Now: Every effect draw
(meshes, particles, trails) takes its material’s
decoded blend state through
PdEffect.blend_spec: all 26 pairs
in the data are drawn by their formula (fixed
Godot blend + source term in the shader, or from
the screen texture); only the two rows below
remain
Where:pd_effect.gd
Lead: see “Effect blend
from the screen texture” and “Effect blend:
frame alpha”
Now: Mesh per-part
animation and particle01/02 (type-0 emitters)
are decoded (FORMATS “Effect particles”); trails
are decoded too; msparticle fields are still
read in one place, marked INFERRED
Where:pd_effect.gd,
effectconv.py
Lead: see the trail,
msparticle, particle and loop/blink rows
below
Effect
blend from the screen texture
Status: INFERRED
(effectdecode)
Now: dst color / one (548
elements; dst × (1 + s) can pass 1 and Godot’s
mul clamps, blend probe) adds d × s read from
the screen copy, so stacked layers give d (1 +
s1 + s2 …) instead of the game’s d (1 + s1)(1 +
s2) … (the s1·s2 terms are missing; exact for
one layer) (effectdecode2: writing d (1 + s)
made each layer overwrite the ones below, the
stepped rings of eft0042). The other pairs
no fixed Godot blend makes (plain
sub src − dst: 2; min/max: 5; a dst
term in the source factor with another dst
factor: 2; reverse subtract with dst factor ≠
one: 1) are computed from Godot’s screen texture
and written over it: exact over the opaque
scene, blind to effect layers drawn earlier in
the same frame
Where:pd_effect.gdblend_spec
Lead: none in Godot’s fixed
blends; a per-layer framebuffer copy would be
exact
Effect blend:
frame alpha
Status: INFERRED
(effectdecode)
Now: The frame’s alpha is
taken as 1 in factors that read it (one element
uses inv dst alpha; none dst alpha or src alpha
saturate)
Where:pd_effect.gdblend_factor
Lead: what the game’s back
buffer alpha holds after the stage draw
Effect
trails: UV format, history rate
Status: INFERRED, narrowed
(effectdecode2)
Now: tale01 is decoded
(FORMATS “Effect trails”: crossed ribbons,
Catmull-Rom head, cut-back, fixed grey, scroll);
the UV halves are read as SNORM (across 1 / 0)
and the trail advances once per game tick
Where:pd_effect.gd_step_trails
Lead: the input layout
bound for the trail’s vertex format (0x1401cbbc0
flags 0x1e09c41); whether
pdEftTale03 runs once per tick or
per drawn frame (its 0x140530034 flag test); the
states 0x8f / 0x40 / 0x90 (shared by most effect
draws, set once per mode 2; not blend: the
draw’s blend is decoded, see FORMATS)
Effect source
files
Status: DECODED
(effects1)
Now: The data build
converts the 496 pack entries (e.i
+ aN.e,
effectconv.py --pack, struct
reads); the loose revision is no longer read.
