a77c5a140796ca337e1a68b709a61e3866d58642
217
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Joe DiPrimaandClaude Opus 5
|
a77c5a1407 |
#149: measure the filed suspect -- the #141 segment sweep is EXONERATED; ship [segperf] telemetry
A/B on identical 2-node beam-heavy runs (scratchpad/night14/segperf.sh),
BT_BEAM_SEGFRESH=1 (swept behaviour) vs =0 (pre-sweep compose at the beam
site, the only per-frame swept call):
fresh(swept) legacy(pre-sweep)
rstat blocks>50ms 0 0
maxDraw worst 730ms 646ms (mission-load spike, BOTH)
segperf dirty-passes 38/s 4/s <- the sweep DOES multiply
segperf accessor ms 0.97/s 0.57/s <- ...by 0.4 ms/s. Noise.
So the invalidation-storm hypothesis I filed on #149 is wrong by three orders
of magnitude, and Oracle's sustained 50-104ms stall window does NOT reproduce
on this rig at all. Refusing to guess a third time: the build now carries the
telemetry to answer it on the machine that actually regresses --
[segperf] calls= dirty= ms= printed beside every [rstat] window under
BT_PERF_LOG (JMOVER counters; two integer increments when unset), and
BT_BEAM_SEGFRESH=0 remains as a one-env A/B for the beam site.
Default stays FRESH (the swept accessor): its measured cost is trivial and it
is the correctness-cautious side while the peer-beam-staleness question is
unmeasured.
Next for #149: Oracle runs one session with BT_PERF_LOG=1. If [segperf] ms is
large inside his stalled windows, segment work is implicated on HIS
configuration and BT_BEAM_SEGFRESH=0 gives the immediate A/B; if it is small
(as here), the stall is elsewhere in the 817->857 delta and we hunt with his
numbers instead of my theories.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC
|
||
|
Joe DiPrimaandClaude Opus 5
|
0d6ed40db9 |
#137 FIXED: restore the binary Reset's +0x58c re-seed -- the respawn freeze was a
teleport-poisoned finite difference, not heat, not the reset, not velocity
The one-line fix is the binary's own SECOND Reset instruction, dropped in
transcription: FUN_00408440(mech+0x58c, param_2) ==
accelPrevPos = origin.linearPosition;
+0x58c is the previous-position memory of the AccelerationLastFrame ring feed
(+0x81c/0x824/0x828/0x82c). The port reconstructed the ring faithfully (ctor
part_012.c:9836, derivative :15169) but Reset never re-seeded its cache, so the
first post-respawn sample computed |newPos - prevPos|/dt = TELEPORT DISTANCE/dt
(~1e5) into the velocity ring; the ring-mean derivative spiked
AccelerationLastFrame (pure forward, with an opposite-sign echo ~15 frames later
as the sample rotated out of the 15-ring); and the myomer integrator @004b8d18
turned it into a one-tick pendingHeat deposit of ~3e9:
termAccel = (1-accEff) * |v| * |a| * m * dt
= 0.2 * 40 * 1.04e5 * 75000 * 0.044 = 2.75e9
snapping the freshly-reset myomers from T=77 to T~9000 against failT=2000 ->
speedEffect 0 -> speedDemand *= 0 -> "respawned unable to move until it cools".
WHY ~8% IN THE FIELD: the deposit needs |v| in the same 1-2 frames, so only
pilots whose throttle was still forward at the respawn -- physical lever /
HOTAS, exactly who reported it -- had gait-republished speed in the spike
frames. Idle-throttle respawns deposit ~nothing. (And frozen-subset-of-
died-hot: died hot == was running hard == lever still forward.)
MEASURED, same abusive bench (0.95 throttle + continuous autofire):
before: deposits up to 3.35e9, every one 3-4 lines after a Mech::Reset;
4-6 of 7 respawns frozen; post-reset myomers T 7700-12100
after: deposits >1e7: ZERO across 7 respawns; frozen: ZERO;
post-reset myomers T 77-178 (vs degradeT=1000)
THEORIES KILLED ON THE WAY, each by operand data, in order:
* stale pendingHeat carryover (bounded to 1 frame, +1.2K -- gates agent)
* slow accumulation during the death window (consumers tick the wreck)
* conduction from a hot neighbour (roster-wide snapshot: ALL partners 77;
flow trap: zero e6 flows into the myomers, ever)
* drag/impulse writers (both trapped: never fired)
* my own earlier "velocity-driven, players accelerate to top speed" close of
the ticket -- arithmetically impossible (input ceiling ~6.5e5/frame ~=
60 deg/s; the observed snap was 1,100-21,000 deg/s) and corrected in
context/subsystems.md, which carried the wrong paragraph.
* the localAcceleration zero-fill (+0x1dc) restored along the way is KEPT --
the binary does it -- but it was NOT the cause; the snapshot is rebuilt
from the position difference one frame later.
Probes kept (all BT_HEAT_LOG-gated): roster-wide [myofreeze] at-death/at-reset/
post-reset (T + heatEnergy + pendingHeat per subsystem), the [heatflow]
conduction trap with full operands, the [myodep] deposit trap with mech
identity + acceleration components. Engine MOVER.cpp traps reverted -- they
proved their negative (drag/impulse innocent) and do not belong in engine
source.
Gotcha #30 records the class: when the binary's Reset writes a cell you don't
recognize, that write IS the spec -- transcribe the whole zero/seed list; any
prev-value cell backing a finite difference must be re-seeded at every
teleport; and derived state (T) sampled at the reset proves nothing about the
backing producers.
The operator called the shape of this two days ago: "maybe the math gets screwy
in respawning while some systems are ticking while values are being reset."
That is precisely what it was.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC
|
||
|
Joe DiPrimaandClaude Opus 5
|
5556329807 |
#137: REPRODUCE the myomer freeze -- reset is innocent, the myomers RUN AWAY after it
First reproduction of #137 in a bench. The reset path is exonerated and the real defect is located, though not yet explained. DECOMP RE-READ (what the reset actually does): Mech::Reset @0049fb74 walks the roster from index 2 calling vtable +0x28 (slot 10 = ResetToInitialState), then @0049f788. Myomers::RTIS @004b8aa4 -> PoweredSubsystem::RTIS @004b0e6c -> ALWAYS HeatSink::RTIS @004ad760, whose first act is param_1[0x45] = param_1[0x4f] // currentTemperature = startingTemperature (byte 0x114 = 0x13C). Our port matches AND additionally resets heatEnergy, which it must, since HeatSinkSimulation derives currentTemperature = heatEnergy / thermalMass. The freeze itself is the derating curve @004b8ac0: temp >= degradation(@0x118) -> falls off temp >= FAILURE(@0x11C) -> 0.0 -> chain MAX 0 -> speedDemand *= 0. Nothing in the reset chain touches Myomers::speedEffect (@0x31C). MEASURED (scratchpad/night14/myofreeze.sh -- hot mech, repeated self-kill): at-reset Myomers T=77 deg=1000 fail=2000 speedEffect=0 <- stale post-reset Myomers T=77.14 speedEffect=1 <- 1 frame post-reset Myomers T=9297.8 fail=2000 speedEffect=0 post-reset Myomers T=11620.4 fail=2000 speedEffect=0 So, in order: * THE RESET WORKS. T is exactly startingTemperature at the reset. My original bench was right about that much; it just stopped looking there. * THE STALE speedEffect IS REAL BUT HARMLESS -- it survives the reset (no RTIS writes it) yet self-heals on the very next tick. Not the bug. * THE MYOMERS THEN RUN AWAY: 77 -> ~11,600 against a FAILURE point of 2,000, nearly 6x over, in seconds. That is not heat earned by running; that is a runaway, and it is what pins speedDemand at 0 until it cools -- exactly Oracle's "maxed heat bar ... unable to move until it cools off". * Rate: 7 of 105 census samples over the failure temp (~6.7%), against the field's 5-of-61 respawns (~8%). Same order, so the bench is reproducing the field condition and not a bench artifact. * The myomers link to Condenser5 (mass=250000 k=190000) -- literally Oracle's "loop 5 and generator D heating up as all the excess heat goes into the loop". NOT YET ESTABLISHED, and the reason this is a checkpoint and not a fix: whether the runaway is CAUSED by the respawn or is a heat-model calibration problem that merely CORRELATES with it (you die when you overheat, so respawns cluster around hot periods). The bench drives at 95% throttle with continuous missile autofire, which is abusive, and the [heat-t] census has no pre-first-reset samples to compare against. Next step is to instrument the heat INPUT and diff a respawn-adjacent window against a steady-running window. Probe added: [myofreeze] prints T / degradation / FAILURE / speedEffect per myomers plus the chain MAX the mover multiplies by, AT the reset and for ~4 s after (armed by Mech::Reset, sampled where the multiplier is formed). The post-reset window is what nothing was watching -- sampling only at the reset is what made this look innocent and got the ticket wrongly closed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC |
||
|
Joe DiPrimaandClaude Opus 5
|
cacca58836 |
#148 is NOT A BUG: a peer mech does not tick before RunningMission -- by engine design
Chased to the bottom instead of stopping. The answer is that there was
nothing to fix, and my bench was lying to me.
Entity::Execute (ENTITY.cpp:556, real engine source [T0]) calls PerformAndWatch
ONLY when the app state is RunningMission/EndingMission or the entity
IsPreRunnable(); otherwise it merely WriteSimulationUpdate()s.
Entity::DefaultFlags is DynamicFlag|MasterInstance -- no PreRunFlag. Only
Player and Director add it, and Mech::Reset sets it for a reset MASTER ("a
reset master must tick"). A REPLICANT mech never gets it.
So a peer mech performs ZERO subsystem ticks until the round actually starts,
however much correctly-replicated data is arriving. Measured on the observer:
235 [perf-first] mech 3:161 master <- own mech, immediately
402 [torso-rec-rx] <- peer torso records arriving
2754 [perf-first] mech 2:55 REPLICANT <- peer's FIRST performance
2758 [torso] PushTwist COPY <- its torso ticks 4 lines later
2761 [ent-exec] state=5 <- RunningMission
The peer starts performing exactly at the RunningMission transition. That is
the engine doing what it says.
WHICH MEANS THE PREFIX WAS A BENCH ARTIFACT. BT_AUTOFIRE starts shooting
immediately, during WaitingForLaunch -- something no player can do in a real
match -- so those 60 leading salvos measured a peer whose torso had never run.
Every "ZZZZ...XXXX" pattern in this investigation was that, and the first X
lands within a few lines of the state transition. #141's fix is unaffected and
remains verified: the segment-cache defect was real and mid-match.
Chain of things ruled out on the way, all measured:
* record CADENCE is authentic -- sends on RATE CHANGE (payloads are the sweep
extremes, rate flips sign), peer dead-reckons between them. 12 records for
12 reversals is correct, not starved. My "only 13 records" premise was wrong.
* ComputeTargetTwist clamp -- limits load fine on the copy (+/-2.44346).
