The "we're off the rails" reinvestigation -- a 5-thread evidence workflow
(logs / PFX data / RES model lists / decomp / port audit) + the i860 firmware
decode, then surgical fixes. Every claim tiered; the wrong turns are on the
record in the KB alongside the corrections.
THE OLD-STYLE SPECIALFX ENGINE [T1, firmware-decoded from VREND.MNG]:
rebuilt scratchpad/i860dis.py (binutils opcode table); mapped the dispatch
(data+0xdd0c; sfx trigger 0xf040cda0, install 0xf040cdc0, step ~0xf0413698,
instance init 0xf04128d8); decoded the heat model EXACTLY: per 30Hz board
frame h *= cool_a; RGB_ch = h_old*(h_new*cook_ch - 0.25) + 0.25 (K=0.25 =
the ember floor @VA 0xF080; kill at h <= 1e-4 @0xF0A0); alpha fades cool_b
x RAW dt (PER-SECOND -- the x30 scaling made laser hits invisible); 7s cap;
y_off = the kill plane; "variance" is DEAD DATA (binary reads "variance",
INI authors "varience"). The 13 descriptors (PPCHit/LaserHit/MissileHit/
Chunks/Sparks/Fireball...) parse from BTDPL.INI and render via the BTPfx
layer -- heat bursts draw the FIERY sheet (brightness over fire), .PFX keeps
GRAYSCALE (authored colours: DNBOOM orange, DDAM gray).
HIT-PACKAGE CENSUS [T1, RES byte-verified]: ppchit=[8] lzrhit=[9] mghit=[7,11]
canhit=[11] mslhit=[10,12,1023] explode=[6]; all 8 mech death lists identical.
Effect routing corrected in BOTH consumers: <100 = specialfx, >=1000 = psfx.
"SILVER MIST x5" ROOT CAUSE: SHKWAVE.PFX (mslhit's 1023) authors maxIssue=5
relPeriod=0.2 rate=1 -- the ONE file where rate contradicts the window; the
emitter trusted rate -> 5 shells at exactly 1Hz. Emission rate now always
maxIssue/releasePeriod.
PER-ROUND DETONATIONS [T1 @004bef78]: every missile round spawns its own
ExplosionModelFile at ITS impact (hull + terrain) -- a volley ripples 12
fireballs like the demos; rack-tube launch spread [T3] (GUIDED rounds only --
the slot-0 deflection sent every AFC100 shell 3.4deg left/2.5deg down: the
phantom "4th gold beam"); replicant salvos detonate too (launcher index rides
the visual push). Missile damage bundles through the shooter's messmgr with
the launcher's subsystemID (mslhit fires at the consolidated point; the
binary's dedup CONFIRMED [T1 @0049b784] -- a workflow agent's per-record
claim REFUTED by direct decomp read).
DAMAGE-BAND SEMANTICS [T1]: MechDeathHandler fires the CURRENT band effect on
any damage rise (the binary's changed-flag coalescing) -- a mauled mech under
fire smokes/burns per hit; bands 3/4 are authored fire plumes.
THE RAM ECONOMY -- CLOSED (3 layers, measured live):
1. armor = POINTS (every zone: damageScale = 1/armorPoints, armor 50-140;
the [zone-armor] BT_DMG_LOG dump);
2. StaticBounce prices rams with authored moverMass ~1.3e6 -> ~59,000 pts
@10m/s; the binary dispatches it RAW but pod MP dropped it on the local
replicant (MECH.CPP:986 warns) -- ram damage was NETWORK-INERT; our
task-#47 replicant forwarding surfaced it as a one-shot. Port
normalizes x1e-3 to the point economy [T3];
3. contact EDGE (ramLastVictim/ramContactLinger): the binary's bounce made
separation implicit; our gait-derived velocity re-priced full rams at
60Hz (the respawn explode-loop). One bump = one hit; pressed = BLOCK.
PLUS: sfx size x0.5 + 1.25m visibility floor + hot-phase occlusion [T3];
particle pool 2048->8192 (missile traffic starved laser bursts -- the
"intermittent effects"); replicant beam aging (a lost beam-END record pinned
a stale beam on forever); UV-variant noise stamps (mask-safe mirror/swap);
no scroll on particle stamps (material-path only); BT_FX_TEST/fxshot.py
self-verification harness ([beam-draw]/[zone-armor]/[collide-tx] telemetry).
KB: rendering.md (specialfx engine + census + closeout), combat-damage.md
(ram economy + band semantics).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The "fireballs like the demo vids" arc, decomp/content-grounded end to end,
plus the live-play UX batch verified over the same sessions:
FIRESMOKE SHEET (the PFX fireball fix): every firesmokeN_scr_tex in BTFX.VMF
maps the SAME 64x64 tileable noise image bintA (variants differ only in SCROLL
rate) and firesmoke1_mtl colours it through the "fiery" ramp (0.3,0.1,0.1)->
(0.9,0.7,0.3). The particle layer now bakes ramp(lum(bintA)) as its sprite
colour (noise detail in alpha) and SCROLLS it at firesmoke1's authored rate
via a texture-transform; the port's radial soft-edge mask moved to a second
CLAMPed stage so the WRAPPED scroll rolls flame through the sprite without
scrolling the edge away. Old grit x radial bake kept as the no-BINTA fallback.
Impact hits, damage-band smoke and death booms all ride this layer.
AUTHORED TEXTURE SCROLL in the model path: the BMF TEXTURE records carry
SPECIAL " SCROLL u0 v0 du dv" (tag 0x2037); the draw path always supported
per-op scrolling (SetTextureScrolling) but the BGF loader never parsed it, so
every scrolling material rendered frozen. Wired TexRef -> MatInfo -> batch ->
L4TEXOP.doScroll: the flame cards (flamebig/fire5) now roll fire noise.
VERTEX-ALPHA EFFECT CARDS (the "twisted drill bit of fire" fix): FLAMEBIG's
verts carry authored float RGBA -- white-hot base (1.0,0.99,0.97) -> dark-red
tip fading to alpha -0.2 (the DPL clamp convention). The loader kept a flat
batch colour and drew it OPAQUE = a solid orange spike. Corpus sweep: exactly
14 shipped BGFs carry vertex alpha, ALL effect cards (flames, MUZFLASH,
EXDISK_A/B/C, TMST_A/B/C, beam models, DECLOUDS). Such batches now keep the
authored per-vertex gradient and route to the alpha-blend pass, unlit,
colour = texture x gradient, alpha = the vertex fade; sky objects excluded
(drawAsSky + alphaTest passes NEITHER pass filter -- DECLOUDS stays in the
sky pass). MUZFLASH/EXDISK render correctly for free when the muzzle-model
work lands.
WRECK DRESSING (the 1996 ExplosionScripts case-4 transcription): pieces spawn
HIDDEN and reveal 0.25s after the boom (the InstanceSwitch delay, behind the
dnboom flash); flamebig hangs over the pile, Y-BILLBOARDED at the camera
(SetOffsetYaw + a camera-pos getter -- the dpl_SetDCSReorientAxes analog);
the MakeDCSFall settle arms at the reveal with the two authored rates (hulk/
debris -0.025 t^2, fires -0.01 t^2 -- the flames ride above the sinking pile
and die with it at burial). EMPTY-PLACEHOLDER hulk guard: THRDBR.BGF is a
153-byte zero-geometry stub that "loads fine" -- vertex-count check now routes
it to the gendbr fallback (a Thor wreck was invisible). Hulk content census
recorded: AVADBR==MADDBR==VULDBR geometry (palette-only prefix diffs),
RAPDBR==SNDDBR==STIDBR byte-identical -- wreck variety is materials + the
dressing, not unique piles.
LIVE-PLAY BATCH: muzzle resolve uses the named segmentIndex (raw +0xdc read
was layout garbage); forward launch frame (authored MuzzleVelocity +Z vs the
mech's -Z facing); dock-bottom single window (gauge strip appended below the
world viewport, 1100x600 default, BT_DEV_GAUGES_WINDOW=1 restores the separate
window); portrait sec surface unrotated CW; ammo counters live via typed
bridges (BTAmmoBinCountPtr/BTAmmoBinFeeding/BTWeaponAmmoBin -- raw bin+0x180
and a hand-rolled link walk were garbage); fourth fire key ('4' = Pinky);
panel/arc probes de-aliased (%61 prime).
KB: rendering.md (vertex-alpha card family + scroll), combat-damage.md (hulk
census + THRDBR stub), gauges-hud.md (ammo bridges).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The interactive 2-node playtest wave -- every fix decomp-grounded and live-verified:
COLLISION ECONOMY (the ram one-shot): StaticBounce mutates worldLinearVelocity
per contact and ProcessCollisionList walks EVERY touched solid per frame; with
2007 terrain-as-solids the reflections compounded x4-x40 within one frame and a
walking bump one-shot a pristine mech for 112,375 pts (62-pt authentic economy).