One pack entry has a texture name with no
texture in the packs (a stage haze mesh, no
skill uses it)
Now: Built: one
camera-facing sprite per element at 5000 ×
min(1, 2 L²) / depth pixels on a 480-line screen
(the Xbox size, setting “Xbox effect sprites”,
default on); off = the PC port’s size (divided
once more by depth)
Where:pd_effect.gd_make_sprite,
sprite_px
Lead: footage of records 42
/ 53 on both platforms to confirm
Effect
draws not compared with the Xbox
Status: INFERRED (docs
11-10 xboxfx)
Now: msparticle, thunder,
flair, blade and filter draws follow (or will
follow) the PC code only
Where:pd_effect.gd
Lead: data, element
functions and shaders are equal on both
platforms; only mseftbb01s was
found translated wrongly. Not compared: the
draws named here; the haze textures not in the
signed format (510 meshes)
Effect
msparticle emitters
Status: INFERRED
(effectdecode)
Now:msparticle00/01 (1061 elements)
keep the first CPU-particle reading (count,
life, size, spread, rise) with the decoded
blend
Where:pd_effect.gd_make_ms_particles
Lead: their own exe classes
MSParticleFunc00..03 /
alfn_FunctionmsPARTICLE00/01
(vtables by RTTI)
Effect
particle emitter type 1
Status: PLACEHOLDER
(effectdecode)
Now: Elements with
raw_m[12] = 1 (106) spawn with the
decoded type-0 spawner
Where:pd_effect.gdspawn_particles
Lead: the type-1 spawner
0x1401a2bc0 and its vertex builds 0x1401a3130
(particle01) / 0x1401a4d60 (particle02)
Effect particle
drift
Status: PLACEHOLDER
(effectdecode)
Now: The acceleration term
global vector × raw_m[10] × 0.01
(524 elements have raw_m[10] ≠ 0)
is left out (vector taken as 0)
Where:pd_effect.gdspawn_particles
Lead: the vector is 0.3 ×
the vector at +0xC8 of the object the exe keeps
at 0x140533308 (set at 0x140222d73; copied at
0x140189a6c into the effect context
+0x115A8)
Effect
particle matrix source 3
Status: DECODED (M)
(effects1)
Now: The acceleration is
turned by the caster’s root matrix (the owner of
the instance; the effect’s own when it has
none)
Where:pd_effect.gd emitters
Effect
dynamic light (light02)
Status: INFERRED, narrowed
(effectdecode-fix, bofxemu, bof2);
stage part DECODED and built (bof4),
capsules not fed
Now: Decoded: color, range,
fade envelope (PdEffect.light_at,
exe 0x14010c990); the pool record’s life
(int(end + 1) ticks after its effect is gone,
then 15 ticks of colour × 0.625, freed at once
past its window;
PdEffect.pool_step, 0x14019a240);
the per-character pick of the best two (range /
d², 0x1402998f0) and the two channels that fade
in/out over 3 ticks when two lights compete, cP0
= channel 0, cP1 not fed in battle
(PdCharacter._step_light_channels,
0x140299be0, 0x140199e70). Corrected
(bofcmp 11-13): the stage does get
effect light — under light set 0, point
light 0 is the most recently allocated effect
record while it is live (0x140199970, index
0x140527cc8); built in bof4 for the stage
materials (row “Stage lit by the latest effect
light”). Still drawn additionally as a Godot
OmniLight3D (lights nothing on our own
shaders)
Lead: left INFERRED: the
pick is measured from the character’s origin
(the exe passes a per-player position, not
traced); the pool update’s order against the
effects’ update inside a tick (±1 tick of life);
with no light the exe binds cP0 to pool record
0x42 (0x1405a8720), the set 0 point light
(bofcmp), where ours writes black, so
st12’s stage flashes
(pd_stagefx._apply_light) and
effect lights still overwrite each other
eft0061 hit
burst
Status: DECODED
(effects1)
Now: The played pack
entry’s hit slot has no petal quads (that was
the loose revision): one point-particle emitter
on a joint 0.2 m up and one particle emitter,
each on its joint
Where:pd_effect.gd
eft0061 skill
cast pose
Status: INFERRED
(bofxemu)
Now: Ours plays the cast
slot with the arm forward at the chest; PC and
xemu footage raise the hand beside the head
Where:data_skill.gd_cast
Lead: which motion slot the
exe picks for this skill’s cast
Now: Decoded: particle01’s
QuadMap corners and UVs (bof2); the