* the torso's own executable flag -- restoring the engine's instance branch
(
|
||
|
Joe DiPrimaandClaude Opus 5
|
f36f0136c8 |
#148: restore the ENGINE's replicant instance-branch in the Mech subsystem tick
Correct on its own merits as a fidelity fix; it is NOT the cause of #148, and I am not claiming it is. Entity::Perform (ENTITY.cpp:733-793, real engine source [T0]) picks the executable predicate BY INSTANCE: if (GetInstance() != ReplicantInstance) IsNonReplicantExecutable() else IsReplicantExecutable() and the two differ exactly on the replicant case (SIMULATE.h:195-206): NonReplicant : (flags & DontExecuteFlag) == 0 Replicant : (flags & DontExecuteFlag) == 0 || lastUpdate >= lastPerformance `ExecuteOnUpdate()` SETS DontExecuteFlag -- it means "do not tick me every frame, tick me when an UPDATE ARRIVES". Mech's reconstructed tick loop used the NonReplicant predicate for EVERY mech, dropping the branch, so on a replicant any ExecuteOnUpdate subsystem could never run however many records arrived. Restored. Measured: it does NOT move #148 (first TorsoCopySimulation call 1014 -> 1006, noise). So the torso's own flag was not the gate. Keeping it because the engine source is unambiguous about what the loop is supposed to do. WHAT #148 ACTUALLY IS, now much better characterised: * The record CADENCE is authentic -- my original "only 13 records" framing was wrong. Payloads are the sweep EXTREMES with `rate` flipping sign each time (atUpd 0.0437, 2.3558, -2.3928, 2.3854, ...): the master sends on RATE CHANGE and the peer dead-reckons `atUpd + rate * elapsed` between them. 12 records for 12 direction reversals is correct, not starved. * The real defect is that the peer's copy torso PERFORMANCE does not run at all until log line ~1006, while its first record arrived at line 205 -- ~800 lines of correctly-replicated twist integrated by nobody. The first tick coincides with the replicant's MODEL bring-up, not with record arrival: [loadclips] end: fScale=0.8 ... hasGimpClips=1 [clipfix] mech 05769358 -> EXTERIOR (lean) [torso] PushTwist COPY node=057A3C68 type=1 twist=-1.52319 so the gate is above the subsystem level, in replicant model/clip init. Not yet found; #148 stays OPEN. Also: [torso-copy] logs on call #0 (s_cl++ % 120), so its first line IS the first Performance call -- that is what makes the 205-vs-1006 gap readable, and it is why the earlier "first copy currentTwist != 0 at 1016" reading was a SAMPLING artifact, not a measurement of when the twist started. Probe additions kept: [torso-copy] now prints limL/limR/enab (which ruled out the ComputeTargetTwist clamp -- limits load correctly at +/-2.44346 on the copy), and [launchframe] now prints the shooter's live torso twist so twistDelta and its driver sit on the SAME line. That pairing is what proved #141 is fully fixed: every zero-twistDelta peer launch reads liveTwist=0, and the first launch with liveTwist=-1.84061 reads twistDelta=-1.83813. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC |
||
|
Joe DiPrimaandClaude Opus 5
|
e6c5ac951e |
#141 sweep: every hand-composed segment->world now goes through the engine accessor
Finishing the audit the #141 fix implied. The unfaithful pattern
mw.Multiply(seg->GetSegmentToEntity(), mech->localToWorld);
appeared at FOUR sites, not one. GetSegmentToEntity only recomputes when
segmentModified is already set (SEGMENT.cpp:262); the thing that sets it is the
binary's FUN_00424da8 == JointedMover::GetSegmentToWorld, which tests
AreJointsModified() and marks the whole segment table dirty. Compose by hand
and you read whatever cache is there -- fresh on the local mech (the render pass
refreshes it every frame), BIND POSE on any replicant.
Swept (the muzzle path was fixed in
|
||
|
Joe DiPrimaandClaude Opus 5
|
f01de8cbfa |
#141 follow-up: do it the BINARY's way -- the muzzle query IS the segment refresh
The previous commit's fix worked but was NOT faithful: it set
ModifyJoints(True) to force the engine's dirty flag before reading the
segment. The binary never does that. Called out by the user; corrected.
WHAT THE BINARY ACTUALLY DOES. MechWeapon::GetMuzzlePoint @004b9948 ends in
`FUN_00424da8(owner, segment, out)`, which is JointedMover::GetSegmentToWorld
instruction-for-instruction:
iVar1 = FUN_00417ab4(param_1 + 0x31c); // GetJointSubsystem()
if (*(int *)(iVar1 + 0xfc) != 0) { // AreJointsModified() <- TESTED
... walk owner+0x300, seg+0xc = 1 ... // ModifySegment()
*(int *)(iVar1 + 0xfc) = 0; // ModifyJoints(False)
}
FUN_0040b104(out, FUN_004244dc(seg), owner+0xd0); // x localToWorld
So in the 1995 image EVERY muzzle query performs the joints->segments refresh,
and the flag is only ever TESTED, never set.
THE REAL DEFECT. BTResolveWeaponMuzzle -- labelled "the faithful FUN_004b9948"
-- hand-composed `seg->GetSegmentToEntity() x localToWorld` and skipped
@00424da8 entirely. GetSegmentToEntity only recomputes when segmentModified is
already set (SEGMENT.cpp:262), so it returned a stale cache. On the MASTER the
render pass refreshes the local mech every frame and hid it; a REPLICANT got no
refresh, so peer muzzles sat at the BIND POSE and the missile left along the leg
facing. Fixed at the muzzle path, where the binary puts it -- and the forced
flag in BTPushProjectile is REMOVED (the launcher calls GetMuzzlePoint just
above, so the cache is already current when the launch frame is composed).
MEASURED -- the faithful path scores exactly what the hack did, so the hack
bought nothing and is gone:
master n=165 max 2.1389 mean 1.2718 >0.1rad 100%
REPLICANT n=165 max 1.9426 mean 0.8051 >0.1rad 64%
AND THE 64% IS NOT A PARTIAL FIX -- I called that wrong last commit. The
failures are a contiguous PREFIX, not interleaved:
ZZZZ...(60)...ZZZZXXXX...(105)...XXXX
and they end exactly when the peer acquires a twist to carry:
first torso RECORD received : line 206
first copy currentTwist != 0 : line 1016
first CORRECT launch frame : line 1054 (38 lines = probe granularity)
Those 60 salvos fired while the replicated twist was genuinely 0, so launching
along the body facing was CORRECT. Once the peer has a twist, 100% of launches
carry it. #141 is fixed.
SEPARATE ISSUE FOUND, not fixed here: the peer's copy torso takes far too long
to first reflect the master's twist -- the master was twisted from the start,
only 13 torso records arrived across the whole run, and the copy's twist stayed
0 until line 1016. That is a torso REPLICATION CADENCE problem, and it would
also make peer torsos visibly lag -- likely relevant to #37 (MadCat torso
backwards) and #70 (twist stops after respawn).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC
|
||
|
Joe DiPrimaandClaude Opus 5
|
05d7b5890a |
#141 peer missiles: the launch frame read a STALE segment cache on replicants
Oracle: "missiles are firing in the direction the mech feet are facing ...
and then coming around to track the target", peer POV only -- the shooter's
own view is correct.
REPRODUCED AND MEASURED (scratchpad/night13/missileframe.sh, 2-node: only A
sweeps its torso and only A fires, so every REPLICANT line in B's log is the
mirror of one A salvo). New [launchframe] receipt prints the yaw of the
launch forward vs the BODY forward on both nodes:
master n=165 |twistDelta| max=2.2962 mean=1.2283 >0.1rad: 100%
REPLICANT n=165 |twistDelta| max=0.0000 mean=0.0000 >0.1rad: 0%
segResolved=1 on BOTH, and segYaw == bodyYaw EXACTLY on the peer.
WHAT IT IS NOT. Both sides already pass the mount segment (mislanch.cpp:363
master, :478 replicant mirror, both `GetSegmentIndex()` from task #67), and
the peer's torso data is fine end to end: records arrive (atUpd=2.44/-2.39,
rate 0.305), the copy extrapolates correctly (cur=-2.13987 target=-2.13987
copy=1), and the copy torso demonstrably writes its joint (PushTwist COPY
twist=-1.49601). Hierarchy is identical too: same seg 18, same parentIdx 4,
non-null parent + joint subsystem on both.
ROOT CAUSE. BTPushProjectile composed the frame BY HAND --
`mw.Multiply(seg->GetSegmentToEntity(), localToWorld)`. But
EntitySegment::GetSegmentToEntity (SEGMENT.cpp:262) recomputes ONLY when
`segmentModified` is set, and the thing that sets it after a joint moves is
JointedMover::GetSegmentToWorld (JMOVER.cpp:136-146), which tests
AreJointsModified() and then marks every segment dirty. Hand-composing skips
that, so you read whatever cache is sitting there. On the MASTER that was
invisible -- the renderer/cockpit camera call GetSegmentToWorld for the local
mech every frame, AFTER the local torso pushes its joint, so the cache was
already correct. A REPLICANT gets no such refresh: its cache stayed at the
BIND POSE, and the twist never reached the launch direction.
FIX. Use the engine accessor, and set the joints-dirty flag first so it
actually refreshes (by fire time the frame's render pass has already consumed
and cleared it -- measured jointsDirty=0 on BOTH nodes).
RESULT (same bench):
REPLICANT max 0.0000 -> 2.1145 mean 0.0000 -> 0.8252 0% -> 64%
PARTIAL, and I am not claiming otherwise. 36% of peer salvos still read the
exact-zero stale signature while the master is 100%. Forcing every per-joint
`jointModified` flag as well (GetSegmentToParent's own gate, SEGMENT.cpp:196)
was tried and moved the number by NOTHING -- 64% either way -- so the residual
is a different cause, most likely frame ORDER (the salvo mirror running before
the copy torso has posed that frame). Cheap form kept.
Also fixes a SAMPLING TRAP in the torso probe: PushTwist sampled one shared
static every 30th call, and with a master torso and a copy torso ticking 1:1
every 30th call is always the SAME instance -- so the probe showed only the
local untwisted torso and hid the copy's writes entirely. Now sampled per
instance-kind, which is what made the copy's correct joint writes visible and
moved the search downstream to the segment cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC
|
||
|
Joe DiPrimaandClaude Opus 5
|
1ae57398f1 |
#147 range caret: NaN poisons a process-lifetime static -- and the caret's input was never logged
Oracle: "no range finder on this drop" + a screenshot -- tick marks present,
moving caret absent, one drop, only tester affected.
Not the host, not the chassis, not his destroyed HUD. From the four field
logs: he WAS hosting (`[lobby] host:` appears only in his log) but range
computed fine on his node (1806 nonzero samples) and the reticle built on all
6 drops; a second tester flew a Thor the same night without hosting and saw
nothing, and the ladder is shared HudSimulation/BTReticleRenderable, not
per-chassis content; his HUD was destroyed twice but for 11s and 26s only, and
a destroyed HUD costs the LOCK (_DAT_004b7ec4 = 0.75), not the caret.
THE DEFECT. sShownRange -- what the caret binds to -- is a function-level
static in mech4's targeting step: one cell for the whole process, shared by
every mech, carried across drops, never re-seeded. NaN is ABSORBING in
step = trueRange - sShownRange;
if (step > maxStep) step = maxStep; // false for NaN
if (step < -maxStep) step = -maxStep; // false for NaN
sShownRange += step;
so one poisoned frame makes it NaN for the life of the process. The consumer
repeats the mistake -- BTReticleRenderable::Draw clamps with the same two
comparisons -- so NaN reaches AddPoint/ConcatMatrix and the caret + its bar
become degenerate geometry that STOPS RENDERING, while every static reticle
element including the tick marks still draws. That is the reported symptom
exactly, and it is sticky until relaunch.
WHY NO LOG COULD SETTLE IT. The caret's actual input had NO diagnostic
anywhere: BT_RANGE_LOG instruments the PICK (#4), and [target]'s `range=` is a
SEPARATE locally-recomputed Sqrt in the weapon-range check -- neither is
sShownRange or gBTHudRangeStorage. Grepping the field logs for NaN returns
nothing because the poisoned variable was never printed. Absence of the
signal was not evidence of absence.
FIX (4 parts):
1. re-seed sShownRange when the viewpoint mech CHANGES, so a new drop starts
at the binary's 1200 default. Deliberately NOT on respawn -- that reuses
the entity, and the binary does not reset the readout on respawn either.