Fix: frameEntryWorldVelocity restore per contact (damage always priced at the
real approach speed -- all the binary's physics ever saw); Mech::Reset zeroes
the mover motion (respawn = teleport); [collide-tx]/[mp-hdlr] telemetry.
Gotcha #16 (engine-facility drift class).
MISSILES: peer-visible salvos (the launcher record extension carries a salvo
counter + aim point; ForceUpdate actually enqueues it -- the dirty flag alone
never serialized), the authentic arc (authored MuzzleVelocity vector + the
Seeker's 200m/0.1/300 loft + gain-4 steering, decoded from @004beae4/@004bef78),
world-impact bursts (rounds detonate on cave geometry instead of phasing
through), contact-only damage (flight-cap expiry = fizzle, no more teleport
damage), live re-lead, and ballistic (unguided) shells for autocannons.
projweap's stale BTPushProjectile extern (the /FORCE signature trap, gotcha #6
corollary) crashed the Avatar's first AFC100 shot -- fixed + sweep rule recorded.
RADAR: two transcription bugs made the scope permanently empty -- FUN_0040b244
is the affine INVERSE (not a copy) and FUN_0040adec writes ONLY the 3x3 rotation
(never the translation); worldToView now Invert(view) built rotation-first.
CulturalIcons sorted out of the moving grid (the phantom red pips were map
props), visible-radius culls on all three draw passes, live pip verified at
|delta| x ppm px. Gotcha #17 (verify the FUN_ body, not its call shape).
WEAPON PANELS (the frozen-dial hunt): the binary's *(subsystem+0x40) means
FAILED -- the recon's 'operating' name was backwards, inverting the destroyed-X
lamps, the panel look, the children enable and the ready-lamp gate (which had
NEVER executed). Polarity chain corrected end-to-end (failedState, fed by real
damage saturation). Root cause of the freezes: MFD page-mode gating -- the dev
composite shows ALL pages at once, so off-page dials legitimately stopped; under
BT_DEV_GAUGES the 15 page-plane bits stay active (the exclusive secondary trio
untouched). The SEH gauge guard now names its kills; repaint-heal resets the
incremental arc after panel repaints; [panel]/[arc] probes added.
COMM/SCORE: MessageBoard LIVE (the engine already shipped the whole
Player__StatusMessage queue; wired the binary's one producer -- the kill branch,
victim's name, 6s -- plus the consumer bridge and a lazy source bind); MP DEATHS
counted via the observed-death tally (each node scores every pilot from locally
observed events, the same model as the KILLS credit) and the -2/-1 engine seed
clamped for display.
DEV UX: node-tagged window titles (-net port), gauge panel reworked (1320x480,
true 4:3 MFD cells, the portrait secondary UNROTATED upright, linear filtering,
BT_GAUGE_SCALE), fixed close spawns via BT_SPAWN_XZ, Boreas flies an Avatar
(first second-chassis live outing).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
TASK #17 -- AUTHENTIC RADAR SYMBOLOGY (the cross-blip stand-in retired).
The recon proved the 'missing pip raster set' never existed as BT game code:
pips are the ENGINE's L4GaugeImage vector-stroke system (T0 source in tree,
L4GAUIMA.cpp == FUN_0046f0c0 line-for-line), the 'pip table' is L4Warehouse::
gaugeImageBin keyed by Entity::resourceID, and BTL4.RES ships 110 type-0x12
shapes (every mech/vehicle/building/tree). The six btl4rdr stubs are wired
to the real facilities: contacts draw their authentic model silhouettes with
LOD selection, the target gets the binary's 4px-inflated box highlight, and
player-name labels resolve via Mission::GetSmallNameBitmap (the prebuilt
64x16 egg rasters keyed by the player's bitmapindex). DECODED: the invented
'VideoObject' was Entity::owningPlayer all along (+0x190; nameID =
playerBitmapIndex@0x1E0, target = BTPlayer::objectiveMech@0x284 -- new
bridge BTPlayerObjectiveMechOf); the 'LabelledEntity' class is Landmark
(cultural.h; label path dormant -- no landmark content ships, no runtime
landmarkID writer). The L4GREND BT_DEV_GAUGES warehouse guard is removed
(its AV had a different culprit, below); resource type 18 corrected to
GaugeImageStream in decomp-reference (was mislabeled 'ModelList').
TASK #14 -- the Myomers ODR duplicate ELIMINATED: the powersub.cpp/hpp
'Myomers' (classID 0xBC3 -- actually Sensor) is retired whole; it duplicated
?DefaultData@Myomers@@ against the real class (dumpbin-verified) and /FORCE
picked the winner by link order. The real Myomers (0xBC6) now chains
PoweredSubsystem's handler set (ids 4-8) and publishes its SEVEN binary
attributes (@00511588: SpeedEffect/Current/Recommended/Min/MaxSeekVoltage-
Index/SeekVoltage/OutputVoltage) -- the old empty unchained index starved
the Myomer engineering panel of every resolve.
TASK #16 -- attribute parity: MechWeapon publishes the FULL binary table
@0x511890 (11 entries; ids renumbered to binary truth -- the port aliases
had squatted the binary's DistanceToTarget/TargetWithinRange ids; the
streamed TriggerState 0x13 binding unchanged). Binary names resolved two
TODO members: pipState -> estimatedReadyTime (attr 0x1A), and the EXT-model
flag is the binary's RearFiring (0x1B). ThermalSight LightState published.
HUD (offset conflicts) + missile-side tables (id encoding suspect)
documented for a re-dump instead of publishing blind.
THE CRASH THIS EXPOSED [T2, cdb-verified]: gotcha #11's dense-table gap
is a LATENT AV, not a guaranteed one -- the old table's 0x0D..0x12 gap
(task #5) survived on heap luck; the renumber reshuffled allocations and
Find() AV'd on a garbage entryName in WeaponCluster's PercentDone resolve.
Fixed with five named PAD entries (the mech.cpp attrPad idiom) + a
static_assert locking the pad base to PoweredSubsystem::NextAttributeID.
Gotcha #11 amended with the proof.
TASK #15 -- stale-ledger sweep: GAUGE_COMPOSITE ('composite not yet built',
Reservoir shadow, PlayerStatus/vehicleSubSystems 'remaining', valve-dormant-
until-0xBD3, sensor guard, the superseded 'Heat MFD near-static' reframe --
all banner-corrected), gauges-hud frontmatter, L4VB16 + powersub comments.
Verified live: 50/50 config attribute bindings resolve, 0 NULLs, 0 parse
skips, mech spawns and simulates 31/31 subsystems, no cross-blip fallbacks,
the pip cache fills through entity registration without the old guard.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The pilot's third heat tool: Condenser::MoveValve (id 4, the Condenser handler
table @0x50E52C -- PE-verified, exactly one entry) registered via the
order-proof function-local accessor and GUARDED by the real FUN_004ac9c8
(task #12's BTPlayerRoleLocksAdvanced rookie-role bridge -- the old deferral
note blamed the 0xBD3 messmgr, a stale misattribution, swept). Each press
cycles the valve 1 -> 5 -> 50 -> 0 -> 1 and RecomputeCondenserValves re-shares
the fixed total coolant flow across all six condensers -- opening one valve
wide starves the others: the pilot's cooling-triage lever.
Desktop: 'C' cycles the selected condenser's valve (BT_VALVE_SLOT picks the
roster slot; default Condenser1). BT_VALVE_TEST scripts a headless press.
Verified live: press -> "[valve] condenser#1 valveState=5 flow=0.5 (total=10)"
-- exactly the authored redistribution math (5/(5+5x1)), through the real
role gate.
The heat-management triad is now complete and all-authentic:
G + fire buttons = regroup weapons across buttons (task #6)
F5-F8 / F9 = reassign weapon generators / mode (task #12)
C = coolant valve triage (task #13)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The pilot's second heat-management tool: the PoweredSubsystem message table
@0x50F4EC (ids 4-8) registered and implemented -- SelectGeneratorA-D
(@004b099c..@004b0a74: FindGeneratorByNumber roster walk on generatorNumber
@0x1E0 -> AttachToVoltageSource re-tap with tap accounting -> modeAlarm
Connected) and ToggleGeneratorMode (@004b0abc: Manual -> Auto -> detach+Manual
cycle). Weapons inherit via the MechWeapon handler chain. Desktop: F5-F8
assign the selected weapon (BT_CONFIG_SLOT) to Generator A-D, F9 toggles
reconnect mode; BT_GENSEL_TEST scripts a headless verify.
Verified live end-to-end: dispatch -> handler -> re-tap ("PPC_1 ->
GeneratorD (tapped)") -> the charging I^2R physically moved (GenD cold ~90K
baseline -> ~1570K carrying the PPC; GenA relieved), stable over a sustained
autofire soak with thermal-breaker trips.
FOUR defects found and fixed on the way [all T1/T2]:
- THE e17 HEAT EXPLOSION: Generator::SourceLevel misread *(this[0x38]+0x158)
as linkedSinks->heatEnergy; it is the engine-base DamageZone @0xE0 ->
damageLevel [0..1] (the same named-member pattern as the bank radiator's
zone read). A breaker-restarting generator emitted (1 - 4e8) x 10000
volts; squared through the customers' I^2R feed, one restart blew the
whole thermal network to e17. Authentic: a damaged generator yields
proportionally less voltage.