size push (bof3): the draw
(0x1401a1c00) uploads cPROJEC = particle matrix
× view (0x1405a5ca0) × the fixed 75° 853:480
projection (0x1405a5c60), so w is the view depth
in game units and the push is a view-space
square of half-size size / (853 cot 37.5°)
units; ours pushes by its projection’s x/y
scales × that (PART1_PUSH,
part1_push), equal to (1/853,
1/480) at 16:9 and square on the Deck’s 16:10
(it was 10 % narrow there). Still inferred: the
game’s rand() sequence from seed 1
(the game’s state is shared and
clock-seeded)
Now: Every element (mesh,
particle emitter, light, trail, sprite) sits on
its joint: local T·Rz·Ry·Rx·S (Z mirrored), then
the owner’s bone of the same name with its
rotation (names → bones from
<data>/effect/bones/<model>.json,
effectconv.py --bones), else the
instance root. Cast instances (slot 0) are
rooted at the caster’s live root rotation at its
place at the cast; release (3) and loops (4 / 5)
at the caster’s matrix at their start
Where:pd_effect.gdjoint_xform,
root_xform
Lead: bones are converted
for the converted characters only (2 models);
others fall back to the root
Effect
draw layer (.eft per-model value 29
/ 27)
Status: MISSING (docs
11-12 effects)
Now: Not read; all effect
models draw in one pass
Lead: decoded field (H): 29
→ draw pass 0x1E, 27 → 0x1C; what differs on
screen not followed
Joint-pair
elements (LHRH, B1RH, … 14 names)
Status: MISSING (docs
11-12 effects)
Now: Not handled (such an
element sits at the effect origin)
Lead: pair table decoded
(M); the consumer 0x1401130e0 not followed
Effect
screen position for heat haze
Status: INFERRED, narrowed
(effectdecode-fix)
Now: The game maps the
projected (pushed) position by x × 0.5, y × −0.5
(c37 decoded, 0x1402f8ab0) plus
cSCREENOFFSET; Godot adds the same
scaled shift to SCREEN_UV, taking Godot’s
SCREEN_UV as y-down like D3D texture space, and
leaves out cSCREENOFFSET
Where:pd_effect.gdshader_code
Lead: the value the exe
writes to cSCREENOFFSET (VS
cb0[13])
Effect
frame copy (opt06) alpha kill
Status: PLACEHOLDER,
narrowed (effectdecode2)
Now:pdeftopt06 (7 meshes) is drawn with
its decoded math, push, format 0x1E, blend and
now its UV scrolls
(PdEffect.opt06_scroll: signed
nibbles of
raw_m[6]/raw_m[7] × t
/ n, FORMATS “Effect drawing”); the per-texture
alpha-kill flags (alphakill) are
left off
Where:pd_effect.gdshader_code(copy)
Lead: who writes PS
cb0[23]: the three flags are set together by
0x1401df810 (reached through the vtordisp thunk
0x1401e5fc4 of an interface vtable; its callers
are not traced); each flag discards its
texture’s zero-alpha texels
Effect mesh
matrix sources
Status: DECODED (M)
(effects1)
Now: All of 0–8 drawn as
decoded: 3 / 4 the caster’s root (live), 5 the
spawn matrix’s position, 7 the caster’s root at
its spawn-time place, 8 the caster’s axes at the
live position; 1 / 2 / 5 / 6 frozen as before.
Frozen matrices are kept for the instance’s life
(not reset when a loop wraps)
Where:pd_effect.gdmesh_xform
Lead: the exe’s table is
keyed by (effect, element): its loop behaviour
is unknown
Effect
blades (blade01,
blade02)
Status: MISSING
(effectdecode3)
Now: 20 elements are
dropped: PdEffect.setup’s element
loop has no branch for them, so nothing is
drawn
Where:pd_effect.gdsetup
Lead: blade01 is decoded
except its spline (FORMATS “Effect blades”:
history, geometry, weights, shaders, blend
0x5401); still to read: the coefficients of the
spline helper 0x14017e120 (a quintic per span
evaluated at t = j / K; coefficients from
0x14017e760 set-up and 0x14017ea90 per span),
then draw it with the frame copy;
blade02 (pdEftBlade02
draw 0x14019fa80, update 0x14010b850, 40-byte
vertices) is unread
Effect flairs
(flair01)
Status: MISSING
(effectdecode3)
Now: 100 elements are
dropped: no branch in
PdEffect.setup’s element loop
Where:pd_effect.gdsetup
Lead: RTTI
alfn_FunctionFLAIR01 (vtable
0x140479d28, update 0x14010c480), draw object
pdEftFlair1 (0x140479d70, draw
0x1401a0dd0, object init 0x14010c2e0, per-tick
step 0x1401a1170); draw list flags 0x1e09c41.