2. producer NaN trap -> re-seed to 1200 instead of propagating.
3. NaN-safe consumer clamp (test x == x first) -> fall back to the authentic
no-target peg rather than rendering nothing.
4. BT_RANGE_LOG now prints the caret's real input at 1 Hz plus a
"[range] NaN TRAPPED" receipt, so the next field log CAN settle it.
STATUS [T3 on the field link]. The defect and the symptom match exactly and
the fix stands on its own merits -- a process-lifetime static feeding unguarded
float geometry is a bug regardless. But the causal link to Oracle's report is
INFERENCE: the NaN source is unidentified and this has not been reproduced.
Field-verify with BT_RANGE_LOG=1.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC
|
||
|
Joe DiPrimaandClaude Opus 5
|
7b003243ae |
#146 respawn: release the DESKTOP throttle -- and close #137, which was never a bug
#137 ("respawn came back with MYOMERS heat MAXED", Oracle; "overheating generator D", Sauron) sent us through a full two-sided audit of Mech::Reset and the whole RTIS chain. Both sides were correct. The answer was in the field log all along, one line after the reset: [respawn] Mech::Reset 3:30 healed+moved to (...) alive=1 [techstat] ... every live condition CLEARED [techstat] Myomers condition 3 SET <- Overheating, immediately [mppr] in thr=1 -> ... [gaitSM] cycleSpeed=14.6 state=12 <- already RUNNING [techstat] Condenser5 condition 3 SET <- "dumping into coolant loop 5" [techstat] GeneratorD condition 3 SET <- Sauron's generator D The mech respawns STILL UNDER POWER and earns the heat honestly. Two facts close it: 1. condition 3 is an OPERATING flag, not an alarm. Census over one match: LLaser_2 33 SET / 33 CLEARED, LLaser_1 31/31, SRM4 26/26, PPC_2 18/18 -- every volley trips it and clears it. EVERY subsystem is balanced (GeneratorD 5/4, Myomers 5/4, Condenser5 1/1; the extra SET is only the log ending mid-heat). cond 6 BadPower behaves the same (Myomers 8/8). Nothing latches. A post-respawn SET is not evidence of anything. 2. Mech::Reset's subsystem loop starts at index 2 and the ControlsMapper is index 0, so the throttle is never reset -- and the BINARY does the same. That is right for a pod: the throttle is a PHYSICAL lever still under the pilot's hand. Respawning under power is authentic and stays. Oracle's read that the myomer heat rate "felt right" was correct. WHAT IS a real defect (#146), desktop only: the glass bridge merely EMULATES that lever, with the static ramp accumulator sLever (mech4.cpp:3250) zeroed ONLY by the X all-stop and a direction-crossing snap. A pad/keyboard pilot is physically holding nothing and cannot see the lever, so they respawned at speed for no reason they could perceive -- and ate the heat load above. The Thrustmaster/RIO path was never affected: InterpretControls (@004d2150) rebuilds throttlePosition every frame from the databound throttleForward. Fix: queue the existing all-stop at Mech::Reset, reusing the proven path (it already clears the zero-crossing detent too). LOCAL VIEWPOINT MECH ONLY -- gBTDrive is the local bridge's state and Reset also runs for replicants, so an ungated write would all-stop the player whenever a REMOTE mech respawned. Pod-safe besides: with a RIO present the key bridge is off and gBTDrive.throttle is never read. BT_NO_RESPAWN_THROTTLE_RELEASE=1 reverts. Benched 2-node (scratchpad/night13/throttlerespawn.sh): the release fires 1:1 with local respawns on both nodes independently (A 2/2, B 1/1) and never spuriously. HONEST LIMIT: the viewpoint gate was NOT stressed -- B ran Mech::Reset 0 times for A's mech, so the remote-respawn path never fired. The gate is correct by construction (the isPlayerMech idiom), not proven. BT_AUTODRIVE cannot test the lever itself (forced mode reads forcedThrottle, never sLever), and the zeroing path is the X button, proven in the field. Also keeps BTReportHeatAtReset (heat.cpp, BT_HEAT_LOG): the [heat-t] census runs on a 5s timer, far too coarse to sample AT the reset. It is what proved every roster subsystem including all six Condensers sits at T=77 start=77, and it corrected an earlier false negative from filtering on IsDerivedFrom(HeatSink). KB: context/decomp-reference.md gains the routine/self-clearing condition semantics + this post-mortem, so it is not re-chased; cross-ref in context/gauges-hud.md. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018SgmXGNMXavXiafKXf9MDC |
||
|
Joe DiPrimaandClaude Opus 5
|
1f0923747b |
Mech::Reset: restore the POSTURE clears the port had dropped (#142)
Oracle: "crouch wasn't resetting on respawn ... mechs always spawn standing".
Correct -- Mech::Reset (@0049fb74) stands the mech up and the port cleared
none of it:
*(this+0x398) = 0 duckState
Set_Alarm_Level(this+0x39c,0) legStateAlarm -> standing
Set_Alarm_Level(this+0x714,0) bodyStateAlarm -> standing
*(this+0x650/0x654/0x658) = 0 death + leg/body reset latches
*(this+0x5ac) = 1.0f idleStrideScale
A pilot who died CROUCHED came back crouched -- leg parked in 'sqd' -- and now
that the cockpit strip works, showing the up-arrow "press to rise" frame on a
standing mech.
Benched (crouchrespawn.sh): A squats, dies while down, respawns -> legLvl 0
(standing) after Mech::Reset. Weak but the failure mode (stuck legLvl=1) is
absent. NB the first attempt was void: force-damage kept A dying before it
could crouch (legLvl 22/24 = death clips), so the run tested a STANDING death.
Switched to self-damage so the mech is stopped long enough to crouch.
Also carries the #142 gauge work: the crouch strip is a BUTTON-STATE indicator
(grey unavailable / orange down-arrow ready / orange up-arrow crouched),
decoded by rendering BDUCK.PCC rather than inferring it; and the gauge
factory's missing-image path no longer uses the no-op DebugStream.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NCJQkvq6G2JNrpVbA75tVZ
|
||
|
Joe DiPrimaandClaude Opus 5
|
03c4d55672 |
#142 crouch: duckState is a THREE-state posture -- the strip is an animation
Operator confirmed on screen: bduck.pcc is a real duck ANIMATION, and stepping
duckState 0->1->2 plays it. So the attribute is not a flag:
0 = standing 1 = moving between 2 = crouched
Everything else was already right -- asset, element (OneOfSeveralPixInt
@004c5204), factory registration, L4GAUGE.CFG:5001, and the attribute binding
(new [gauge] receipt confirms 'bduck.pcc' frames=3x1 attr=BOUND). We were
writing a two-value flag into a three-frame strip, so frame 2 was unreachable
and the cockpit saw a snap: "it lights up and sticks, no animation".
The handler is back to the binary's exact write (duckState = 1, @0049fa00).
That value now MEANS the middle frame, so the press gives immediate visual
feedback and the earlier toggle divergence is retired.
Needed a separate duckRequest cell, which I tried twice to avoid:
* duckState cannot be both the request and the display. Settling it to the
real posture destroys the request, so on the frame the squat clip parked
the consumer read "crouched + pending" and issued the opposite direction --
69 transitions from 2 presses, benched, twice.
* reading the CACHED legAnimationState instead of the alarm made it worse:
the cache refreshes only at the top of AdvanceLegAnimation, so right after
SetLegAnimation it still reads the old state. Read the alarm.
duckRequest is port-only, appended, never read by offset.
Also fixes a silent failure in the gauge factory: the missing-image path used
DebugStream -- the no-op ReconStream (project gotcha) -- so a strip that failed
to load reported NOTHING. Now DEBUG_STREAM, plus an ungated one-line receipt
per element naming the image, frame grid, port and whether the attribute BOUND
or came back NULL. That receipt is what proved the element was healthy and
sent me looking at the value instead of the plumbing.
Benched (crouch142.sh, madcat): 2 presses -> exactly 2 transitions,
SQUAT -> parked (settles to 2) then RISE (settles to 0). Refusal while moving
still holds (posture=0, authentic per Lynx).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NCJQkvq6G2JNrpVbA75tVZ
|
||
|
Joe DiPrimaandClaude Opus 5
|
fcd1a0ca8d |
#142 crouch: refuse-and-snap, not queue -- and honour the must-be-stopped rule
Fixes a regression I introduced in
|
||
|
Joe DiPrimaandClaude Opus 5
|
59f53da07b |
#142 crouch: duckState is the POSTURE the cockpit animation reads
The crouch symbol animation is fully present and we were starving it.
content/GAUGE/BDUCK.PCC the 3-frame strip
OneOfSeveralPixInt @004c5204/@004c52d8 the element, reconstructed
btl4grnd.cpp:144 registered in the factory
L4GAUGE.CFG:5001 oneOfSeveralPixInt(
E,ModeAlwaysActive,
bduck.pcc,3,1,DuckState)
ATTRIBUTE_ENTRY(Mech, DuckState, duckState) attribute 0x37
A 3-frame mech symbol beside the CROUCH button, indexed by duckState -- the
standing<->crouching animation Lynx and Draco describe. It never played
because the consumer zeroed duckState the frame after the press, in BOTH
directions, so the strip sat on frame 0 with a one-frame blip to frame 1.
That is the field report verbatim: "button flickers sometimes on press ...
state does not change. Remains in stand mode."
THE ZEROING WAS OURS. Every writer of +0x398 in the export is the
DuckRequest handler (=1) and Mech::Reset (=0). FUN_004a9b5c -- the master
perf, which contains the address the old comment cited as "the DuckRequest
consumer (@0x4aa011)" -- does not reference 0x398 at all. mech.hpp's own note
already said "duckState has NO code reader anywhere in the decomp ... whatever
consumes it consumes it through DATABINDING". The databinding consumer is
this gauge strip, and we were clearing it behind the gauge's back.
Restructure: drive on DESIRED vs ACTUAL. duckState is the desired posture;
the parked leg alarm is the actual. Act only on a mismatch -- no re-fire, and
nothing clears the attribute. A frame where mapPosture is not ready now
RETRIES (throttled [duck] WAITING) instead of silently dropping the request,
which retires the old "request consumed, posture=N" miss as well.
ONE DOCUMENTED DIVERGENCE: the handler now TOGGLES. The binary writes a bare
1 and clears the cell only in Mech::Reset, with no per-frame reader, so a
second press could never rise -- and a pod pilot's second press must un-crouch
(Lynx: "Mech is immobilized until crouch is pushed again, and mech rises").
One cell, same meaning, noted at the site.
Benched (crouch142.sh, madcat):
duckState -> 1 (crouch) -> SQUAT -> [holds 1 while crouched] ->
duckState -> 0 (rise) -> RISE
Value now persists across the crouched period instead of blipping, so frames
0/1 of the strip are reachable and stable. Also removed the interim REQUEST
DROPPED receipt: after the restructure nothing is dropped, and a receipt that
says otherwise is a trap for the next session.
STILL OPEN on #142: no immobilization while crouched (Lynx) -- nothing gates
movement on duckState or the parked leg alarm. A driven mech with a parked
leg channel is the [skate] signature (#52), so it may not be cosmetic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NCJQkvq6G2JNrpVbA75tVZ
|
||
|
Joe DiPrimaandClaude Opus 5
|
0530366687 |
#142 crouch: the mech is fine -- it is a missing PANEL ANIMATION
Benched solo AND in MP on the same chassis: both presses reach
DuckRequestMessageHandler, zero drops, SQUAT -> squat clip parked -> RISE.
Locomotion is not the bug, and MP is not refusing it.
Added an ungated [duck] REQUEST DROPPED receipt at the consumer's silent miss.