- FUN_004ac9c8 is NOT "IsDamaged": raw body = owner -> mech+0x190 player ->
roleClassIndex(+0x274) == 0 -- the ROOKIE-role lockout for advanced
cockpit systems. New bridge BTPlayerRoleLocksAdvanced (NULL player =
unlocked [T3]; bring-up role 2 = unlocked). The old stand-in gated the
handlers off permanently (healthy subsystems have simulationState==1).
- MESSAGE_ENTRY tables must be FUNCTION-LOCAL statics inside the accessor:
as namespace-scope arrays they are read by other TUs' static-init chains
before their own initializers run -- Build copies zeros and every id in
the table is silently dropped (ids 9/10 only worked by TU-order luck;
both tables relocated; gotcha recorded, reconstruction-gotchas #9).
- The AutoConnect hunt scanned GetSegment() -- the raw @004b0bd0 walks the
subsystem ROSTER (+0x124/+0x128); EntitySegments were being cast to
Subsystems (the classic +0x128 gotcha).
Also: the dense handler table's GAP slots (skipped ids) are uninitialized
heap -- the name-based Find strcmp-walks them and AVs; diagnostic probes use
the id-based Find only (the 1995 binary's own tables carry the same holes).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
TASK #6 -- the pod's in-cockpit weapon regrouping, fully reconstructed and
live-verified (hold-configure -> tap fire button -> toggle membership ->
release-commit):
- MechWeapon handlers id 9 ConfigureMappables @004b9550 / id 10 ChooseButton
@004b95b8 (binary table @0x511860; the old "Myomers fns" mislabel swept)
+ the GetMessageHandlers() accessor chain through Emitter/ProjectileWeapon/
MissileLauncher/GAUSS/PPC (empty per-class sets swallowed dispatch).
- Mapper vtable truth: +0x38 EnterConfiguration / +0x3C ExitConfiguration /
+0x40,+0x44 AddOrErase evt/dir. No "secondary vtable @0050f498", no
"CreateTemporaryEventMappings" virtual (RP-name drift; swept incl CLASSMAP).
L4 Enter/Exit rebuilt complete: StartMappableButtonsConfigure (the
NonMapping 0x10000 <-> Mapping 0x8000 mode flip + the gauge's active-weapon
latch) + the held-button re-arm + the 4 fire-button temp maps; only the RIO
mapper implements the toggle (Thrustmaster no-ops = can't regroup).
- MechSubsystem +0xE8/+0xEC corrected to controlDestination/controlMessageID
(ex hostEntity/subsystemId2 mislabel); MechWeapon ctor defaults the
destination to &fireImpulse (@004b99a8).
- ConfigMapGauge state loop reconstructed (PE-recovered DAT_00518eb4 table;
buttonGroup GetMapState sampler -- the "needs ModeManager" guard rationale
was wrong). Finding [T1]: the shipped binary NEVER enables this gauge (no
SetColor caller) -- authentically dormant; BT_CONFIGMAP=1 is the dev enable.
- Dev harness: HOLD 'G' opens the session (BT_CONFIG_SLOT picks the weapon);
BT_CONFIG_TEST scripts a headless verify. @004afbc4 corrected to its real
Fail-trap body (the guessed AddOrErase(NULL) body would corrupt groups).
- Bonus [T1]: the binary MechWeapon ATTRIBUTE table has ELEVEN entries
(PercentDone..WeaponState) -- "TriggerState is the only one" was wrong.
TASK #11 -- the user-reported live-play regressions (both real bugs):
- PHANTOM FIRE: the mech4 bring-up shot block painted an explosion at the
victim on its OWN 0.3s cadence whenever fire was held -- desynced once the
authentic recharges landed. Retired (the real impact visual flows from
each discharge via the messmgr SubmitExplosion).
- WEAPON BRICKING (the "cuts out" bug): the Loading->Loaded snap window
(+-0.01 around seekV) assumes the pod's LOCKED 60 fps; one port dt-spike
jumps it, the byte-verified >1.0 clamp (_DAT_004ba830=0) zeroes readiness,
and the weapon sticks in Loading at level ~10000 forever (observed live).
Fixed with pod-frame (1/60s) sub-stepping of the binary's own Loading tick
-- also fixes the I^2R integral over-heating generators on big steps.
NEW GOTCHA CLASS recorded: reconstruction-gotchas #12 (frame-pacing trap).
- Duty-cycle measured (max-rate autofire): generators equilibrate in the
degradation band (recharge stretch ~3.5-4.4x), the FailureHeat breaker
never trips solo, PPCs sustain ~11.8-damage full-charge fire. The slow
kill pacing and recharge gaps are the AUTHENTIC heat economy; the pilot's
counters (this regrouping UI, generator reassignment, coolant valves) are
the reconstruction queue.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Task #10 set out to fix "scrambled linked-sink routing"; a [heat-link] attach
log proved the routing was NEVER scrambled (every subsystem links its authored
condenser exactly; the "pools in Condenser1" read was the diagnostic-sampler
aliasing trap). The real defect: emitter.cpp's local FUN_00417ab4 stub
returned NULL, so the whole electrical model was inert and the E7 force-charge
recharged every emitter in ONE frame (~0.3s cycle, 1501 fires/90s, ~1.7e9
heat/s -- the "runaway").
Landed authentically [T1: disasm + byte-verified constants]:
- Emitter ctor @004bb120: seekVoltage = authored fraction x generator
ratedVoltage (10000); EC = energyTotal/(seekV^2 x 0.5); voltageScale@0x310 =
(RechargeRate / -ln(1 - 1e-4 x seekV)) / EC -- charge reaches seekV[rec] in
EXACTLY the authored RechargeRate (PPC 5s, ERL 4s, SRM 3s, ERM 2s) cold.
Owner-flags ctor gate (the usual gotcha; the this-flags read never armed).
- PoweredSubsystem::ChargeTimeScale (@004b0d50, was a =1.0 stub): voltageScale
x (1 + thermalResistivity x srcTempRise) -- hot generators charge slower.
("voltageScale is never read back" was wrong; corrected + swept.)
- TrackSeekVoltage @004ba838: charging I^2R -> the GENERATOR's pendingHeat
(~3.5e8/full PPC charge) -- generators self-heat, conduct to their authored
condensers, and throttle further charging. The feedback economy closes.
- FailureHeat consumers found: this+0x184 == 2 gates BOTH weapon families
(@004baa88 emitter: reset firing + hold charge 0; @004bbd36 ballistic:
recoil=rechargeRate + alarm 7). Emitter::GetFaultState un-stubbed.
- ProjectileWeaponSimulation @004bbd04 opens with call 0x4b0bd0 (disasm) --
launchers now run the powered/heat step (their firing heat previously
accumulated in pendingHeat forever).
- heat.cpp: per-instance BT_HEAT_LOG census (the old shared-static 1-Hz
sampler aliased); [heat-link] attach log; the stolen-else Verify restored.
Verified live (120s max-rate autofire): PPC ~470-500 (was 55,000), bank
plateaus ~600 and sheds to ambient, generators 1100-1400, ERMLaser
self-regulates at the authored 2000 failure threshold (shutdown-cool-resume).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Emitters: heatPortion closed form = heatCostToFire x 1e7 x (charge/seekV)^2
(PPC 1.1e8 -> +632K on its own sink); the missing projectile/missile heat
adds (raw -- resources author pre-scaled 1e7 units).
- The bank's AMBIENT RADIATOR @4ae73c reconstructed (the system's ONLY heat
exit; conductance x 0.1 x HeatSinkCount -- _DAT_004ae974 float80 = 0.1) +
the link-attach guard corrected (skip = the 0xBBE bank, not Condenser; the
inversion blocked condenser->bank links and closed the system).
- Constant corrections (all byte-verified float80s): coolant epsilons
0.0025/0.003/1e-4 (was a single 1e-4 serving three sites);
CoolantCapacityScale 0.05 (was 1.738).
- Verified: heat flows + exits; max-rate autofire overheats weapons into the
authentic heatLoad range-cutout (thermal spam unsustainable by design).
- KNOWN REMAINING: linked-sink routing scrambled (heat pools in Condenser1;
authored map says PPC->C4/C6) -- the heat-stream offset audit, filed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three root causes, all fixed:
1. kDamageScale was 1.0 -- _DAT_004bafbc is an x87 float80 = 1e-7, cancelling
the ctor's x1e7: damagePortion = authored DamageAmount x (charge/seekV)^2
(closed form at fire time; madcat AC=25/LRM=50/ERLL=6, bhk1 PPC=12/SRM=35).
The observed "0.25" was the degenerate EC=1 fallback, never authored data.