Read so far: the update stores the element, its
window length, colors and blink colors, then by
int(raw_m[3]) takes 0 the full
effect matrix, 1 the position only, 2 the
position and the matrix, 3 the position and the
effect-context +0xA0 matrix;
int(raw_m[12]) goes to +0x108.
Shaders
Flair1V/Flair1P: plain
transform, texture × clamp(vertex color), the
usual alpha test
Effect joint
rotations
Status: DECODED
(effects1)
Now: Applied (row “Effect
element joints”)
Where:pd_effect.gdjoint_local
Effect
joints named after a joint the owner lacks
Status: INFERRED
(effects1)
Now: Such a joint (the shot
joints of flight slots; any named joint of a
shot, hit or ownerless instance) rides the
instance root. The hit (slot 2) is spawned with
no owner on the host, the net client and the
blast path alike (round 2)
Where:pd_effect.gdjoint_xform
Lead: the exe uses the
joint’s local matrix alone there (0x140188fd0);
which matrix the shot instance’s draw multiplies
it with is not followed
Looping
character effect root
Status: INFERRED (M)
(effects1)
Now: Slots 4 / 5 are rooted
at the caster’s matrix at their start and do not
follow a walking caster (named joints do ride
its bones)
Where:defense_look.gdstart_loop
Lead: whether 0x140259540
refreshes the instance matrix per tick
Point-sprite
size limit
Status: INFERRED
(effects1)
Now: 64 px on a 480-line
screen (Xbox setting only)
Where:pd_effect.gd SPRITE_LIMIT_PX
Lead: the console’s
point-size cap (render state, not in the
exe)
Point-sprite
colour envelope (raw_m[15])
Status: MISSING
(effects1)
Now: Sprites take the
element’s colour lerp and the linear weight; the
per-element envelope field is not applied
Where:pd_effect.gd_apply_sprite
Lead: the sprite update
loop past 0x1401a5940
Effect
cube-map textures
Status: MISSING
(effects1)
Now: The pack revision
gives the reflective effect shaders (second
texture of 05 / 07 / 08 and the test shader; 23
references in 23 effects) a 64 × 64 cube map;
pdconv does not export cube maps (listed as
skipped_textures in effect.json)
and our drawing does not sample a second
texture
Where:effectconv.py,
pd_effect.gd
Lead: export through
Esper-RE’s texture path (DDS cube) or bcdec,
then the shaders’ reflection term
Hit
slot of melee and blade records
Status: MISSING
(effects1)
Now: With the packs many
melee / blade effects have no slot 2 any more;
where the game plays a melee hit slot is not
followed
Where:melee_attack.gd
Lead: the melee hit routine
(step3 11-13) decoded: close-range
objects never reach the end routine with a hit
reason (they end on their counter, reason 0), so
the game spawns no slot 2, no end-routine hit
sound and no cue 6 / 7 for a melee / blade /
pursuit hit; a defense that stops one gives the
attacker reaction action 26. Rebuild: drop the
hit slot for close-range hits. Open: the reader
of the reaction row’s sound field (row + 6)
Card
eraser (record 0x5C / 0x5E / 0x60)
Status: DECODED
(skills3a: played)
Now:PdCardEraser.run after the damage
of every set-0 hit (shots and their kinds, close
range, auto), before the hit effect; skipped
with the hit effect for a type-1 attack on a
slot-31 holder; cue 0x5b; 0x4E = 2 cards stay
(button flash ticks), 0x4E = 1 cards pay
instead; the erased card goes on the used list;
slots reach a client in the snapshot. Open: rows
“Card eraser: button flash and client cue”,
“0x4E = 1 payment”
Where:card_eraser.gd
Card
eraser: button flash and client cue
Status: MISSING
(skills3a)
Now: A guarded card sets
Combatant.button_flash (16 ticks)
but nothing draws it; the erase cue 0x5b plays
on the host only (a net client sees the slot