The squatCapable==0 path skips the consumer entirely AND leaves duckState
latched at 1 with NO log today; the posture-gate miss logged only under
BT_DUCK_LOG, which no player sets. Neither fired on madcat.
What the pilot sees, traced with BT_LAMP_LOG: the button lamp is momentary
press feedback, not state --
PRESS -> [lamp] 0x13 <- 0x3c
SQUAT -> mech crouches, clip parked
RELEASE -> [lamp] 0x13 <- 0x14 <-- while still CROUCHED
so crouched and standing look identical.
Era testimony corrects the scope: the button should ANIMATE A MECH SYMBOL
beside it, standing <-> crouching (operator). Lynx: 'When a mech stops,
crouch button lowers its stance and plays crouch animation. Mech is
immobilized until crouch is pushed again, and mech rises.' Draco concurs.
Two real gaps, neither fixed here:
1. no immobilization while crouched -- nothing gates movement on duckState
or the parked leg alarm. NB a driven mech with a parked leg channel is
the [skate] signature (#52), so this may not be cosmetic.
2. no stance symbol -- no gauge element draws one, and the decomp carries no
crouch/squat/stance/duck graphic string, so it is an authored IMAGE on the
secondary MFD; find it in that gauge's element list, not by string.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NCJQkvq6G2JNrpVbA75tVZ
|
||
|
Joe DiPrimaandClaude Opus 5
|
61f21107b4 |
#108 THE EJECT GHOST: the death-edge latch tested the wrong field
One substitution, three field symptoms. Mech::TakeDamageMessageHandler arms
the whole death tail -- kill report, VehicleDead, death blast -- from a
was-alive-at-entry latch:
const int deathBlastArmed = !IsMechDestroyed(); // graphicAlarm >= 9
The binary tests movementMode 9|10 there (@0x4a0303). The port swapped in the
graphic alarm and justified it: "the death transition sets mode 9 synchronously
with the structural flag on every path through here, so the edges coincide".
True of every DAMAGE path. False of the one that matters:
Mech::EjectPilotMessageHandler raises graphicAlarm to 10 (the EJECT state)
BEFORE dispatching its self-damage, while movementMode is still 1. So on an
eject the handler entered already reading "destroyed", the latch never armed,
and the death tail was skipped entirely -- including VehicleDead, which IS the
respawn trigger.
Everything the field reported on night 13 follows from that:
* "they all self destructed with panic button and didn't respawn properly"
-- no VehicleDead, so no drop-zone hunt, so no respawn;
* the EJECT GHOST -- the peer wrecks the mech and never un-wrecks it, because
the un-wreck rides the master's respawn. Normal deaths replicated fine all
along (9 deaths -> 8 un-wrecks, benched), which is why only ejects ghosted;
* the manual chart's "-1000 ejecting" never materialised -- the negated kill
award and the death cost both live in the tail that never ran.
Fix: use the binary's own predicate. MovementMode is untouched by the eject's
alarm write, so the latch arms on an eject exactly as on a combat death.
WHY SEVEN RIGS MISSED IT: the punch-out was being REFUSED, not undelivered.
EvaluateEjectPermission (@0049fa1c) grants only on
liveWeapons < ejectMinWeapons || liveGenerators == 0 || coolantFrac < 0.05
|| (leg-gimped && !simLive)
-- armour damage satisfies none of them, and every bench ejected a healthy
mech. An [ejecttest] receipt (2 lines) proved the dispatch fired every time
and the handler declined; the "[eject] REFUSED (mech not crippled enough)" line
was sitting in the very first bench log, ungrepped. BT_KILL_SUBSYS's
comma-list form ("GeneratorA,GeneratorB,...") was already built for this bench.
Verified 2-node (scratchpad/night13/ejectreal.sh), before -> after:
PUNCH-OUT landed 0 (785 refusals) -> 1, charge=500
peer wreck-enters 1 -> 1
peer UN-WRECKS 0 (the ghost) -> 1
eject score (type=2) absent -> award=-1000.00, score 1000 -> 0
death cost never ran -> APPLYING, penalty=500
That -1000 is the manual chart's eject row to the digit: killBonus 500 plus the
500 self-damage tally, negated.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NCJQkvq6G2JNrpVbA75tVZ
|
||
|
Joe DiPrimaandClaude Fable 5
|
f44be87ab2 |
#60 PART 2: the RE-EXPORT -- dark code 90KB -> 41KB, coverage 87.3% -> 93.5%
Installed JDK 21 + Ghidra 12.1.2 (no admin, %LOCALAPPDATA%\bt411-tools beside DXSDK/cmake; runner uses 8.3 SHORT paths because Ghidra's .bat expands %JAVA_HOME% unquoted and the profile has a space). New tooling: reference/ghidra_scripts/ExportGaps.java -- ExportAll's exact output contract PLUS a gap-fill pass (force disassembly + createFunction at E8 call targets outside functions, data->code pointers at a plausible prologue, and the census's discovered starts; iterated to a fixpoint, logged to gapfill_report.tsv). tools/ghidra_reexport.sh (headless runner, 'reprocess' mode) and tools/gapdiff.py (score two censused exports); gapcensus.py now censuses any export dir. Results: 6267 -> 6472 functions (+205 created in 2 rounds: 195 census starts, 6 call targets, 4 data pointers; 56.1KB newly covered), ZERO decompile failures. Dark real code 90.4 -> 40.8 KB (54.8% recovered); game-side dark 53.1 -> 21.1 KB; regions 428 -> 321. EVERY historically dark function now has pseudocode -- including @0x4c05c4 VehicleDead, the absence that opened this issue. VALIDATION: the new pseudocode confirms this week's hand reconstruction of the crouch field-for-field (mapPosture/duckState/squatCapable/myomerEff/ novice gate/SetLegAnimation/ForceUpdate/stability alarm) -- and exposed one branch the raw pass missed: AIRBORNE AUTO-RISE (mode 3|4 && legState 1 -> forced squ), now implemented in mech4.cpp and re-benched un-regressed. PROMOTION: the re-export is canonical reference/decomp/; the previous export is preserved at reference/decomp/archive_2025export/ so old `part_0NN.c:LINE` citations still resolve (addresses are stable across both; line/shard membership is NOT -- cite @ADDR). New lead recorded: @0x4c0904 is the MASTER BTPlayer Performance (team resolution, EndMission console post, score heartbeat) -- our @0x4c083c PlayerSimulation attribution needs a re-check. KB: source-completeness, gotcha #20 (the rule is cheap now -- look it up), CLAUDE.md router/layout. Log: phases/phase-04-gap-census.md. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
e1c3f2db6a |
Myomer factor: correct the false 'feeder unreconstructed' claim + wire the
crouch gate to the LIVE drive value The myomer system was ALREADY COMPLETE (2026-07-31 seek audit): Performance wrapper @004b8b9c, AvailableOutput @004b8ac0 (gear clamp x quadratic heat degrade x (1 - zone damage)), and the master-perf chain walk + speedDemand scale + turn freeze in mechmppr.cpp:990 -- the same @0x4a9cf2-0x4a9da4 bytes the crouch dig re-decoded. The 2026-08-05 banners calling the feeder dark were an export-gap-blind grep (named members, not offsets). Fixes: mechmppr publishes the chain MAX into mech->myomerEffectiveness (the binary's +0x79C home) so the crouch posture gate reads the live factor (dead/overheated myomers now genuinely refuse squat/rise -- previously the gate read a neutral 1.0 and never fired); the duplicate speedDemand multiply in the posture block is removed (mechmppr's is the one application); banners and locomotion.md corrected. Squat re-benched green on the live wiring. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
591d205b19 |
CROUCH complete: full cycle + MP replication verified
The WIP's 'pose does not hold' was a chain of bench-instrument errors, not a code bug: every capture ran in COCKPIT view (the pilot cannot see their own legs; the eye-height residual masked as reversion). Joint probes prove the park holds indefinitely (knee 1.138, root -2.219 steady); the 2-node bench shows the observer's replicant fully crouched and held (duckmpA_031 -- the type-3 state record carries it with zero new replication code); the second scripted press (new BT_BTNTEST2 env) verifies RISE -> standing zeros. MP button delivery confirmed mode-mask-clean (the one miss was round-start jitter). Diags added, all BT_DUCK_LOG-gated: SetLegAnimation re-arm tracer, 1 Hz joint probe, RIO press mode-mask, BT_TREE_LOG topology dump, and the squat-park log. RESIDUAL filed: pilot's own cockpit eye does not ride the root drop (DPLEyeRenderable chain composition; cosmetic). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
dd70061e0e |
CROUCH reconstruction (WIP): the master posture/duck machine, decoded + wired
Raw-disasm of the dark master-perf region (@0x4a9cf0-0x4aa0af): the myomer effectiveness factor (+0x79c, MAX over heatables' +0x31c, scales speedDemand -- feeder @004b8be3 unreconstructed, neutral 1.0 [T3]), the posture selector (+0x3f8: mode/novice/leg-state/myomer gates -- novices cannot crouch), and the DuckRequest consumer (standing -> SetLegAnimation(2) 'sqd', ducked -> SetLegAnimation(3) 'squ', ForceUpdate 8+1 ships the type-3 state record, stability alarm flips, request consumed). +0x1DC = mountSegment... er, the searchlight learned that one; here: mapPosture @0x3f8 + myomerEffectiveness @0x79c members land; value-space note (port normal mode == 1, binary 0). VERIFIED: request->consumer chain fires ([duck] SQUAT), the sqd clip plays (22kf/7joint, ends root -2.22 crouched -- clip data parsed from BTL4.RES, squ is its exact mirror; loader slot map re-verified byte-exact). OPEN: the parked crouch pose does not HOLD on screen (reverts ~1 frame after clip end with NO SetLegAnimation re-arm logged) -- the hold's render path is the remaining dig; [duck] re-arm tracer left in SetLegAnimation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
91bd28669e |
#45/#134 authentic score/death report tail -- replaces the scoring stand-ins
Reconstructs the dark-gap tail of Mech::TakeDamageMessageHandler
(@0x4a02f4-0x4a0890, raw disasm): the three id-0x16 score reports (kill to
the shooter's player / type-0 wire-fidelity / received to the victim's
player) and the BT 0x38-byte VehicleDeadMessage extension {killed-by player,
kill zone} dispatched from the death tail. Retires BTPostDamageScore /
BTPostKillScore and the per-hit inflicted credit (never existed in 1995:
@0x4c0200 is bound in no handler-table entry -- byte-scan receipt in
decomp-reference). Suicides now dispatch and the handler negates the award
(the #134 panic penalty). Collision divert falls through to the death tail
per @0x4a0375 (wall deaths respawn + blast; no score). ScoreMessage fields
renamed to decoded truth (vitalHit/zoneIndex/subsysID) + wire asserts;
console VTVDamaged points_transfered corrected (Round(award), not Now()).