2. CheckFireEdge NaN latch: TriggerState carries ControlsButton INTS; the
release value (-65) is a negative NaN. The binary's x87 unordered compare
read it as "released"; IEEE-correct float compares latched the edge
detector shut after the first release -- the reason the emitter discharge
chain NEVER fired in-game. Fixed with bit-pattern sign compares.
3. The weapon-side submission LIVE: Emitter::FireWeapon fills damageData
(amount + damageForce=target-muzzle [the gyro directional-bounce feed] +
impact) -> MechWeapon::SendDamageMessage (@004b9728 real body) ->
messmgr consolidation with per-weapon records + explosion bundling.
The mech4 bring-up damage block + flat kShotDamage retired to diag hooks.
Bonus: LODReuseHysteresis 0.82 -> 0.33 (double misread). Zone model
verified byte-exact (no change). Solo end-to-end: 5-record volleys (2 PPC
+ 3 ERML with authored amounts), per-weapon zone granularity, explosions,
kill. Heat stays bring-up scale pending the heat-calibration audit [T3].
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ConsolidateAndSendDamage @0049b784 fully implemented: the beam path submits
into AddDamageMessage; each messmgr tick builds ONE Entity::
TakeDamageStreamMessage (id 0x13; wire-verified 0x34+4+Nx12) with the
common impact record + appended {type, amount, subsystemID} entries and
Dispatches it at the victim -- the T0 handler (ENTITY.cpp:817) re-splits
into per-record TakeDamage; replicants reroute cross-pod. Real
ResolveExplosionID (firing weapon's +0x3E4, guarded) + SubmitExplosion via
the Explosion::Make port; TWO chain-purge bugs fixed (the "iterator dtor
clears it" assumption was wrong -- records re-applied every tick, observed
1->2->3->4 double-counting). @004b9728 identity corrected (SendDamageMessage,
not DrawWeaponPip). Verified solo: 25 clean single-record consolidations ->
kill -> death transition. Residue: weapon-side submission awaits the
damage-economy reconciliation (authored 0.25-scale vs kShotDamage=12).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The recovered system: fire channels = LBE4ControlsManager buttonGroups
(0x40/0x45/0x46/0x47); default groups = the per-mech type-6 controls-map
resource in BTL4.RES, installed by the T0 CreateStreamedMappings the port
already called -- it needed only the TriggerState attribute (id 0x13 PINNED
to the binary value; fireImpulse@0x31C is the binary's TriggerState) and an
input feed. Keyboard/harness now push press/release edges into the button
groups; the gBT*Trigger bypasses, per-type keyboard split and 1,0 pulse
hack are retired -- weapons sharing a button fire TOGETHER (madcat Trigger
= 4 weapons). Myomers @4b9550/@4b95b8 misattribution corrected (they are
MechWeapon ConfigureMappables/ChooseButton). Verified 2-node: kill through
the authentic chain (12 hits vs ~36 pre-groups). Config-mode session
(regrouping UI) = the remaining stage, KB-scoped.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two size errors (animationPrefix char[12] vs 4 bytes; a FICTITIOUS 64-byte
skeletonName block -- the record has no skl name) skewed every member after
word 0x10 by +2/+15 words and pushed the fieldXX block past the record end.
Full writer map recovered (FUN_004a2da8 + Mover FUN_004238bc + Reticle
FUN_00435ac8 chains) + runtime reader map (ctor pass @0x4a24d1-0x4a2968,
capstone-verified). Struct rewritten as the exact overlay (static_assert-
locked); all Pass-2 consumers renamed to real fields; retired: ctor +
replicant + master-gait floor-25 hacks, the throttle guard, the fictitious
classID stamp. Verified: accel=30 superStop=10 throttleAdj=1 (was ~1/OOB/0),
deadbands byte-identical via named fields, full walk->run lifecycle clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The binding was in the zone ctor all along: Ghidra dropped the two arg
pushes @0049d0e1 (Slot::AddImplementation(subsystemArray[streamedIndex])),
making it read as a bare Resolve(). DZSlot stand-in -> engine SlotOf<T>;
SendSubsystemDamage rewritten to the recovered @0049c9a8 body (allotment
into the subsystem's OWN private zone; vital -> graphicAlarm 9); CriticalHit
-> real ApplyDamageAndMeasure; parentArtifactZone.Add revived (LOD damage
averaging); videoObjectFlag renamed vitalSubsystem (+0xE4). BT_CRIT_PROBE
diag added. Verified: 66 plugs bound/mech; probe-destroyed zone -> crits
damaged/DESTROYED, statusAlarm + destroyed-skin chain fire; MP kill + solo
un-regressed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
25-hit MP autofire trace: kick -> damped oscillation -> settle, peaks well
inside clamps, no NaN/drift; measured Y:X frequency ratio 7.75x matches
sqrt(springK.y/springK.x) exactly.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Community reports said the camera looked too far back and the cockpit frames
were never this revealed; the user's hypothesis (our projection widens the
view) was exactly right. The original Tesla projection call, preserved in a
comment at DPLRenderer::SetView, is
dpl_SetViewProjection(view, -1, -aspect, +1, +aspect, 1/tan(viewangle/2))
i.e. BTDPL.INI's viewangle=60.0 is the HORIZONTAL field of view with the
vertical derived by aspect: the pod's frustum was 60 x 46.8 deg at 4:3. The
port fed the 60 to D3DXMatrixPerspectiveFovRH as the VERTICAL fov -> 75 x 60
at 4:3 (worse on widescreen): the world rendered ~25% farther away and far
more canopy was visible than the pod ever showed.
Fix: BTFovYFromHorizontal(viewangle, aspect) at all five projection sites
(main, per-frame resize, sky pass, WM_SIZE rebuild, fallback) -- the
horizontal field stays the authentic 60 deg at any window shape. The aim
boresight reads the live projection (gBTAimP22) and adapts; combat
re-verified (kill chain, no NaN). Cockpits now show the frames at the
pod's framing: viewports dominant, frames at the periphery.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The user caught the inversion (blocked viewports, see-through frames); the
answer came from the rasterizer board itself: an i860 disassembly of the
VREND.MNG firmware decoded the damageize handler (@0xf040f6f8) -- cmd 0x20
(vr_damage_action) writes the dpl_Punchize token triple onto the punch
geogroup's first three geometry chunks IN FILE ORDER:
chunk1 = the detailed window-aperture shapes -> an invisible per-pixel MASK
chunk2 = the coarse hull -> the ONLY colour-drawn geometry (where not masked)
chunk3 = byte-identical hull twin -> the damage-reset record, never visible
Visible canopy = hull-minus-apertures. Both earlier readings were wrong ways
around the same three chunks: dropping all punch left floating fragments (no
hull); hiding the duplicate pair hid the HULL and drew the MASK -- the exact
inversion observed.
Port: bgfload tags per punch patch {non-paired pmesh = mask (role 1),
first twin = hull (role 2), second twin = skip}; L4D3D DrawMesh runs
mask->stencil-1 (no colour/z-write, z-tested), hull->stencil-NOTEQUAL,
mask->stencil-0; device D24X8 -> D24S8 + stencil clear. The cut never
touches z, so later-drawn entities show through the apertures.
BT_COP_PLATES=1 disables (diag).
In-game, all 8 mechs on pure defaults: connected dark frames with clear
viewports -- thor (the footage mech) shows the central viewport + surrounding
frame exactly as filmed. Combat regression clean (kill + no NaN).
Arena textured punch uses the same firmware kit; the black-texel alpha path
there is a working approximation, flagged in the KB.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Resolved the non-Madcat blocked views. Rendering every *_COP shell offline
from its mech's AUTHORED eye (SKL [jointeye] translation -- incl. the Thor's
+1.13 offset cockpit) with faces coloured by punch flag showed the answer
plainly: the PUNCH-tagged patches sit exactly over the window apertures in
all 8 mechs. On these texture-less black-diffuse shells dpl_Punchize renders
the tagged geogroup as holes -- the punch patches are the transparent
WINDSHIELDS, and the visible frame is only the non-punch geometry.
The loader now drops the punch batches of _cop meshes (bgfload finish();
BT_COP_KEEPPUNCH=1 keeps them for diagnosis). In-game sweep on pure defaults:
- madcat: the dark dome frame (matches gameplay footage)
- thor / owens / sunder: authentic brace/dash/pillar remnants
- bhk1 / loki / vulture / avatar: authentically FRAMELESS (all-punch
canopies -- the pod showed no frame for those mechs)
(An early-session test that "disproved" drop-punch ran with the broken
pre-inverse eye; its conclusion was wrong and is corrected in the KB.)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Re-disassembled the unexported gap 0x4b2980-0x4b2d8b byte-exact (capstone over
section_dump.txt) and reconstructed Gyroscope::ApplyDamageResponse:
- zero-damage + Collision(type 0) no-op; hit direction = Damage::damageForce
or a RANDOM horizontal fallback when ~zero (per-component sign roll then
value roll -- binary-legal until senders fill damageForce)
- direction rotated by the yaw-only torso-twist frame (placeRot=(0,twist,0)
-> euler->matrix, the FUN_00408744 raw row-dot reproduced verbatim)
- per-type scaling amount / damageMultiplier[type] * damageResponse[type]
.{trans,pitchRoll,yaw,vibration}; Explosive alone multiplies burstCount;
each clamped at 1.3f
- four kicks: impulse(trans, dir), torque(pitchRoll, dir), impulse(vibration,
vibrationDirection@0x390) and verticalImpulse(yaw, dir)
Layout corrections (byte-verified): gyro+0x390 is ONE Vector3D vibration axis
init (0,1,0) -- was mis-split as scalars animationOffset/Scale/Phase; mech
+0x5c4 is a FLOAT rumble-period countdown (gyroRumbleTimer) -- was mis-typed
int clipLoadGuard.
