Cast rumble
Status: MISSING (docs
11-10 audio)
Now: None
Where:skilldata.py,
player.gd
Lead: decoded (H): cast
profile fields 0x0A (end mode), 0x0C..0x33 (five
steps), 0x34..0x4B (release steps) drive a
per-pad sequencer, local player only; output
chain and gates (FORMATS “Controller
rumble”)
Shake-source
and stage rumble
Status: MISSING (docs
11-10 audio)
Now: None
Lead: decoded (M): shares
the camera’s positional shake sources (falloff ×
fade × pattern), EF00 screen-shake entries and
the collapse add right-motor rumble. Open: the
sources’ numbers (with camera shake).
(collapse 11-12) the stage object
break, support-lost and collapse-family handlers
start no shake or rumble themselves (M); it can
only come from the effect records they spawn
(effects decode)
XACT
bank (media/Sound.xsb,
Wave.xwb)
Status: — (docs 11-10
audio)
Now: Not converted
Lead: decoded: unreferenced
by the game; nothing to do (FORMATS “The XACT
bank”)