Benches: scorekill.sh cross-node kill (kills=1 award=4.88, killedBy=2:1
zone=3, single death cycle), scoreself.sh suicide (type=2 award=-39.00
kills=0), deathblast2.sh re-verified (72 bursts at ~9u).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Fable 5
|
297127d0d7 |
#89 DEATH BLAST reconstructed: a dying mech splashes its neighborhood
The missing half of Advanced Damage, found by call-scanning Explosion::SplashDamage @0042fad0: TWO callers, not one -- Missile::Perform (the known #62 path) and 0x4a0bda, the UN-EXPORTED tail of Mech::TakeDamageMessageHandler itself. Raw disasm @0x4a07b8-0x4a0bda: when the victim ENTERS dead(9)/eject(10) during the applications, the binary sets the wreck burning (id 0x17, deferred -- handler not yet reconstructed), spawns the death Explosion (model 0x31 -- our death-list visuals stand in), and SPLASHES: gates : owning player's advancedDamageOn (+0x264) AND NOT suppressConsole (+0x258 -- eject sets it: PUNCH-OUTS NEVER BLAST, the authentic anti-suicide-bomb rule) damage: type 2 Explosive, amount = deathSplashDamage (mech+0x520), bursts = round(0.001 * moverMass * 15.0) -- scales with tonnage radius: deathSplashRadius (mech+0x524); per-victim falloff bursts/dist^1.25 in the shared core Draco's collision-divert suspicion is settled: the blast is TYPE 2, the divert never touched it -- the tail was simply never reconstructed. Port: deathSplashDamage/Radius PROMOTED from the Wword scratch bank to named Mech members (the bank is one GLOBAL array -- authored per-chassis values were clobbered to the last-loaded mech); BTSplashCore split out of the #62 weapon splash and shared; BTApplyDeathSplash + the death-edge arm in the handler tail; BTPlayerConsoleSuppressed bridge (friend). Bench (2-node, B parked 8.9u from a self-destructing A): blast fired with authored madcat data (radius=50, amount=5, mass=75000 -> 1125 base bursts), B took 73 bursts (falloff exact: 1125/8.91^1.25), cross-pod delivery + cylinder spray verified on B's own log ([dmghit] type=2 burst=73 across zones). ~365 damage at 9u -- Draco's 'double kills on drops' economy restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
546aabd5ba |
#131 false lock FIXED: miss-means-miss -- the pick answers only for drawn geometry
Night-12 field report (Ronin/Conn Man/Oracle, blackhawk-correlated): lock ring lit with the reticle visibly off the mech + no-reg complaints. Root cause: TWO port stand-ins answered where the 1995 card (which cast against the DRAWN geometry) would miss -- the pick's any-object sphere fallback and the caller's whole-mech AABB fallback. The regime that exposes them: a LEVEL boresight over a SHORT mech -- the blackhawk's mesh tops out below eye-ray height, so the ray clears every triangle but pierces the fat cull spheres; the ring lights with the reticle above the mech's head (the operator watched exactly this on the sweep bench). Careful aimed-down fire rides triangles, which is why Oracle's per-panel audit passed on the same build. Fix: MechSegmentPick returns 1=drawn-geometry hit / 0=TRUE MISS / -1=no render tree; the sphere may answer ONLY for a mesh the reader cannot parse (pm==0 -- currently none exist: counters objs/invFail/noTri all clean); the AABB survives ONLY as the pre-tree replicant grace. A readable mesh the ray misses is a MISS -- no lock. Verified (2-node vs bhk1 at 100u, all runs on force-relinked string- verified exes after today's stale-link flake): - LEVEL lock-sweep: 0 locks all run (pre-fix: lock band from 168 sphere answers; picksrc tri=0 sphereFB=168). - DOWN-PITCHED sweep: locks return 100%% tri-sourced (tri=158 sphereFB=0), landing on real parts (rarm/ldleg/rdleg) with honest gaps. - Full zone-walk matrix: tri=18874 sphereFB=0 box=0; victim took 156 hits across 16 zones incl. both side torsos -- combat un-regressed. New instruments (all env-gated): BT_LOCK_SWEEP=<axis> torso pan (the operator-visible lock-envelope bench), [locksweep] transition log, BT_LOCK_ENVELOPE synthetic unit-sweep probe, [picksrc]/[pickbox] source telemetry with objs/invFail/noTri localization counters. Bench: scratchpad/night12/zonewalk_bhk.sh. NOTE for the field: locking is now strictly TIGHTER (ring = reticle truly on the machine). If era testers feel the pods were more forgiving, Draco's "slight lock linger" memory becomes a deliberate investigation (sourced hysteresis), not an accidental sphere halo. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
26ab2fee3a |
Respawn-reset audit: valve restore + peer smoke cleanup (both binary-grounded)
Operator reports audited vs the binary (full matrix in RESPAWN_REARM_PLAN addendum): - VALVES (real gap): Condenser reset @004ae534 was missing from the decomp export -- raw disasm shows it chains HEATSINK (coolant refill runs; the old body chained HeatableSubsystem per the stale TCP shard) then, respawn- side, resets valveState to detent 1 and restores massScale from refrigerationFactor. Mech::Reset now also runs the binary's tail call (@0049f788 BTRecomputeCondenserValves) so flow fractions rebuild from the reset detents. Bench: detent 5 -> death -> "[respawn] Condenser1 valve detent 5 -> 1". - #129 SMOKE (real gap, peers-only): the replicant un-wreck edge rebuilt the model without the @004d0c14 per-entity effect cleanup, so the observer's last 10s wreck-plume window rode the teleport onto the fresh mech. BTStopEntityPfx now runs on the edge; bench shows no plume line after any un-wreck until the next death. - AUTHENTIC (no fix): weapon->generator taps persist (@004b0e6c only resolves the link) and MFD display/control modes persist (mapper vtables 0050f45c/0051e440 slots 8-11 = plain root bodies, read from the exe). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
1f995ee35a |
#124 twist-sign VERIFIED correct (live missile bench) + the rig to do it
The frame-adapter's inferred twist-sign flip (SelectSlice theta -= twist, vs the binary's += pre-reflection) was the last unverified half of #124 -- every earlier probe ran at twist 0. Bench: stationary madcat target with the torso PINNED at 0 / +140 / -140 deg, LRM salvos from a fixed shooter (missiles = the authentic cylinder path; the binary DROPPED zone -1 beam damage). Result: slice picks track the physically-facing flank in BOTH directions (twist-left -> right-family zones for left-flank impacts, twist-right -> left-family), deterministic, wrong-sign outcome (slice 7 vs observed slice 1) clearly excluded. No game-code change needed. Instrumentation added (all env-gated): - torso.cpp BT_FORCE_TWIST=<-1..1>: HOLD the sim's analogTwistAxis (the input-level pin was dead -- live input rides the CONTROLS.MAP device push, and Basic mode auto-centers; sim-level is plumbing-independent). - dmgtable.cpp [slice] line: rot flag, live twist, thetaIn/thetaAdj, chosen slice -- the weighted leaf roll made zone-only logs ambiguous. - [dmgresolve] now names the zone (BTMechZoneSegAndName). - mech4.cpp BT_FORCE_TWIST input pin + one-shot mode cycle (kept as doc of the dead path), mechmppr [mppr] mode probe. - scratchpad/night11/twistsign.sh: the 3-config bench. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
42b0691592 |
#119 leak-voice stutter SETTLED: authentic -- the wooHoo latch is authored OFF
Deterministic leak bench (BT_KILL_SUBSYS partial form, "Condenser4=0.5":
zone level without the crit -- a clean leak source) + the aud-tail receipts
decode the voice stutter end to end: the warning is a phrase-sequenced
voice patch (Warnings01 zones as notes), and every techstat leak-bit edge
restarts/stops the sequence mid-phrase (authored zero-release = hard cut).
At a drained tank the draw HUNTS the authored 0.0025/0.003 band -> edge
streams -> progressive clipping as more systems hunt.
The anti-spam wooHoo latch would bound exactly this -- but no armer exists
anywhere in the flat export, and the authored tuning is minDur=0 range=0
chance=0 (dumped live): DORMANT BY AUTHORING. The clipping is the 1995
experience; the port's one real bug here was the 2x hunt cadence, already
fixed by the 28 Hz filter (#119,
|
||
|
Joe DiPrimaandClaude Fable 5
|
0089b3284f |
Eject firsthand prep: G = coolant Flush binding + BT_FLUSH_HOLD scalpel
The Flush ACTION existed in the input engine but no key was bound in the shipped CONTROLS.MAP -- desktop players could never hold the flush. G (free) now maps to it. BT_FLUSH_HOLD=<frames> extends the BT_FLUSH_TEST scalpel past its 60-frame edge test for tank-dry benches. Dry-run finding (fl.log): a full flush drains the tank to 0/14 in ~30 s, but does NOT arm eject by itself -- InjectCoolant moves tank coolant INTO the loops (the bank RISES), and idle heat draw is too slow to burn the loops down to the 5% arm threshold in minutes. Firsthand arming needs sustained fire after tank-dry (or coolant-loop leak damage / generator loss). Matches the binary's intent: eject arms when LEAKED nearly dry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
8165cee9da |
#118: the eject option comes alive -- per-frame arm chain closed (FUN_004a9b5c+0x10)
The evaluator's "unexported caller" found by E8-scan: the Mech MASTER PERFORMANCE (FUN_004a9b5c, in the 004a977x..004ab188 export hole) calls EvaluateEjectPermission as its FIRST act every frame. Without that call the port's ejectPermitted stayed at ctor-0 forever -- the panic-armed mode never raised, so the PANIC lamp never lit and testers "never saw the option for eject come alive". Restored at the top of Mech::PerformAndWatch (masters only, like the binary). Mislabel swept: mech+0x414 is ejectPermitted, NOT "missionReviewMode" -- the mapper's @004d196c edge-watch arms PANIC mode 0x200000 from it (the real review mode is the GLOBAL DAT_004fd550; btl4pb now reads that global, the GetMissionReviewMode stub is retired, mapper member renamed previousEjectPermitted). Verified live: kill 4 generators -> "[eject] panic-arm mode ON" -> the miniconsole Panic button lights dark->bright (pixel captures; the pad panel shades it from the same PadRIO lamp state the pod's physical button uses) -> BT_EJECT_AT press -> PUNCH-OUT -> respawn heals -> mode disarms -> healthy presses REFUSED. BT_EJECT_LOG=1 logs the arm edges. Note: no shipped gauge rides mode 0x200000 -- the 1995 eject indication is the button lamp; the weapon-eng MFD "UNJAM/EJECT" is the ammo-jam indicator. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
440cee1e4f |
#87 dig closure: host response is linear (raw decomp re-read); the pixel curve was div-card firmware
Re-read the material damage watcher RAW (@004573e4 ctor / @00457784 change-push): 16 floats snapshotted, damaged = pristine x 0.1 (literal at the MakeMechRenderables call site, watching zone+0x158), linear lerp, 13 floats written back, then FUN_0048d4d4 -- which is "flush_material": a dpl command-stream marshal (opcode 100) to the i860 division card. The material->ramp->texel response curve was CARD firmware and is not in BTL4OPT.EXE; logged as an open question (ask: firmware/DIV docs/period screenshots). Field reconciliation, measured live: one 25-pt hit (ERLG class) snaps a 77-pt arm 0 -> 0.3247 in a frame -- "a single missile leaves visible armor damage" holds under the linear curve for heavy hits; 2-pt ERS creep (the leg-audit weapon) is the invisible case. Bench additions: BT_WALK_ZONES=<name-list> walker filter (concentrated per-region audits, scratchpad/night10/legwalk.sh) + BindArmourDamage ORPHAN inventory under BT_ARMOR_LOG (proved every drawn op is zone-claimed on the full tree). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
00dfbf8f9f |
#124 RESOLVED against the authored skeleton: the pick's zone is the
struck drawn segment's dzone, ALWAYS -- the cylinder embellishment removed
The open question ('what zone did 1995 designate for hull hits') was
already answered in the KB's #73 dig [T0+T1]: no software pick writer
exists -- the pod's pick was a dpl scene intersection, struck triangle
-> DCS segment -> its SKL-authored dzone; the cylinder lottery was only
ever the UNAIMED path (missiles/splash/rams). The live segpick dump
completes it with the authored data: the madcat's pickable skeleton is
18 segments -- hip, shoulders, guns, thighs, knees, ankles+toes, and
ONE hull piece (jointshakey -> dtorso). No ltorso/rtorso/utorso/rear/
door/searchlight segments exist. Aimed fire at any torso panel
authentically designates CENTER TORSO; the side/rear/upper panels strip
via the unaimed path.