Call sites wired (binary-gated):
- take-damage hub: mech.cpp TakeDamageMessageHandler, FIRST action (@0x4a0264
in hub FUN_004a0230) -- an invalid-zone hit still bounces
- crushable-icon crunch: mech4.cpp, torque 0.4 along the bounce delta-v + up
impulse 0.2 (@4aa81e/@4aa86c)
- firing recoil: projweap.cpp FireWeapon, damage>3 -> impulse (0,0.6,-1.5)
x damage/16 (@4bc136-4bc19c) via the new Mech::GetGyroSubsystem accessor
DEFERRED (recorded in KB with the byte recipe): the alternate-gait engage
jolt + 0.4s rumble (@4aa158-4aa365) -- its gates are [T3]-flagged and it
mutates the #49/#50-stabilized gait machine.
Verified live: [gyro-dmg] fires per hit with correct type scaling; eyePosition
shows the damped bounce (ramp to ~0.015u, oscillating Y, decay to zero); kill
chain un-regressed; no NaN.
Also recorded in KB: the OFFSET-COCKPIT verification -- Thor's authored
jointeye tranx=+1.13 matches its canopy X-center exactly AND our in-game eye
reproduces it to the hundredth (the placement model is correct per-mech); and
the playable roster is 8 models (strider/raptor/firstrtr/blkjack have assets
but no model resource -- BT_FORCE_MODEL with those names = NULL-resource AV,
cdb-verified, not a code bug).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The Gyroscope subsystem is LIVE again (was a stub since the NaN revert).
Reconstructed byte-exact from @004b3778 / @004b2ec0 / @004b30ec / @004b33e0 /
@004b34ec + the byte-verified dispatch in the unexported Mech performance
FUN_004a9b5c (calls @0x4aaf74/0x4aaf83):
- ctor field map corrected: springConstant@0x1E8 / dampingConstant@0x1F4 were
mislabelled (the old 'eyeOrientation = r->springConstant' poison line), the
0x254-0x2B3 block (externalPitchPtr, work matrix, placement) was missing so
everything after was mis-offset, and clamps/accumulators were 0xCD fill --
every field now initialised per the binary; 33 offsets static_assert-locked,
sizeof == 0x3D0 exact
- integrators: state-minus-target displacements, componentwise damping
OVERWRITE (FUN_004086d0 is a component multiply, not a cross), position step
without dt, clamp to [posSpring, negSpring]; IntegrateBody's X/Z-crossed
force + integration terms
- writers: WriteMechJoint drives ONE node ('jointeye', type 5): TRANSLATION =
eyePosition spring, ROTATION = bodyOrientation spring; WriteEyeJoint is a
multiplicative sway attenuator on 'jointlocal' (post-anim only)
- dispatch moved out of GyroscopeSimulation into the Mech performance tail
(GyroFrameJointWrite bridge, mech4.cpp) with the binary gates; gyro<->torso
pitch link wired (gyro+0x258 = &torso currentTwist) via complete-type-TU
bridges, retiring the SubProxy::linkTarget landmine
- Mech ctor: deathAnimationLatched/legResetLatch were never initialised --
0xCDCDCDCD gated the writers off silently
Runtime-verified: joints resolve (both BallTranslation), Performance installs,
WriteMechJoint finite from frame 0 (no NaN), Madcat cockpit un-regressed,
combat targeting healthy. Empirical: spring targets are SYMMETRIC -> eye
equilibrium (0,0,0), clamps +/-0.1-0.15u -- the gyro is the hit-BOUNCE
mechanism, not a steady eye offset (hypothesis disproven, recorded in KB).
Pending (task #56 tail): the damage->gyro fan-out FUN_004b2980 (unexported
gap) so hits actually kick the springs; the mech+0x3F0 overspeed sway model;
the torso-pitch EyepointRotation writer (FUN_004b66b4) + glance-look states.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The first-person cockpit now shows by default and matches gameplay footage
(verified on the Madcat). Two decomp-verified reconstructions:
EYEPOINT (FUN_004579a8 + caller part_014.c:5525, FUN_004c22c4):
- eye offset = siteeyepoint GetBaseOffset (not GetSegmentToEntity, not the
upright hack), parented on the PARENT segment DCS (btl4vid.cpp)
- live view = affine INVERSE of the eye world matrix (L4VIDRND Execute) --
replaces the hand-rolled LookAtRH whose +Z-forward/+Y-up row guess aimed
some mechs into the canopy; combine order fixed to baseOffset * R
- chase camera basis + aim boresight re-expressed for the inverse convention
(row2 = back); boresight now derives from the view matrix (the stale rows
aimed the pick ray at the sky = no target, no discharge)
CANOPY (*_COP.BGF, dpl_Punchize geogroups):
- every shell is an open strut lattice (38-59% boundary edges, all 12 mechs);
rendered single-sided with per-face INWARD winding (l/r torso patches are
mirrored, so no global winding works) the struts are the dark frame and the
openings show the world
- frame colour = the 'softer' ramp near its dark end (texture-less unlit ramp;
BLXSKIN.BMF has no texture -- the old "punch texels" theory was wrong)
- scoped by filename (meshIsCop); wrong drop-punch + view-cut paths removed
- BT_FORCE_MODEL=<name> forces the player mech for per-mech bring-up
KB: new context/cockpit-view.md topic; punch split corrected in bgf-format;
gotchas 14 (LookAt axis guess) + 15 (per-patch edge namespaces); env gates +
eye/punch addresses in decomp-reference.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Scouted the "-net node render glitch". Two prior framings were wrong and are
corrected in the KB: it is NOT device contention (retracted earlier), and it is
NOT a from-load glitch -- the trigger is a RESPAWN (user correction). It's a
post-respawn render-rebuild fault on the MP path.
Could not reproduce in 9 automated 2-node loads + a forced death/respawn (all
clean). Debug uninit is a deterministic 0xCDCDCDCD, so the intermittency implies
a race -- and the task #52 respawn fixes (LoadMission re-entry, viewpoint relink,
death-latch) plausibly already closed it. Parked as WATCH; re-open if it recurs.
Tooling (general, off by default): BT_SHOT=<path> dumps a per-instance backbuffer
PNG from the world-render path (L4VIDEO.cpp) every 90 frames -- lets us capture a
specific node's frame without foregrounding the window. Used with
scratchpad/hunt53.sh (N-run divergence hunter) to verify a fix if the glitch
returns.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The death/respawn "blue whirlwind" (tsphere) now matches the original cabinet
photo (capture.png): a smooth spinning lavender vortex with a bright core.
Root cause of the long-standing "radial spokes" artifact was NOT the warp code
but a general engine bug: L4D3D::SetTextureScrolling computed its texture-matrix
offset as scrollDelta * absolute_time, which grows unbounded and collapses UV
float precision -> a smooth scrolled cloud shatters into grainy radial steps.
Wrapped with fmodf(..., 1.0f) (identical under REPEAT tiling, full precision).
This also cleans the scrolling bexp beam grit and any other SCROLL material.
Visual reconstruction (verified against the real 45-vtx TSPHERE.BGF bicone,
offline-rasterized then ported):
- view ON-AXIS (eye centred on the throat) + spin in place -> concentric rings
(decomp FUN_00453dc4 does spin-about-local-Z + submit; the port had stubbed it)
- bintA cloud through a WIDE lavender ramp at full contrast (drawn as SKY);
no geometry "bands", no log-polar twist, no off-axis tornado (all discarded)
- tessellate the 12-facet bicone smooth; isotropic + trilinear; ramp baked into
the texture and drawn SELECTARG1(TEXTURE) to avoid double-tinting
Env knobs (BT_WARP_*) default to the verified values; BT_WARP_SELFTEST/SELFSHOT
are an off-by-default visual-verification harness (backbuffer frame dump).
Docs: new context/translocation-warp.md (geometry/material/visual/lifecycle/env);
reconstruction-gotchas.md gains the accumulated-time precision-collapse bug class;
rendering.md / multiplayer.md / decomp-reference.md / CLAUDE.md cross-linked.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
BTTranslocationRenderable was a no-op stub; now reconstructed over the engine's
POVTranslocateRenderable: loads tsphere.bgf and runs a collapse-on-arrival /
expand-on-reveal sphere (scale 100->1 over 1.3s, then 1->150 over 1.0s, rotating)
keyed on the player's SimulationState dial. Drawn direct from the render loop by
BTDrawTranslocationSpheres (beside BTDrawBeams, PASS_ALPHABLEND) -- the same
accommodation the beams/reticle use.