empty, hears nothing)
Where:card_eraser.gd,
slot_hud.gd /
pd_hud.gd
Lead: the draw side of the
HUD button flash (0x1401509f0, kind 4); the game
plays the cue where the target is simulated, so
the client’s own body would need the event
0x4E =
1 payment instead of the erase
Status: INFERRED
(skills3a)
Now: The card pays aura
(and level) and stays; its use count is left
alone
Where:card_eraser.gd erase_slot
Lead: the pay routine
0x1402621f0 as called from the slot erase
0x140262fc0 (use count not followed)
Auto
cards (0x46 = 1): the erase guard
Status: DECODED
(skills3a: played)
Now: Template “auto_card”
(set 4, 0x46 = 1, read from the record’s raw
bytes); a press is refused (“auto card”); an
erase-type (3) set-0 hit on a holder releases
the first idle payable one (aura, use, release
effect slot 3 on the holder) and is cancelled
(PdCardEraser.erase_guard, in every
set-0 hit path before the damage)
Where:card_eraser.gd,
data_skill.gd
Lead: “can release” is our
can-pay test (INFERRED: the release 0x1402626d0
was not followed for an unpayable card)
Set 6
(cast the remembered attack)
Status: DECODED
(skills3a: played)
Now: Template “set6”: the
press casts the remembered attack’s catalog card
(0x54 = 0 Combatant.last_attack,
written at an attack’s release; 1
last_hit_attack, written by every
landed type-1 hit) with its profile, power and
objects, the set-6 card’s own cost; “nothing to
recall” while none. Both reset at a new round
(INFERRED: the game’s reset point not
followed)
Status: DECODED, played /
INFERRED (skills3a 11-13)
Now:Combatant.stance: 0x15 heals 1
health per 91 idle ticks, 0x16 takes no health
from hits (reaction, eraser and hit effect still
run); any stick input ends both
(Player.simulate). INFERRED: no
stance animation loop is played (the body stays
in its idle pose); a new cast or a reaction does
not end a stance; the CPU and the auto attack
path do not read the guard stance apart from
take_hit
Where:combatant.gd stance,
player.gd simulate
Lead: whether a reaction or
a cast ends them; the stance’s own
animation
Deck view
(status effect 20)
Status: MISSING (step3
11-13; still refused in skills3a)
Now: Not played (1
record)
Lead: state 0x18 and
routine 0x1401582f0 not followed
Hurt
voice and hit sound of a reaction
Status: MISSING (step3
11-13)
Now: Reactions play no
voice from the action row
Where:combatant.gd
Lead: decoded: row + 8 =
voice group (cue table at gsdata 0x33004, ten
per group, random); row + 6 = a sound id whose
reader is not found
Effect mesh roll
sign
Status: INFERRED
(effectlook)
Now: Counter-clockwise in
screen space after the Z mirror; unverified
Where:pd_effect.gd roll
Lead: a side-by-side with
footage of an asymmetric rolling mesh
Effect
material cull bit (0x80)
Status: INFERRED, narrowed
(effectfix)
Now: Culling follows the
element’s facing (raw_m[3] = 2
two-sided, else back faces culled with the
game’s clockwise front, triangle order
kept)
Where:pd_effect.gd_make_mesh
Lead: decoded for
pdeftbb01: its setup re-sets the
cull state after the material bind from the
element field (2.0 → none, else 0x901 = D3D11
cull back, FrontCounterClockwise 0), so the
material bit is overridden there; the other
effect shader setups not checked
Now: Local frame wraps over
the window, so the element repeats past its end
until the effect ends
Where:pd_effect.gdlocal_frame
Lead: the element update’s
loop branch. Contradiction: 145
meshes have a fractional raw_m[5]
(0.2 to 0.8), which a 0/1 flag does not explain
(FORMATS calls it a loop flag): read how the
update uses the value (a rate or a fraction of
the window?)