Changes:
* the shared-carrier -> cylinder override (yesterday's embellishment)
is REMOVED: triangle hit -> that segment's dzone, period;
* SegAimPoint: the walker's hull anchor moves from the joint ORIGIN
(crotch gap -- rays sailed between the legs to the terrain sentinel)
to the carrier segment's cull center (the chest);
* [picktri]/[pickwin] diagnostics grew tri/seg/zone fields -- three
consecutive 'mystery' regressions in this stretch were the STALE-EXE
trap (CWD-relative cmake --build from content/ resolved to a
nonexistent dir and the old binary kept running) -- the harness doc's
warning, hit again.
Final zone-walk matrix (spinning target, real MP, full cycles):
limbs 55/70 in-zone (legs/feet 6/6; arm strays = hull/occlusion);
hull-family aims -> dtorso 56/69; victim consumption matches. The
night-10 field table now reads as this model's fingerprint: 'only LCT
gets hits' from aimed energy IS the pod -- one hull piece, dtorso.
What was actually broken and is now fixed: the sphere mis-picks
(rgun-from-a-dtorso-aim class), the unaimed path's 180-degree frame,
and the vertical scatter.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Fable 5
|
1a3c268278 |
#124: the aimed pick now intersects the DRAWN GEOMETRY -- sphere
approximation retired; shared-hull hits route through the cylinder The zone-walk matrix caught the pick red-handed: with the servo verifiably holding the reticle on the dtorso segment, the per-segment BOUNDING-SPHERE pick returned rgun/ruleg -- the gun pods and legs thread the ray before the torso from most angles (its own comments admitted a foot could be unhittable behind its own knee). The 1995 pick was a dpl scene intersection against the drawn geometry. Restored that semantic: * BTGetPickMesh caches each segment d3d_OBJECT's triangles CPU-side once (its own BGF VB/IB, managed-pool locks); MechSegmentPick now runs sphere PRE-FILTER -> Moller-Trumbore nearest-hit across the threaded segments' posed meshes; the old smallest-sphere selection survives only as the no-triangle fallback. * Segments claimed as CARRIER by 2+ zones (the shared hull: madcat seg 4 carries dtorso+ltorso+rtorso+utorso+rears) cannot resolve one zone from geometry -- those hits return zone -1 WITH the accurate triangle point, and the victim's bit-verified (frame-fixed) cylinder assigns the panel by band/wedge. Unique carriers (legs, feet, arms, gun pods) keep the direct zone. * ZoneAimPoint: the walker aims at a zone's VISUAL center (largest pick object's cull-center) instead of the segment origin -- joint origins made feet/lower legs strike the part above. * Walker upgrades from live operation: 3-column truth (aim/pick/land), engage gate, damped servo with polarity watchdog, approach port. Zone-walk verdict (full cycles, spinning target, real MP): limbs 5-6/6 direct in-zone; every hull panel routes CYL; the victim's landed zones now include ltorso/rtorso/reardtorso/rearutorso/rearrtorso -- the panels night-10 reported unhittable. Residuals tracked on #124: rtorso/utorso aim-anchor placement, doors/searchlight small-zone sample, twisted-torso twist-sign verify. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
16e75d6e1a |
#124 bench: the ZONE-WALK MATRIX -- systematic per-panel targeting, operator-watchable
The precision rig the operator specified: two nodes, target visible, every zone aimed at deliberately through the REAL reticle/pick path. * BT_SPIN_SELF=<deg/s> (target node): rotates the node's own viewpoint mech in place (the BT_SPAWN_AT write pattern per frame) so every aspect passes the shooter's boresight; the operator watches this node's paper doll take the hits. * BT_ZONE_WALK=<secs/zone> (shooter node): walks the target's damage zones in order, resolves each zone's carrier segment's live world position (BTResolveSegmentWorld + the new BTMechZoneSegAndName bridge), SERVOS the torso twist + aim elevation until the centered reticle ray (BTGetAimRay) points at the segment, fires 3 laser pulses, advances. [walk] ZONE/FIRE/HOLD on A pairs with [dmghit] zone/level lines on B. * scratchpad/night10/zonewalk.sh: launch both + relay, map sed'd to grass/day, NO kill timer -- the session stays up for observation. Hard-won servo constraints (documented in test-harness.md so they are never relearned): the ENGAGE GATE (ray live + range<150 + bearing<1.1; outside it RELAX the twist -- servoing at the twist limit while the goto marches is a limit-clamp fight that visibly shakes the mech, and a 0.55 gate deadlocks against the goto's ~0.55 resting bearing); YAW POLARITY -1 (the twist cell's angular sense is opposite atan2(x,-z) world yaw -- operator-observed live, the SECOND witness for the #124 SelectSlice twist-sign flip) with a divergence watchdog that self-flips; damped correction (gain .40, cap .025/frame -- the aim ray lags the twist write a frame); and a one-shot approach PORT to 100u off the target instead of a cross-map march. First live run: walker cycles zones, settles, fires; the target's luleg climbed 0.458->0.523 under its own aimed pulses -- aimed shot, correct panel, damage consumed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |
||
|
Joe DiPrimaandClaude Fable 5
|
b0b7adc4ea |
#124 CORRECTION: the frame adapter is a Z-REFLECTION, not a pi rotation --
proven by the mech's own pods
The first fix (
|
||
|
Joe DiPrimaandClaude Fable 5
|
e996be249d |
EJECT/PANIC wired: Mech::EjectPilot (id 0x19 @0049f854) + the crippled-mech gate
The button died twice before reaching game logic: no handler (0x19 was
unregistered) and no sender (the pod's panic was a control bit, not a
mappable). Both halves reconstructed from raw disasm -- the handler AND
its permission evaluator sat in export gaps.
@0049f854 EjectPilot: press-only; gated on ejectPermitted (@0x414) and
!IsDisabled; console eject notice (relay wire = tracked tail; the
suppressConsole@0x258 latch that prevents the death double-notify IS
wired); graphicAlarm -> 10, which kills via the >=9 predicate; then a
self TakeDamage {inflicting=SELF, zone -1, Explosive, amount =
ScenarioRole::killBonus}. Role layout byte-settled via the role reader
@00429bec dest offsets + the ctor record copy: +0x1c IS killBonus, +0x20
is the 4.10-only SpecialCaseDeathPenalty, +0x28 returnFromDeath (all
prior citations reconciled). KillBonus authors NOWHERE in shipped
content -> the charge is 0 and the ALARM does the killing: our bench
outcome is the pod outcome.
@0049fa1c EvaluateEjectPermission: eject only from a CRIPPLED mech --
bank coolant fraction < 0.05 | zero live generators | live weapons
below mech+0x448 (no exported writer: zero, clause inert) | leg-gimped
novice. A healthy mech REFUSES the button; no free resets.
Input: binding-engine "Eject" action -- Backspace / pad LeftThumb
(default profile + shipped CONTROLS.MAP). Bridges per the databinding
rule: weapon/generator/bank/player reads land in their complete TUs;
MechSubsystem gains the both-cells destroyed accessors (gotcha #22).
Bench scalpels: BT_EJECT_AT=<frame> (path-identical synthetic press),
BT_KILL_SUBSYS now takes a comma list.
Verified single-node: healthy press REFUSED (x2), four generators
killed, next press PUNCH-OUT -> death, wreck smoke, respawn; the
respawned mech refuses again (permission re-evaluates after Reset).
Death rides the normal damage/death chain, so MP replication is the
proven path. Tails tracked: console relay notice, RIO 0x38 panic
control, alarm-10 eject audio/canopy, SpecialCaseDeathPenalty consumer.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Fable 5
|
152249cb76 |
#95: the SCOREBOARD banked one missile per salvo -- credit the DELIVERED amount
Two consumers of one impact, only one was ever verified: the victim gets
TakeDamage{amount=per-missile, burstCount=cluster roll} and applies it
burstCount times (armor was always right); the score post on the next line
sent the bare per-missile amount. An LRM10 salvo dealing 7-35 armor banked
3.5 points -- Rajel's "~3 points to score", to the digit.
Ground truth: the binary's score is the victim handler's tally (amount once
per applied burst @0x4a04da + crit bonuses) reported to the INFLICTING
player (the id-0x16 tail, deferred #45). The shooter-side stand-in now
posts amount x burstCount -- the identical figure handed to the victim, at
the identical one-post-per-TakeDamage granularity. Same pass:
* splash never credited score at all -- the binary tallies every
TakeDamage; now posted per splash victim (amount x falloff bursts);
* the bridge credited the LOCAL player for ANY registered hit -- AI-master
fire on the player, a dying mech's death-blast splash; now refused
unless the shooter IS the local vehicle (MP unaffected: only local fire
carries live damage on a node);
* direct-fire unchanged -- beams author burstCount=1 (emitter.cpp:355).
Verified per the harness doctrine: single-node field composition (madcat,
real fire, real enemy) 22/22 impact credits paired at damage x burst, zero
bare 3.33 posts; two-node replicant-victim run BOTH directions 31/31
paired across three missile authorings (3.33/2.0/5.0 per-missile) + 25-pt
ballistics, leftovers all burst-1 beam amounts. test-harness.md gains the
map=grass note (MP.EGG authors cavern/night; GOTO mechs shoot rock).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Opus 5
|
9b1353bfbf |
ONE Explosion per salvo -- the port detonated every visual round, amplifying explosion sound + smoke + audio demand by the missile count
The user challenged the 7% clock story ("i think its not just a 7% discrepancy.
did you base your analysis in the decomp") -- correctly. The stop-path trace
it forced found the real amplifier, and it is byte-grounded:
THE ARCADE (byte-verified in the #95 work): FireWeapon @004bcc60 spawns ONE
Missile entity per trigger -- no loop, missileCount never read at fire. One
Missile -> one detonation (@004be078) -> ONE Explosion entity: one 4-voice
sound + one authored smoke package per salvo. The cluster's N-fold nature is
delivered as DAMAGE (burstCount), not as N explosions.
THE PORT: the N-round visual ripple (a deliberate [T3] port visual; its rounds
are damage-0 tracers) called BTSpawnRoundDetonation at EVERY round's contact --
N full Explosion entities per salvo, N x BigExpInt x 4 voices, N x the
explosion PFX, N=2..20. Measured in a 100s solo missile bench: 559 explosion
sound setups; BigExpInt 208. On observer nodes the stale-point mirror bug
(fixed in
|
||
|
Joe DiPrimaandClaude Opus 5
|
6dbeb3d266 |
wreck smoke dies WITH the wreck; the audio/effect clock is calibrated to the byte-proven 28
USER REPORT: "when you destroy a mech, it continues to smoke for a very long
time after the mech disappears... up to 30secs to a minute."
MEASURED (new [pfx] timed census with per-def particle attribution): the
death plume (psfx 1 DDTHSMK) was re-armed on a 10s cadence as a WORLD effect
(ownerTag=0) -- so neither the burial transition nor the respawn cleanup
(StopAllEntityEffects) could touch it. The last-armed window kept emitting
over the empty spot for up to 10s after the hulk vanished, plus 6+-2s authored
particle lives, and back-to-back windows stretched the visible tail to the
reported range. The re-arm itself is a marked PORT ADDITION [T3]; the binary
fired DDTHSMK once -- its 10s window + particle lives ended right around the
~17s hulk sink. The authored design: the smoke dies WITH the burial.
FIX: the plume is spawned ATTACHED (BTStartPfxAttached, tagged to the dying
entity -- it also rides the sinking hulk now), and the burial transition calls
BTStopEntityPfx alongside the respawn cleanup that already did. Verified over
a 175s run containing FOUR wreck events (the enemy kill + three player bay-
fire deaths): every plume's emission ends at its wreck's burial or respawn,
particles fade within ~8-11s, no orphan emitters remain.