The asset loads by FILENAME (tsphere.bgf), not through the RES table -- which is
why every resource-name search missed the effect for three rounds.
Trigger wiring: btl4vid.cpp MakeEntityRenderables builds the sphere for the LOCAL
player (the authentic wiring builds it only for replicants + a POV fade for self,
but peer player-attribute replication isn't wired -- on a replicant SimulationState/
DropZoneLocation read uninitialised). btplayer.cpp pulses SimulationState
DropZoneAcquired->VehicleTranslocated at respawn (a 1.4s flip timer stands in for
the engine's +1s drop-zone re-post) and writes the respawn origin into the
DropZoneLocation attribute.
Verified 2-node (BT_TLOC_LOG): on respawn the sphere collapses (~100->5) then
expands (->150) at the valid drop-zone origin, deduped to <=2 active, respawn
cycle un-regressed, no render errors. Visual appearance (colour/size) still needs
a live look; the authentic see-others'-spheres path needs player replication.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
My earlier 'not a shipped asset' verdict was wrong. The effect is POVTranslocateRenderable
(engine) / BTTranslocationRenderable (our port), which loads tsphere.bgf BY FILENAME (not via
the RES table, so every resource-name search missed it) and runs a collapse-on-death /
expand-on-respawn sphere keyed on the SimulationState dial. Wiring already exists in
btl4vid.cpp; the renderable is a no-op stub in btstubs.cpp. That's why it never shows.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Searched the full pod install (nick-games/tesla4pod_extracted REL410/BT;
BTL4.RES byte-identical to ours) and the whole nick-games archive: no
whirl/warp/tunnel/vortex asset by filename or RES resource name, no such
psfx in BTDPL.INI. The decomp drop path (DropZoneReply FUN_004bffd0) fires
no effect — only Mech::Reset + the level-2 "translocated" cockpit alarm.
BTPLAYER source is absent from all archives (binary-only). RP411 analog
uses InitialStasisFlag (sim gate, no visual). Verdict: the pod build's
respawn is mech-redrop + translocated alarm; the "blue whirlwind" is not
a shipped REL410 asset.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Death transition (mech4) now dispatches VehicleDead(-1) to the owning
player; BTPlayer::VehicleDeadMessageHandler restructured to the authentic
@004c05c4 three-way branch: -1 = death bookkeeping + sever playerVehicle
(wreck stays) + 5s re-post; >=0 = engine drop-zone hunt -> DropZoneReply
-> CreatePlayerVehicle (new mech); -2 = the acquire probe. Guarded on a
DropZones group so a zone-less mission stays dead instead of aborting.
Three latent bugs the respawn path exposed, all fixed:
- IsMechDestroyed latched on graphicAlarm>=9 alone; a later leg hit on the
wreck rewrites the alarm to 4/3, un-latching -> the death transition
re-ran (double kill/score, abort). Now latches on movementMode 2||9.
- Score handlers dereferenced the severed playerVehicle during the dead
window; guarded.
- The console score flush routed a NetworkClient::Message through the
player's Entity::Dispatch, which stamps entityID past the smaller struct
-> /RTC1 stack overflow. Sent via application->SendMessage instead.
- Renderer LoadMission re-entry (per viewpoint-make) re-read the env INI;
its light block Fail'd on stale sceneLightCount. Reset it in
DPLReadEnvironment so the respawn's second read is clean.
Verified 2-node self-drive: B killed repeatedly by A respawns each time
(wreck stays), ticks + reloads + takes fresh damage on the new mech, no
abort/hang across multiple death->respawn cycles.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
User report from 2-window play: after dying, the pilot could still steer the
wreck around invisibly. Zero throttle/turn/fire and the forced harnesses at
the gBTDrive source when IsMechDestroyed() -- deliberately INPUT-only, so the
gait/animation machinery keeps running and the death collapse plays out.
Also recorded in the KB: the killer's screen shows the wreck SWAP but not the
death SINK (needs the Mech-level death/knockdown update records, disasm
preserved from task #51), and the respawn-cycle scouting (type-12 intro
scripts are binary ref-records; adrop; DropZone + Mech::Reset machinery) for
the recovery-whirlwind reconstruction (task #52).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User report: lasers only visible on the window firing them. The peer's
replicant emitters never learned the master fired.
THE AUTHENTIC PIPE (decomp-verified):
- FUN_0041c350 (the "beam keepalive" ServiceDischarge/ContinueDischarge call)
does TWO things: queue the LOCAL deferred beam-effect callback (@0x4bac0c)
on the app+0x34 manager -- our per-weapon render walk already plays that
role -- and set the subsystem DIRTY bit, which maps to the 2007 engine's
updateModel / ForceUpdate().
- Replication rides SUBSYSTEM UPDATE RECORDS inside the mech's update message:
the roster walk already hands the entity's stream to every subsystem's
PerformAndWatch; Simulation::WriteSimulationUpdate serializes each requested
updateModel bit; Entity::UpdateMessageHandler routes received records by
subsystemID to the subsystem's ReadUpdateRecord. All engine machinery --
the missing pieces were the Emitter's serialize/apply pair + the triggers.
CORRECTIONS to the dormant task-33-era transcriptions (never exercised --
nothing ever set updateModel -- so the latent misreads never surfaced):
- The weapon-family VTABLE SLOT MAP was swapped: slot 6 = ReadUpdateRecord,
7 = WriteUpdateRecord, 9 = TakeDamage (evidence: Mech hierarchy symmetry +
body semantics; @004ba568 resolves an EntityID at rec+0x30 through the
entity index -- record semantics, not Damage). Renamed across MechWeapon /
Emitter / ProjectileWeapon; the real Emitter::TakeDamage @004bafc8 is
undecoded (inherits MechWeapon for now).
- Emitter/MechWeapon Write: `*record = 0x38/0x18` is the record LENGTH, not
recordID; rec+0x30 is the TARGET's EntityID (GetEntityID()), not a colour --
the old `CopyColor(targetEntity+0x184)` was also a databinding trap.
- OVERRIDE-SIGNATURE TRAP: the decls used each class's own shadowing
UpdateRecord typedef as the param type, silently NOT overriding the engine
virtual (the base ran instead; nothing would ever have serialized).
Base-typed params (Simulation__UpdateRecord*), casts inside.
- Emitter::ReadUpdateRecord reconstructed (@004ba568): target EntityID resolve
(drop unknown non-null targets), MechWeapon alarm apply chain, beam fields.
- ServiceDischarge/ContinueDischarge: ForceUpdate() per keepalive tick + one
final record at beam end (turns the peer's beam off).
- Mech::DrawWeaponBeams extracted from the player-only drive block so the walk
runs for REPLICANTS (+ per-mech gun-port cache -- the process-wide statics
would have served the player's segment pointers as the replicant's muzzles).
VERIFIED 2-node: A fires 57 volleys -> 225 emitter records -> B applies all
225 -> B draws 414 beams (PPC blue / laser red, from A's replicant's own gun
ports). Solo un-regressed (150 beams, kill chain, no crash).
Also preserved: the full Mech::WriteUpdateRecord @0x4a0c2c recovery
(reference/decomp/mech_writeupdate_004a0c2c.disasm.txt) with all 9 record
types decoded (pose/alarm/leg-state+heat with the body-channel write-through
re-sync, knockdown, death, impact, movementMode) -- transcription deferred;
it replicates remote knockdown/death/heat and was not needed for beams.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User report from the first live 2-window session: throttle worked, A/D turn
did nothing. [mppr] showed the input REACHING the mapper (stickX=-1) but
turnDemand=0 and controlMode = pointer garbage (73583232 / 217478632 --
different per instance), from frame 1, MP only (solo mode=0).
Hunted with a cdb hardware write-breakpoint (ba w4 armed at ctor time on
&controlMode, compiled offset +0x128 from the [mode-ctor] probe). Caught the
writers red-handed: MechControlsMapper::BuildPilotArray / FillPilotArray.
ROOT CAUSE -- a shrunk-span array: the binary reserves a FIXED 10-slot pilot
block @0x15C..0x183 ((pilotArrayBuilt@0x184 - 0x15C)/4 = 10); the recon
declared `Pilot *pilotArray[1]` ("variable length"), so every write past
slot 0 stomps the members declared after it. In MP, FillPilotArray wrote the
PEER's Player pointer into slot 1 = over controlMode -> the turn shaping
dispatched on pointer garbage -> turnDemand 0. MASKED in solo: the zero-loop
overrun wrote 0 there, and 0 == BasicMode. This was ALSO the real cause of
task #48's "turnDemand zeroes out in -net" observation (worked around then
with the direct goto-turn; attribution corrected in the KB).
Fix: pilotArray sized to the binary's own 10-slot span; ctor zeroes all
slots; BuildPilotArray clamps pilotCount to capacity (local + 9 remotes; an
8-pod game fits). Diagnostics kept (BT_MODE_LOG ctor probe; [mppr] line now
prints mapper/&mode).