Now: Odd frames (counted
from the window start) use the blink color pair
through the color curve
Where:pd_effect.gdblinking
Lead: which frame counter
the exe tests. Contradiction:
361 meshes have raw_m[14] = 1, so
FORMATS’ “not used by meshes here” is wrong;
these meshes blink with our reading,
unverified
Capsule
look without converted data
Status: PLACEHOLDER
Now: Built-in see-through
bubble, only when
<data>/capsule/ is absent
(the decoded look is in
capsule_look.gd)
Where:orb.gd
Lead: run
capsuleconv.py; playtester to
switch the data dirs over
Capsule
glow quad (vkcolanim)
Status: INFERRED
(capsulelook)
Now: Decoded and drawn:
sat(t0·t1(flash u, uv2.y)·colour), own = type
colour added, others = white darkening
(zero/inv_src_color), flash u = (age − raw_f[1])
/ (raw_f[0] − raw_f[1]) from the capsule’s
appearance; culling from the material bind,
flash texture point-sampled, not touched by the
capsule’s near-camera fade
(capsulelook-fix) (FORMATS “Capsules”,
Glow). INFERRED: flash u clamped to 0..1 (no
sampler address state is set for t1 in the draw
or the bind), VS c0.x taken as 0, one counter
step per 1/30 s tick
Lead: t1 sampler address
mode (who last set states 0/1 of stage 1); VS
c0.x source
Capsule core
light
Status: INFERRED
(capsulelook-fix2)
Now: Core math decoded
(FORMATS “Capsules”, Core). Light values
decoded: light set 0 of the stage’s light script
(no stage selects another set; character draws
switch to set 1 and back), vectors in the game’s
(reversed) order, uploaded unchanged and applied
in the object’s own frame (raw model-space T, B,
N); the zeroed default record when the stage’s
chosen set is empty. Normal maps decoded: DXT1
(format 0x0C, uploaded as BC1), read as RGB ×2 −
1 with no z rebuilt and PS c0.w = 0.5; their
blue ≈ 0 is in the data, so the cores’ darkness
against the Xbox reference is the PC shader’s
own result (see FORMATS, “Why the cores shade”).
INFERRED: that nothing between the player pass
and the object draws switches the set (the
object list that calls 0x14021b840 was not
followed); that the matrix-stack top in c20 is
the drawn object’s matrix (capsuleview: VS
c15, the eye, decoded: FORMATS “Capsules”,
Core)
Lead: follow the
object-list draw that reaches 0x14021b840 and
confirm the slot pointers (0x14057f3d0..) hold
set 0 there
Capsule
core light without a converted stage light
Status: PLACEHOLDER
(capsulelook-fix2)
Now: Grey light 0.45 / sky
and ground 0.3
(CapsuleLook.UNCONVERTED_LIGHT)
when the stage has no light.json or with
--fitted-light; every converted
stage (st01–st08, st12–st17, st22–st26) has one
with a used set 0, so play never reaches it
Where:capsule_look.gd core_light
Lead: run
pdconv --layout for the stage (the
game always has its set)
Now: Corners at ±1 (a ±1
corner table in the exe’s data, not tied to the
QuadMap upload); keys interpolated linearly;
ramp and dot sampled linear, clamped; the clip w
the size is divided by taken as the view depth
in game units (the camera’s view/projection
carry no scale, not traced) — so the push is
scaled by 0.1 for our metres
(dustfix)
Where:capsule_form.gd
Lead: whether the dust’s