THE COUPLING (user relay of Oracle: "prolonged smoke and explosion INCLUDING
SOUND... same lever?") -- CONFIRMED as a family: the binary's audio layer
times in FRAME COUNTS converted by the engine frame-rate global:
AudioTime::Seconds_To_Frames == fmul [0x52140c]=28.0; fadd 0.5; round
(@0x42c611, @0x42dd86 -- the 0x42xxxx cluster of the ~90 global consumers)
Our engine's DefaultRendererRate said 30: every audio duration, sequence
delay and compression window ran ~7% fast. Calibrated to 28 [T1]. NOT yet
explained by this: a truly SUSTAINED/looping explosion sound (7% is not
"prolonged") -- the effect stop-path trace stays open on #114/#51.
Tooling: BTPfxParticle carries defIndex; [pfx] census (BT_PFX_LOG, 2s) prints
emitters + particles PER EFFECT SLOT with a timestamp; wrecksmoke.sh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Opus 5
|
da70bd58b2 |
#84: the replicant salvo now HOMES AT THE LIVE TARGET -- the stale-point double detonation is gone
Oracle (674): "missile appear to register hit explosions twice, once where
target was and again where the target is". Rajel (693): "once where the mech
was when I fired and on the expected impact".
MECHANISM, measured on two nodes (BT_PROJ_LOG "DET at" now logs EVERY contact
detonation, including damage-0 rounds, which were invisible before): the
replicant salvo mirror pushed rounds with NO entity handle at the FROZEN
fire-time aim point, and the pool's contact test is proximity to p.targetPos --
so mirror rounds "contacted" the empty air where the target USED to be and
detonated there, on every peer, every salvo (45 salvos -> 234 frozen-point
air bursts in a 200s bench). The true impact appears at the live position
(the shooter's own homing rounds on his node; the victim's damage reaction on
peers) -- two sites per salvo, exactly as reported.
THE BINARY'S MODEL (CLASSMAP, Missile flight-entity cluster): missiles are
ENTITIES with authoritative/GHOST Performance variants (Projectile
PTR_LAB_005129e8 / 005129f4) and their own WriteUpdateRecord (@4bef4c, slot 7)
-- peers ran ghost missiles updated from the wire and saw the TRUE trajectory.
One explosion, correct place, every node. The frozen-point mirror is the port
infidelity (the entity Missile itself stays blocked by the documented 2007
Entity-base mismatch; mislanch.cpp:301).
FIX: the MissileLauncher update record now carries the locked target's EntityID
(salvoTargetID; EntityID::Null = point fire). The mirror resolves it on the
receiving node via HostManager::GetEntityPointer (engine T0, an index-socket
Find -- NULL-safe on any ID), refuses non-mechs via BTIsRegisteredMech, and
pushes the mirror rounds WITH the handle -- they re-lead on the live local
replicant exactly as the master's own rounds do, and detonate at the true
position. Guards: record-length gate (a short record from an older build has
no ID field -- its tail would be garbage and must not resolve) + Null check +
registered-mech check.
VERIFIED (two nodes, 200s, missile autofire both ways):
* every record arrives len=52/52 with sane aims ([mlrec] diag)
* 22/46 salvos resolve (tgtID=3:22 -> the victim's local master) and their
mirror rounds home: the frozen-air DET population fell 234 -> 102, the
live-homing population rose to 155/207 per node
* the unresolved remainder is BENIGN and correct: EntityID::Null point fire
plus hostID=-1 LOCAL entities -- i.e. the victim's WRECK, which autofire
keeps shooting after the kill; a wreck does not move, so the frozen aim IS
its true position
Downstream expectations: Ronin's smoke-screen (#114) should drop by ~the mirror
half, and the audio census (#32) loses the duplicate Static3DPatchSource
explosion sounds on peers -- both to be read from the next field logs.
Also in this commit: [projectile] DET diagnostic (every contact detonation with
position/damage/handle/aim), [mlrec] record-arrival diagnostic (len/recID/
subsys/counter/aim, capped 200), scratchpad/night9/mp_double.sh.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
Joe DiPrimaandClaude Opus 5
|
3b7c19c232 |
#95: REVERT the salvo-damage hack; deliver the cluster the way the binary does
Supersedes the mislanch change in |
||
|
Joe DiPrimaandClaude Opus 5
|
efc3e9ff1a |
#95/#84: missile salvos deliver their FULL authored damage, and stop double-exploding
THE SALVO DAMAGE (#95). Players measured an LRM 15 landing "3ish points". The logs agreed: [projectile] IMPACT damage=3.33333 (Oracle, LRM15) and 3.5 (Rajel, LRM10). Those are right PER MISSILE -- 50/15 and 35/10 -- and the bench shows why the salvo still under-delivers: each launcher pushes N rounds of which exactly ONE carries damage. Two individually-correct changes composed into an N-fold shortfall: * The ctor does what the binary's MissileLauncher ctor does (@0x3ac/@0x3d4): damageData.burstCount = missileCount; damageData.damageAmount /= missileCount; The record holds the PER-MISSILE amount plus the count; the arcade reconstitutes amount x burstCount when it applies the hit. * Task #62 then correctly stopped the port applying the hit once per visual round (that was ~missileCount-x too lethal) by damaging only the lead round -- but handed it the already-divided amount. Our DamageZone::TakeDamage is `damageLevel += amount * scale` and drops burstCount, so the salvo delivered amount/missileCount. Since the port collapses the cluster to one damaging round, multiply the count back in there. Bench: an SRM6 salvo now lands amt=35 (the authored total) taking a zone 0 -> 0.556, where it previously landed 5.83. THE DOUBLE EXPLOSION (#84). Oracle: "missile appear to register hit explosions twice, once where target was and again where the target is." There are two spawn sites: BTSpawnRoundDetonation at the round's own impact point, and the message manager's bundled explosion at the CONSOLIDATED point a frame later. The duplicate was known and thought harmless -- "among a rippled volley it is invisible" -- which held only while a salvo landed N detonations. A projectile now marks its weapon (MarkRoundDetonated) and the consolidation skips queueing a second blast for it; direct-fire weapons never mark, so lasers/AC keep the bundled explosion they rely on. Bench: 4 missile impacts -> 4 SKIPPED, while 11 direct-fire hits still queue normally. SWEPT CONTACT (#84 tail). Contact was a 10-unit sphere sampled only at the END of each step. With the authored thruster live (#84) field rounds arrive at v=955 -- a ~16 unit step at 60fps, larger than the radius -- so samples can straddle the target. (Pre-#84 rounds flew ~100-300 = a 1.7-5 unit step and could never skip it: the velocity fix exposed this, it did not cause it.) Now tests the whole segment travelled and bursts at the point of NEAREST APPROACH, which also stops the detonation being flung past the target at speed. RETRACTION: I posted tunnelling as the leading explanation for the lost salvo. The bench disproves it -- zero fizzles, and the "missing" rounds are the dmg=0 visual rounds of the cluster, which never registered damage by design. The sweep is kept as speed-independent robustness, not as the #95 fix. Bench: scratchpad/night8/salvo.sh (BT_PROJ_LOG + BT_FIRE_LOG). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
Joe DiPrima
|
137c151951 |
#87: mech armour PANELS now darken with damage -- the .DZM material system was loaded and unused
Players: "the actual enemy mech in external view is not showing darkened armor panels". We swapped destroyed LIMB meshes but never darkened a panel. The 1995 game darkens armour through the MATERIALS. MakeMechRenderables (FUN_004cef28) built, per (damage zone, material), a watcher FUN_004573e4(material, &zone->damageLevel, 0.1f) that snapshotted the materials |
||
|
Joe DiPrimaandClaude Fable 5
|
5410371b0c |
#86: destroyed weapons can no longer fire -- the fire gates read a
never-written cell (the split-cell gotcha) Both weapon fire gates (ProjectileWeapon gate 1 @4bbd36, Emitter hard-failure @4baab9) test the binary's subsystem+0x40 for Destroyed(1). In the 1995 layout that offset sits INSIDE the embedded status alarm (statusAlarm@0x2C + level@+0x14 = 0x40) -- ONE cell, written by ForceCriticalFailure when a zone's crit cascade kills the subsystem. The port models the same address as TWO members: the AlarmIndicator AND a plain int simulationState@0x40. Every destruction path writes the ALARM (so the MFD draws its X correctly) while the gates read the plain int, which nothing ever writes -- so a weapon on a blown-off arm showed destroyed on every panel and kept firing and scoring, locally and on peers (night-7: all three testers, screenshots of a missile leaving a destroyed pod). The tell had been sitting in our own logs for weeks: [ammo] SRM6_1 -> NoAmmo (gate1): destroyed=0 ... on a mech whose launcher was X'd out. Fix: both gates now read statusAlarm.GetLevel()==1 as well as the int. Also added: the crit-cascade log names the destroyed subsystem, a BT_SELF_DAMAGE_ZONE=dz_* named-zone bench mode, and BT_KILL_SUBSYS=<name> (force ForceCriticalFailure on one named subsystem -- the exact call the zone cascade makes, so a bench can ask 'the panel says dead, does it still shoot?' without hunting for the zone that carries a given weapon). Verified A/B in ONE run: before the kill both SRM6 launchers fired 2 salvos each; after, the destroyed launcher fired ZERO (gate log destroyed=1) while its twin kept firing normally. KB: new gotcha #22 (the SPLIT CELL -- one binary offset, two port members, only one written; sibling of #1) + combat-damage entry. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J |
||
|
Joe DiPrimaandClaude Fable 5
|
4704ab3d41 |
#84: missiles ACCELERATE like the binary -- the pool flew at constant
rack-eject speed; the authored MissileThruster was never applied The binary Missile hosts a MissileThruster subsystem: authored acceleration for BurnTime seconds after the MuzzleVelocity eject. Authored values decoded from BTL4.RES (type-15 missile model records, reached from the AmmoBin's ammoModelFile via the type-1 MODELLIST indirection): SRM 2.5s @ 600 u/s^2, LRM 10s @ 300 (climb 50), Streak 3s @ 300 (turn 360deg/s), NARC 10s @ 300; splash 30 all (port constant was already right); authored drag ~0.001 = negligible. The port pool flew every round at CONSTANT |MuzzleVelocity| -- SRMs 100 u/s, LRMs 30 u/s (10x slower than authored) -- the entirety of the night-7 "missiles are very slow" report, and the driver of the "explosion before the missiles arrive" perception (the salvo-lead detonates while the slow spread rounds straggle in). Fix: BTMissileThrustOf (mech4.cpp) lazily parses + caches the RES thruster table (ModelList ids aliased to their type-15 member's burn/accel); both launch paths (master FireWeapon + the replicant salvo mirror -- peers see the same speed) resolve through the launcher's bin and pass burn/accel into the pool; the advance integrates speed += accel*dt while burn remains, heading preserved. Ballistic rounds (autocannon/gauss) pass 0/0 -- unchanged. Impact log now prints v=/burnLeft= evidence. Verified live (solo, BT_SPAWN_ENEMY + BT_AF_MISSILE): thrust table 8 entries cached; Black Hawk Streak resolve burn=3 accel=300; impacts at v=277 u/s (was a constant 100). Fun authenticity note: the Black Hawk's "SRM6" fires strk (STREAK) ammo per its authored bin. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J |
||
|
Joe DiPrimaandClaude Fable 5
|
ac7c46b87a |
#93: fix the end-of-round PerformAndWatch crash -- the STATIC projectile
pool held dangling entity pointers across round teardown Root cause (disasm-pinned on the shipped 4.11.659 exe): RajelAran's crash (call to 0x65676769 = ASCII 'igge', EBP-walk return btl4+0x2b91a) is the tgt->Dispatch vtable call in BTUpdateProjectiles' NON-MECH impact branch (mech4.cpp:1598; the site matches the disasm literally -- the 0x12/0x64 TakeDamageMessage ctor, the EntityID::Null ternary, the getenv gate after). The trap: the mech branch is guarded by BTIsRegisteredMech(tgt) -- but at round teardown a destroyed mech is DEREGISTERED, so a round still in flight (the pool is a static array that outlives the round; missiles are slow, #84) holding it as p.target now FAILS the mech check and falls into the !BTIsRegisteredMech branch, which treats the freed mech as a cultural icon and dispatches into freed memory. The liveness check itself routed the dangling pointer into the unguarded branch. Freed heap reused by a string -> vtable slot +0x10 read 'igge' -> call 0x65676769. (The EBP walker explains the stack shape: the faulting call pushed its return address on ESP, but the walk reads [EBP+4] = PerformAndWatch's frame.) Fix -- scrub at the source of truth: - BTProjectilesDropEntity(e): every pool entry drops a dying entity from p.target/p.shooter; called from ~Mech and ~CulturalIcon (the only free paths for targetable entities). Flight + impact code is already null-guarded on both fields. - BTProjectilesClearAll(): kills all rounds at RunMissions exit (cross- mission hygiene for the static pool). Verified: 6 short-mission cycles (45s missions with a spawned dummy + autofire, mission expiring mid-combat) -- 0 exceptions, 6/6 clean 'RunMissions returned'. The original crash was a heap-reuse race with no deterministic repro; the scrub eliminates the dangling-pointer class by construction. (Bench note: blind autofire never launches missiles -- the launcher needs a target lock -- so the exact in-flight race was not re-created; the non-regression + the pinned mechanism carry the verdict.) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J |
||
|
Joe DiPrimaandClaude Fable 5
|
cf8618bcb9 |
myomer system COMPLETE -- the 'WAVE 6 mover cutover' was a phantom built on
a misattribution; the authentic couplings are all live
THE CORRECTION [T1]: @004b9550/@004b95b8 ('ConnectToMover'/'Disconnect
FromMover') are MechWeapon::ConfigureMappables/ChooseButton -- their +0x31C
is the weapon TriggerState and **(mech+0x128) is the CONTROLS MAPPER's
button roster (the MECHWEAP.CPP assert string + the trigger-edge detector
@004b9608 sit adjacent in the image). The binary has NO dynamic myomer->
speed feed: AvailableOutput has exactly TWO callers in the whole image --
RegisterMaxOutput (assembly) and the SeekVoltageGraph sampler -- and
speedEffect@0x31C is a published GAUGE attribute.