VERIFIED: MP session holds mode=0 throughout, turnDemand flows nonzero, the
mech turns (heading -1.33) and closes on the peer; solo kill chain intact.
KB: new "shrunk-span array" gotcha (reconstruction-gotchas 5) with the cdb
ba-w4 recipe; multiplayer.md task-#48 note corrected.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A peer's replicant mech slid around as a frozen statue: DeadReckon moved
localOrigin but nothing animated the legs. Reconstruct the replicated-motion
display loop: after DeadReckon, derive a SIGNED speed demand from the
replicated velocity (the master publishes worldLinearVelocity in every update
record, Mover::WriteUpdateRecord MOVER.cpp:759; the replicant stores it in
updateVelocity.linearMotion :712; sign = forward dot against the -Z facing),
write it to controlsMapper->speedDemand (the LEG channel's live demand
source), and advance the LEG state machine for its JOINT writes only --
travel stays with DeadReckon, so position always follows the master and the
clip cadence matches the replicated speed.
Two replicant-only initialization holes found live (the master's equivalents
are primed per-frame in the PLAYER-only drive block):
- globalTimeScale/idleStrideScale read 0 -> clip advance time dt*0, the leg
machine engaged (state 11) but the clip froze (legCycle stuck 0);
- reverseSpeedMax2 (the run-cycle rise-clamp @0x7a0; LoadLocomotionClips does
NOT set it) read debug-heap garbage -> legCycleSpeed clamped to -4.3e8 on
entering run state 13. Primed both; legCycleSpeed sane-banded once.
VERIFIED 2-node (BT_GOTO driver + BT_REPL_LOG observer): the replicant runs
the full gait lifecycle -- stand -> walk(5,7) -> run(10,12), legCycle 7.5->34
tracking the master's speed, wind-down(8), Standing(0) when the master stops.
Solo un-regressed (goto->aim->kill chain intact, no crash).
projectedVelocity source-swap (the binary publishes the BODY channel's
smoothed projection instead of the actual travel velocity) evaluated and
DEFERRED: it risks a verified working replication feed for a smoothing
nuance; recorded in multiplayer.md.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User report: the shadow broke MID-SESSION -- suddenly solid opaque black and
buried on any slope. The session log pinned the trigger: pressing V. The
view toggle (SetViewInside) RELOADS every segment mesh through the object
loader and swaps them into the render tree -- including blh_tshd.bgf -- and
the reload path never called SetIsShadow(1) (only the original tree-build
call site tagged it). Untagged, the entire shadow pipeline disengages: no
translucent TFACTOR state block (solid black), no depth bias (buried), wrong
pass. The damage-swap reloads had the same hole.
Fix: tag *tshd* filenames in LoadObjectBGF itself, next to the existing
all-shadow-material detector (which misses the mech proxy -- plain black
material, not shadow_mtl). Every load path -- build, damage swap, view
toggle -- now yields a tagged object; the drawOps alphaTest routing reads
the flag downstream. User-verified live: the shadow survives repeated V
toggles, translucent and slope-correct.
This also likely explains the RECURRING "shadow vanishes on inclines"
reports: any V toggle silently broke it; every relaunch "fixed" it. KB
(locomotion.md): loader-level tag recorded as co-dependent part 5 of the
shadow arrangement; the vanish-diagnostic note now lists all three causes
(reload untag / gait desync / genuine burial) with check-first guidance.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User report: the ground shadow painted OVER the mech's feet. Root cause: the
flat quad + big depth-bias (-0.004 ~= 2-4 world units) workaround for slope
burial -- a bias big enough to beat the TERRAIN also beat the FEET in the
depth test, and the shadow drew in the late alpha-blend pass (after the mech),
so z-arbitration was the only layering control.
Fix, four co-dependent parts (user-verified live: feet on top, opaque,
survives inclines/declines):
- TILT (mech.cpp UpdateShadowJoint): apply the SKL's own contract for
jointshadow ("apply terrain angle to pitch and roll") -- quad up aligned to
the surface normal via an orthonormal basis fed to the ENGINE's own
LinearMatrix->EulerAngles conversion. Hand-derived Euler signs are exactly
how the prior tilt attempts "dug into the hillside"; the engine conversion
is convention-proof. ~35 deg cliff-guard cap.
- SAMPLER (mech4.cpp): surface gradient from the collision probes PLUS
BTVisualGroundLift per probe -- the quad hugs the VISIBLE terrain, whose
lift varies across a slope; world->local rotation by the TRUE yaw from
localToWorld, not the drift-prone gDriveHeading mirror (the task-#48 bug
class). BT_SHADOW_LOG traces the normals.
- DRAW ORDER (new PASS_SHADOW, l4d3d.h/L4VIDRND/L4VIDEO): shadows draw
between the STATIC terrain and the DYNAMIC opaque bodies -- the classic
decal order. The mech, drawn after, z-passes over the shadow: the bias
can never paint the shadow over the feet, structurally.
- BIAS pairing (L4D3D.cpp): tilt on (default) -> decal epsilon -0.0008;
BT_SHADOW_TILT=0 flat fallback -> the old -0.004. BT_SHADOW_BIAS overrides.
KB (locomotion.md): the four-part arrangement recorded as co-dependent, with
the gait-desync-symptom diagnostic hint retained.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The shadow report was a desync symptom, not a bias/tilt bug: the ~2-4 unit
depth-bias margin is tight enough that v4's out-of-phase leg pose showing
through could push the flat quad under a slope. No shadow code changed;
both symptoms cleared together. Noted as a diagnostic hint (check
[sync]/BT_SYNC_LOG before touching bias/tilt for this report class).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
User-reported recurrence of "legs stutter and lose sync". Live [sync] showed
advSum/legSum drifting 5->92 during walk-up churn, then locking. Reproduced
headlessly (BT_FORCE_OSC throttle feathering + BT_FORCE_TURN + the new
per-frame BT_SYNC_LOG probe) and root-caused in two parts:
- SEED: the bring-up turn-in-place trigger arms only the LEG channel, so the
walk re-entry lands one frame apart -> a persistent clip phase offset.
- AMPLIFIER: with offset clips the two end-of-clip callbacks fire on different
frames; a demand change landing between them picks DIFFERENT next states
(one channel winds down, the other keeps walking) -> opposite-phase churn
(observed bs=6 ls=7 -- body on walk-R, leg on walk-L).
Under v4 (display+travel = BODY) the out-of-phase LEG pose showed through on
every frame the body didn't write joints (Standing/wind-down) = the visible
stutter/pop.
DISASM GROUND TRUTH (neither Advance fn has a static decomp caller; found by
byte-scanning the CODE section for e8 calls):
- master perf (0x4a9b5c gap) -> AdvanceLegAnimation @0x4aa399 (air @0x4aa388)
-> -dist/dt into localVelocity (+0x1cc): the LEG drives LOCAL travel and
(writing last) the displayed pose;
- IntegrateMotion (0x4ab1c8, body advance @0x4ab312, caller FUN_004ab430 =
the projected-origin updater) -> -dist/dt into projectedVelocity (+0x2a0):
the BODY is the dead-reckoning PROJECTOR -- locally invisible.
v4's reading of FUN_004ab430 as the travel source was wrong.
FIX: advance body FIRST (projection; its writes get overwritten), leg LAST
(displayed pose); travel = legAdv under the two-channel split. Display ==
travel through the LEG by construction -- body-channel drift can no longer
become visible, structurally (not by input-symmetry luck).
Verified: trn+feathering repro clean; straight drive locked (adv==proj);
solo goto=enemy kill chain intact; MP 2-node drive-to-range cross-pod kill
intact (241 hits, DESTROYED); no crashes. KB corrected (locomotion.md roles
+ P3_LOCOMOTION.md v5 entry). New gated instruments: BT_SYNC_LOG,
BT_FORCE_OSC.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The BT_GOTO beeline never actually closed on a target -- three separate faults:
1. It was gated behind `if(gBTDrive.forced)`, and `forced` was set ONLY by
BT_AUTODRIVE -- a chicken-and-egg where the block that would drive was gated
on the drive already being on. So BT_GOTO alone just turned in place (the
"stall"). Fix: BT_GOTO enables forced mode itself AND supplies its own
forward throttle (a beeline must drive, not only steer), cutting the throttle
inside a stop radius (weapon range for "enemy", so it holds + shoots).
2. Steering used the gDriveHeading SCALAR MIRROR (seeded to 0), but MP pilots
spawn facing a non-zero heading -> it steered the WRONG way. Fix: heading
error from the mech's ACTUAL forward (localToWorld's -Z axis), signed angle
to the target.
3. The goto turn was routed through controlsMapper->turnDemand, which zeroes out
in -net mode (the mapper key-bridge only shapes the local viewpoint mech
there) -- the heading froze and A drove straight past B. Fix: apply the goto
turn to the orientation integration DIRECTLY (if gBTGotoActive, turn=
gBTGotoTurn), after the mapper read.