vertex shader reads the QuadMap (its upload is
now decoded, FORMATS.md “Effect particles”: UV
(0, 0) at the (−1, −1) corner, a vertical flip
against ours that a symmetric dot hides), the
animation sampler, the vkPSPRITE sampler states,
the camera’s projection set-up
Capsule
refill dust without converted data
Status: MISSING
(capsuleform)
Now: No dust when
<data>/capsule/form/ is
absent (the timing still applies)
Where:arsenal.gd,
capsule_form.gd
Lead: run
capsuleformconv.py
Capsule
take / move one-shot models and sounds
Status: MISSING
(capsuleform)
Now: Not drawn: system
entries 2 (take) / 3 (move) at the slot with a
random turn, cues 0x55 (back), 0x5c (move) of
sound bank 0 not played
Where:arsenal.gd
Lead: located (FORMATS
“Capsules”, Refill); export models 2–3 with
st00.dat records 2–3 (auraglow 11-14: the
pickup itself plays battle cue 0x56 from the
slot-fill routine 0x140261a00 (M): played since
the auraglow build (Orb.took, row
“Capsule pickup cue gate”); a pickup raises no
aura glow event)
Capsule
refill dust frame vs the take
Status: INFERRED
(capsuleform)
Now: Our arsenal sees the
take on the tick after it (the orb leaves its
group); frame = ticks since then − capsule
delay
Where:arsenal.gd
Lead: none needed unless
takes move into the arsenal
Capsule
near-camera fade: stipple bit positions, step
rate
Status: INFERRED
(effectfix)
Now: Decoded and played:
alpha −16 within 1 m of the camera / +16
outside, hidden at 0, stipple level alpha
>> 3 keeping level² of 1024 pixels, row
offset re-rolled per frame (FORMATS “Capsules”,
Near-camera fade, Stipple). INFERRED: the
generator’s rand() seed (fixed 1; the game’s is
its clock-seeded shared state) and one step per
1/30 s tick (the game steps once per drawn frame
of its loop)
Lead: the PC build’s draw
rate vs the 30 Hz logic tick
Capsule
UV scroll rate, st00 model order
Status: INFERRED
Now: Scroll and offsets
decoded (frame counter × element floats, FORMATS
“Capsules”, UV scroll); INFERRED: the counter
steps 30 times a second (it steps once per drawn
frame); models fill the table in file order
(counts match). Blend decoded per material
(capsulelook)
Where:capsuleconv.py,
capsule_look.gd FPS
Lead: the PC build’s draw
rate; the model-table fill loop
Capsule
depth and draw order
Status: INFERRED
(capsuledepth)
Now: Decoded: material
flags bit 3 picks the list (set: per-shader
lists drawn first with depth test + write, so
cores write depth; clear: a keyed list drawn
after with test on, write off: glow and shells).
Inferred: the tested byte is the material flags
(the queue reads it through a pointer whose
record layout is not traced), the key is camera
distance drawn far to near, the shells of one
capsule keep file order (Godot render priority
k), and the vkCAP04M4 core (0x10001, unblended
in the keyed list) draws test-only. Decoded: the
tree flush sets test + write once and the walk
loop does not reset it. Untraced: vkCAP04M4’s
own per-item draw function (it may set depth).
If it doesn’t, the core writes depth only if no
blended item precedes it in that tree’s order;
that order (key) is the inferred part.