The authentic myomer system (now fully live):
1. drive heat (#85, this morning);
2. VITAL death -- myomers are the only authored vital=1 subsystem
(BTL4.RES res+0x48 scan): destroying them kills the mech via the #80
vital-crit path (this is also the ram-death mechanism -- myomers are a
0.35-weight collision-rattle target);
3. the assembly-time gait cap: RegisterMaxOutput @004b8ef0 raises
mech+0x7A0 to AvailableOutput(top gear) = base x ~1.4284 -- now called
(idempotent max, per tick; the 0x358 layout lock leaves no latch room)
with real OwnerBaseSpeed/OwnerMaxSpeed bridges; the old per-frame
reverseSpeedMax2 heal that would have clobbered it is garbage-guarded;
4. the ENG-page power curve at correct absolute scale (real base speed);
5. electrical sourcing (low gears run on a browned-out generator).
REMOVED (inventions): the mechmppr `speedDemand *= speedEffect` demand
multiplier + BTMyomersDriveOf + the Myomers ConnectToMover/MoverAttach
family. Damaged-but-alive myomers heat faster and drop the power curve
but do NOT slow the mech (gitea #75's premise is falsified by the binary;
the real damage slowdown is the LEG-damage gimp gait). Seek gears change
heat/sourcing/graph -- never speed.
Verified live: a mech at myomer dmg=0.234 (speedEffect attr 0.766)
beelines at full commanded speed with dmgGain=1.234 heat; 0 faults.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J
|
||
|
Joe DiPrimaandClaude Fable 5
|
1570983fde |
#85: myomer drive heat LIVE -- the five Mech motion operands were return-0
shims, so the reconstructed integrator @004b8d18 accumulated zero forever
The integrator, its gates and the heat plumbing were all reconstructed and
running -- but OwnerVelocityMag/OwnerDriveMag/OwnerVelocityY/OwnerMotionGain/
OwnerMass were cross-family compile shims returning 0.0f, so the myomers
priced ratio^2 * damageGain * 0 every tick: ice cold at any seek (Oracle's
night-7 panel-confirmed report).
Operands re-grounded against the raw disasm + decomp and mapped to NAMED
members via a complete-Mech-TU bridge (BTMechMyomerMotionSample, mech4.cpp),
per the databinding rule:
+0x1C4 vec -> localVelocity.linearMotion
+0x82C vec -> localAcceleration.linearMotion (the authentic 15-ring
smoothed AccelerationLastFrame the port already maintains)
+0x20C -> moverMass (the collision divert's cell; the old "motion
gain" label was a misread)
*(+0x250) -> environment gravity (FUN_00421e2c: vy -= **(+0x250)/tick),
via GetEnvironment()->gravityConstant
So the heat physics reads: 0.005 * 1/2 m v^2 (kinetic work, per-tick/no-dt
in the binary -- kept verbatim, frame-rate note documented) + 0.2 * m g
|vy| dt (climb power) + 0.2 * m |v||a| dt (acceleration power), all scaled
by ratio^2 and (1 + zone damage). Restored the binary's fabs(v.y) (the
draft had signed vy -- descending would have COOLED).
Authored tuning extracted from BTL4.RES (18 mech records, one shared
tuning): VelocityEfficiency=0.995, AccelerationEfficiency=0.8, gears
0.3/0.5/0.7/0.9999 rec idx 2; myomer thermal profile 77/1000/2000,
conductance 1.9e5, thermalMass 2.5e5.
Live-verified (solo arena, BT_MYO_LOG + BT_HEAT_LOG): standstill generates
~0; hard circling drove Myomers T 77 -> 1225 (past the 1000 degradation
line) and the coolant loop conducted it back to ~520 equilibrium on
slowing; damageGain live at 1.048 from grind rattle.
Known-inert remainder (filed in open-questions): the climb term reads g=0
because Mover::localEnvironment is never populated in the port
(EnvironmentZone res type 23 unwired); the mover feed (speedEffect ->
locomotion) stays WAVE 6.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J
|
||
|
Joe DiPrimaandClaude Fable 5
|
37dd7f9663 |
collision-pricing audit: restore the binary's crash economy -- two fixes
A. KNOCKDOWN PACING: the 0.4s gBlockCooldown contact-hysteresis is gone. It suppressed the crash knockdown for as long as contact was held, letting the kinematic drive re-slam the obstacle at full commanded speed EVERY FRAME -- continuous full-price crash self-damage, a healthy 65-tonner dead in ~2-3s of grinding (the night-7 SAURON ram death, 110 priced frames at |v|~34.7). The binary needs no guard (crash threshold @0x4ab178 read from the exe = 0.0, dispatch per blocked frame is authentic): the bmp clip zeroes the drive, and a pressed mech re-triggers the knockdown at v>6.3 long BEFORE its grind can price anything -- the type-0 divert's 0.5-pt free floor needs v>~18.5 at 65t. Wall grinding is free taps + paced staggers; only genuine fast arrivals pay. B. RAW RAM DISPATCH: the *0.001 "ram economy normalization" (2026-07-12) predated the #83 type-0 divert and became a double-normalization -- the victim's divert scale (~4e-5) compounded with it, so NO physically possible ram could clear the 0.5-pt free floor: rams were a no-op against the victim while the rammer's own un-scaled crash path rattled for real. The binary dispatches raw [T1 @part_012:15324]; the divert IS the normalizer. Also un-starves cultural-icon crunches (trucks now crush on contact per their authored armor, not after ~19 bumps). Bench (scratchpad/night7/mp_rampricing.sh): - wall leg: 150s full-throttle wall push = 125 paced knockdown binds (~1.2s cadence, iv2 40-43), 4767 crash frames ALL under the free floor, 0 rattle, 0 deaths (pre-fix: dead in seconds). - ram leg: BT_GOTO=enemy + BT_GOTO_STOP=2 rams the peer; raw amount arrives byte-identical on the victim (66239.1 tx == rx), victim prices 2.55 pts internal rattle (first ram damage to a victim in the port's history), pressed follow-ups free, 0 deaths both nodes, 108 knockdowns -> exactly 108 type-5 records on the observer (no skating regression). KB swept: combat-damage.md (the 1.3e6 moverMass mis-attribution corrected to the measured 60-90k tonnage scale), locomotion.md, rendering.md. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J |
||
|
Joe DiPrimaandClaude Fable 5
|
054c3f2eee |
#94: replicant un-wreck now rides the DEATH-STATE EXIT edge -- the zone
falling-edge latch was blind to zone-less (collision) deaths The peer wreck lifecycle listened to two different signals: wreck ON came from the replicated death explosion (unconditional), wreck OFF from a damage-zone falling edge (wasWrecked latch, armed only when a zone crossed >= 1.0 on the observer). The #83 collision damage prices as internal rattle and never moves a zone, so a fresh mech killed by ramming could NEVER un-wreck on peers -- the live respawned mech drove around wearing the hulk (night-7 field report: SAURON's Loki, screenshotted by both observers at the exact minute the log analysis places the ram death). MechDeathHandler::Tick now tracks prevMode and fires the rebuild + warp when the replicated MovementMode leaves the death modes {2,9} -- only Mech::Reset ever does that, and the state rides every record header, so the trigger is cause-agnostic. The zone falling edge just refreshes the cache. Side effect fixed for free: the peer respawn warp vortex now plays for zone-less deaths too (same gate). Bench (scratchpad/night7/mp_zoneless.sh + the new BT_SELF_DAMAGE_TYPE=collision harness option, which dispatches type-0 through the real TakeDamage divert): 14/14 zone-less deaths un-wrecked (mode 9->1) with zero zone rises on the observer; explosive regression leg 12/12 with zone destruction active. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Zh7PTkFy4KwTzVighLR9J |
||
|
Joe DiPrimaandClaude Fable 5
|
e91d447405 |
#82 FINAL LAYER: the peer body-channel trn-lock -- a circling limper skated because the peer's turn-arm had no walk-demand yield and the trn speed exit read a dead replicant mapper cell
The master limped clean (218x gimp-cycle records, zero trn) -- a _ReturnAddress trap on SetAnimationState(4) proved ALL peer trn arming came from the port's own mech4 turn-step block, not the type-3 reader. Two dead ends, one root: - mech4 peer arm: gated on !wantsWalk (standSpeed < bodyTargetSpeed), exits trn when walking demand appears -- the leg twin's authentic precedence (part_012.c:12013, the Standing speed test outranks the turn test). - mech2 body case 4: bspd reads bodyTargetSpeed on ReplicantInstance; the local mapper's speedDemand is a dead cell on a peer (reads 0 forever -> no exit). Verified 2-node timeline: arena circle replicant 218x state-24 / 0x state-4 (was 116x trn churn), grass circle 235-clean, knockdowns bounded, 0 deaths. Diagnostics kept: BT_TRNTRAP ra-trap, BT_ANIMIND_CAP, [gimpfeed], [replgimp] extension. KB: locomotion.md trn-lock entry + the dead-mapper-cell lesson. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QRjrTQpJd6u9XyfnUbTB3v |
||
|
Joe DiPrimaandClaude Fable 5
|
14ff3519ac |
#83: the COLLISION DIVERT -- type-0 damage prices as internal rattle, never armor
The crash self-damage reconstruction (
|