Verified 2-node one-box: A with BT_GOTO="enemy" ALONE (no autodrive) drives
dist 1673->231m, holds at weapon range, mechPicks=59, autofire -> B hdlr=192,
DESTROYED -- a fully automated human-style cross-pod KILL via the real beam
path (not the FORCE_DMG hook). Solo un-regressed (plain autodrive still drives;
goto="enemy" resolves the enemy, mechPicks=59, no crash). BT_GOTO_LOG traces
the beeline.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The aim/pick ray was the active RENDER EYE's view direction, which made a
mech unable to aim at a distant enemy:
(a) it inherited the mech BODY pitch -- on a slope the body tilts ~8 deg to
conform to terrain, so the boresight pointed that far DOWN and the pick
ray hit the ground ~50-100m short of the target (groundPicks, never
mechPicks); and
(b) in CHASE view (the default) it was the elevated behind-the-shoulder
camera, so the ray started over the mech's shoulder and flew high.
Fix (L4VIDRND DPLEyeRenderable::Execute, the BTSetAimCamera publish):
- LEVEL the boresight: drop the view-direction pitch and rebuild an upright
basis (world +Y up) -- the guns fire along the mech's gyro-stabilized
HORIZONTAL heading, not the terrain-pitched body. Reticle X still carries
the torso twist (BTTwistToReticleX); reticle Y any elevation.
- ANCHOR the ray origin at the mech's TORSO (myEntity->localToWorld
translation + ~5) instead of the render eye, so it is correct in BOTH the
chase (default) and cockpit views.
Verified end-to-end: solo, enemy 120m ahead, BOTH chase and cockpit view ->
[target] "MECH under boresight ... mechPicks=59 groundPicks=0" -> beam ->
*** DESTROYED. Also proved PickRayHit resolves a REPLICANT when aimed at it
(hit=1 at 757m, 2-node run).
mech4: BT_GOTO="enemy" test aid now chases the nearest peer of ANY instance
(solo dummy OR replicant), not just replicants.
Still open (locomotion, not aim/MP): BT_GOTO/BT_AUTODRIVE stall short of the
enemy, so an AUTOMATED 2-node kill via the real beam isn't demonstrated yet
(the FORCE_DMG hook still is). A human can now aim correctly in the default
view. Solo un-regressed; build clean; KB validator clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Re-measured interactive -net combat on a 2-node run. The KB's "aim ray dead /
drive dead in -net" was STALE: A's aim ray is live (noRay=0, the active eye
publishes the aim camera) and A drives (BT_AUTODRIVE/BT_GOTO move it). The real
blocker to a human cross-pod kill via the beam path is NOT the netcode:
- Boresight inherits the mech BODY pitch. The aim ray = the active eye's view
dir; on sloped terrain the body pitches to conform and the boresight points
~8 deg down ([EYE] up=(-.03,.99,-.14) == [pick] ldir.y=-0.145 every frame),
so at range the ray hits terrain ~50-100m short of the enemy (mechPicks=0).
- Drive/goto stalls ~700m short before closing to pick/weapon range.
- PickRayHit vs a replicant is thus unconfirmed, but the replicant localToWorld
IS correct (lstart magnitude == true A->B distance; the DeadReckon +
localToWorld=localOrigin block already tracks the replicated position).
Change: add a gated BT_GOTO="enemy" test mode (beeline toward the nearest live
replicant) -- MP spawn coords vary per run so a fixed "x z" can't reliably face
the peer. KB (multiplayer.md) updated with the corrected findings + the real
remaining work (torso-pitch/level-boresight aim ray + drive-to-range).
Solo un-regressed; btl4.exe builds; KB validator clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Full cross-pod KILL now works: instance A (host 2) shoots B's mech (host 3)
dead over the network; B runs the death transition on its own screen. Verified
on a 2-node one-box run (BT_MP_FORCE_DMG on A): B's mech takes ~36 remote hits
of 12 and is DESTROYED, with wreck swap + explosion.
ROOT CAUSE (one bug): B's own MASTER mech was INVALID. Entity::Receive(Event*)
does `if(!IsValid()) event->Defer()` -- so a cross-pod-delivered TakeDamage that
arrived, resolved to B's real mech, and Posted, was DEFERRED forever; the handler
never ran. The reconstructed Mech::Make force-set ValidFlag only for replicants
(the P6 bring-up hack for the partial MakeReady handshake); a locally-created
master never got validated.
FIX (one line): Mech::Make now sets ValidFlag for the master too -- the
reconstructed ctor builds the mech synchronously, so it's safe. Nothing else
changed: targeting, the virtual victim->Dispatch, the replicant reroute, the
wire, receive, and id resolution were ALL already correct.
Two prior root causes were BOTH mis-diagnoses from the WRONG messageID:
Entity::TakeDamageMessageID is 18, not 21. Every earlier probe filtered
msgID==21, never matched a real TakeDamage, and mis-correlated the stray 21
hits' 3:19 / classID-48 against the real 3:22 hit -- inventing (a) a "cross-TU
Entity layout divergence" (disproven: offsetof identical in both TUs, sizeof
444, entityID@0x17C) and (b) a "localID corrupted on the wire". Both FALSE; the
id is correct end-to-end (3:22 = B's master).
- game/reconstructed/mech.cpp: validate master + replicant at Make.
- game/reconstructed/mech4.cpp: BT_MP_FORCE_DMG cleaned to a concise gated
2-node cross-pod reproducer (real virtual Dispatch path).
- context/multiplayer.md, reconstruction-gotchas.md: corrected root cause +
the validity-defers-cross-pod gotcha. Engine diagnostics reverted to clean.
Solo un-regressed (the solo enemy was already force-valid; the change only ADDS
ValidFlag). btl4.exe builds; full mission loop clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The prior commit (c78662a) filed cross-pod damage failure under a
"cross-TU Entity/EntityID layout divergence" (and the P5 base-region
audit). A compile-time offsetof probe injected into BOTH translation
units disproves that: game/reconstructed AND engine/MUNGA compute an
IDENTICAL Entity layout --
sizeof(Entity)=444 offsetof(entityID)=380 (0x17C)
offsetof(ownerID)=388 offsetof(simulationFlags)=32
offsetof(Mech,entityID)=380 (Entity subobject at 0)
So Entity::Dispatch reads this->entityID at the SAME offset mech4's
GetEntityID() does -- there is no per-TU read difference, and the
logged 3:22 (mech4 candidate) vs 3:19 (engine Dispatch) cannot be one
object read two ways. Those lines were different objects/messages,
mis-correlated. The distinct, REAL P5 "base-region layout divergence"
(HARD_PROBLEMS.md) is about raw-offset stomps at this+0x2d4..0x2f0 --
far above entityID -- and is unrelated to #47.
Changes:
- Revert the committed BT_MP_NET engine diagnostics (ENTITY/EVENT/
NTTMGR/RECEIVER/L4NET) to the clean a9c3e96 baseline -- those are
the very instruments that produced the mis-correlated data.
- mech4 BT_MP_FORCE_DMG hook: dispatch via the real m->Dispatch(&td)
path (no false id stamping); comment records the offsetof finding.
- context/multiplayer.md: layout-divergence RULED OUT; 3:22/3:19 marked
unconfirmed; leading hypotheses reframed as H2 (wire host-relative
(de)serialization) vs H3 (replicant id != master's registered key),
to be distinguished by a 2-node run with per-message correlation.
Solo un-regressed (mech walks + targets, 0 faults). Cross-pod delivery
remains open, but the investigation is redirected off the wrong (large,
structural) layout-audit path onto the EntityID wire/id-assignment path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Traced the dispatched-message delivery end to end with BT_MP_NET /
BT_MP_FORCE_DMG. Everything works except one wire step:
- A's Entity::Dispatch reroutes the replicant's TakeDamage
(application->SendMessage(ownerID=3, EntityManager, msg)); POST-Dispatch
the message carries entityID=3:22 (host:local) -- the replicant's own
ID, matching B's master (GetEntityID()==3:22). VERIFIED sent.
- On B the message ARRIVES, GetEntityPointer finds an entity, Posts it,
the event drains (ProcessEventTask = ProcessOneEvent(0)), Event::Process
runs, Receive finds+calls a handler. VERIFIED the full deliver chain.
- BUG: B receives entityID=3:19, NOT 3:22 -- the localID dropped by
exactly the hostID (3) between A's send and B's receive. So
GetEntityPointer(3:19) returns the WRONG entity (classID 48, not the
mech 0xBB9), whose base handler ignores the unaimed hit -> 0 damage.
Auto-replicated UPDATE records (msgID 18) arrive with the correct 3:22
and find the mech, so the corruption is specific to the dispatched-
message wire path/direction.
Next: the host-relative EntityID (de)serialization on the dispatched-msg
path (RoutePacket / packet EntityID encoding) -- diff vs the update path
which translates correctly. Diagnostics retained (all BT_MP_NET-gated,
off in solo -- verified: solo 22 hits, 0 probe noise).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