Where:capsule_look.gd DRAW_LIST_BIT,
opaque_list
Lead: queue
0x140295616..0x1402957c7 (pointer at item
+0x54), the key at item +0x10 and the tree walk
order; trace vkCAP04M4’s per-item draw function
for depth calls; for vkCAP04M4 the per-item key
order within one capsule (shells/glow vs
core)
st03
white block on placed prop 4 (model 11)
Status: MISSING
(capsuledepth)
Now: A small 255-white
block on placed prop 4 (model 11) in the st03
head render, which the flat-colour render
attributes to surface 1 (m11_mat1, sbts). The
probe (corner/centroid in rect, no depth test)
has not yet tied the white pixels to specific
triangles, so the texel/vertex-shade values
behind them are unproven; the textures load and
turning the ramp shine off leaves it white
Lead: per-pixel probe
(point-in-triangle + depth, barycentric
UV/vc/texel at the white pixels ~x 528..535, y
110..134); dump the active material and uniforms
of that surface; compare to the PC build at that
view
Capsule
type colours for erase/environment (fallback
look)
Status: PLACEHOLDER
Now: Guesses
Where:orb.gd
TYPE_COLORS
Lead: goes with the
fallback once the data look is everywhere
Capsule look
Status: PLACEHOLDER only
without data (capsules, merged)
Now: Decoded models,
shaders, own/enemy model when
<data>/capsule/ exists; the
see-through bubble only without it (row “Capsule
look without converted data”)
Where:orb.gd,
capsule_look.gd
Lead: done; playtester to
switch the data dirs over
Capsule
waiting-state model (model 0)
Status: MISSING
(capsules)
Now: Not drawn while a slot
refills (its glow time (refill −
timer)·420/refill is decoded)
Where:capsule_look.gd
Lead: draw model 0 during
the refill
Capsule
type colours for erase/environment (old
look)
Status: PLACEHOLDER
Now: Guesses, used only by
the fallback look (same item as above)
Where:orb.gd
TYPE_COLORS
Lead: replaced by the
decoded look
Xbox
look: core combiner constants
(xboxlook)
Status: INFERRED
Now: The Xbox vkCAP02 math
(decoded) with the PC’s constant values:
renormalise 0.5, k offset 0.5, alpha 1/124,
light records. (xboxlook round 2)
Checked: the shader definition’s own constant
arrays are all zero and its stage/final constant
mappings point every used constant at run-time
constant registers 0, 2, 3, 6, 7 (FORMATS “Xbox
capsule shading”), so the values are uploaded by
the game, not stored with the shader
Where:xbox_look.gd
Lead: the Xbox code that
uploads the pixel constants 0, 2, 3, 6, 7 for
the vkCAP draw (default.xbe, not traced)
Xbox
look: capsule texture/blend states
(xboxlook)
Status: INFERRED
Now: Same as the PC draw
(material bind)
Where:capsule_look.gd
Lead: the Xbox material
bind’s render and texture-stage states
(default.xbe)
Xbox
fog render states (xboxlook)
Status: INFERRED
Now: The light script’s fog
call taken as the D3D fog states (mode,
start/end or density, colour); D3D fog formulas
on the view depth
Where:xbox_look.gd
Lead: the Xbox handler of
the fog call (default.xbe) and its
SetRenderState writes
Xbox
height fog (YFOG shaders)
(xboxlook)
Status: MISSING
Now: Not fogged in the Xbox
look
Where:xbox_look.gd FOGGED
Lead: the TMAT0 upload and
which fog range the Xbox uses for the height
coordinate
Xbox
fog on the cloud layers (SBTSCLOUD2/3)
(xboxlook round 2)
Status: DECODED
(skybright 11-15, H Xbox code)
Now:(skyglow
11-13) Drawn in the Xbox look with light
set 2’s fog (st02 exp 8e-5 black, st03 linear 1
602..12 822, st22 exp² 2e-5): the sky model
draws under set 2 and selecting a set applies
its fog, a partial fade, not set 0’s full fog.
(skybright 11-15) Decoded in
default.xbe (H): the sky draw 0xb1580 selects
set 2, the fog apply 0x89440 enables that set’s
fog for the whole sky model, and the cloud
vertex program’s fog coordinate is the view
depth
Where:xbox_look.gd CLOUD_FOGGED /
set_fog
Lead: none
Xbox
fog on effects with fog (PDEFTBB03, crash
shaders) (xboxlook)
Status: MISSING
Now: Not fogged (only stage
materials and capsule cores are)
Lead: route effect
materials through XboxLook.set_fog