radar: composite the OVERLAY plane -- SECTOR/SCALE were blank on every desktop path

Playtester DOS-binary comparison: our radar showed blank SECTOR:/SCALE: values. Root cause [T1]: the secondary CRT's low byte is authored as TWO ports (L4GAUGE.CFG:4395-4396) -- sec (0x3F, btspal) + overlay (0xC0, btopal, TransparentZero) carrying the btsec1ov graticule, the SCALE numeral (RadarRange) and the SECTOR numerals (sectorDisplay). Both numeral writers are reconstructed and RUN -- their pixels were in the shared buffer's 0xC0 bits all along -- but every desktop expand indexed the LUT with pixel & 0x3F, stripping them. The arcade DAC indexed the FULL low byte. Key mechanism: clut0 ALREADY holds the DAC's combined 256-entry table -- BuildSecondaryPalette (AllChannels, sec) fills all 256 indices and the overlay's TransparentZero pass overwrites groups 0x40/0x80/0xC0 -- so the composite is exactly a mask change, no blending code: - BTSecCompositeMask(): sec->GetBitMask() | overlay's (NULL-guarded, same display), used by BTDrawGaugeSurfaces + BTDrawCockpitPanels (surround), SVGA16::Update case 0 (pod parity), and the glass BlitSurface palette expand. - GlassWindowToken: the palette-expanded radar window's dirty token now includes the overlay port -- the SECTOR numerals tick on that plane, and a pixel-only 0x3F token would never repaint them. The overlay's ColorMapper re-stomp over combined indexes is authentic (the shipped machine ran the identical palette code) and is left alone. Verified live (glass panels): SECTOR: 497.503 ticking with movement, SCALE: 1000 at spawn (radarRange init 250*2^2 -- the DOS reference's 4000 is max zoom, not a delta), and the btsec1ov graticule rules now draw. KB: GAUGE_COMPOSITE row 28 + polish list updated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-08-13 07:18:43 -05:00
36 changed files with 243 additions and 3933 deletions
@@ -61,13 +61,3 @@ content/*.log
 content/matchlog_*.txt
 content/btl4.exe
 content/*.EGG
 # scratchpad: session working files -- NEVER bulk-add (the 2026-08-13 6a744ef
 # incident: a stray `git add scratchpad` pushed a session handoff carrying the
 # gitea credential).  Bench .sh scripts are added INDIVIDUALLY by full path.
 scratchpad/**/*.out
 scratchpad/**/*.exe
 scratchpad/**/*.dll
 scratchpad/**/SESSION_HANDOFF*.md
 scratchpad/**/AGENT_FINDINGS*.md
 scratchpad/**/message*.txt
@@ -199,23 +199,14 @@ Fixed-torso mechs (the BLH: `TorsoHorizontalEnabled=0` — no jointtorso in the
 boresight dead-ahead. Once locked, `Emitter::FireWeapon` converges with NO aim/arc test
 (part_013.c:7758). `MechWeapon::UpdateTargetState` (`FUN_004b9bdc` [T1]): `targetWithinRange =
 dist < (1 − hostZoneDamage) × weaponRange`. The 0x388 WRITER is in the same un-exported gap.
-⚠ ~~`weaponRange` is fire-control data only~~ **RETRACTED 2026-08-13 (the #171 hunt): that
+⚠ `weaponRange` is **fire-control data only** (#168 audit [T1]): projectile FLIGHT is never
-was MISSILE-only truth mis-generalized to ballistics.** The AC's own FireWeapon @004bc104
+range-capped — a plain shell (AFC) lives exactly 5.0 s (@4bddec) and a Missile lives
-(recovered by the 2026-08-06 re-export; #168 worked from a stale "not recovered" banner and
+BurnTime+10 s with a y<−1 kill-plane (@4bef78); FireWeapon @4bcc60 has no range gate, so an
-generalized from the missile launcher) is TRIGGER-TIME HITSCAN: batched rounds, dist vs LIVE
+AFC50 (authored range 750) landing hits at 2000+ u is AUTHENTIC arcade behavior. Constants +
-effectiveRange at fire time (the same gate energy uses), ONE single-panel damage message to
+the whole flight/expiry chain: [[decomp-reference]] §5 "Projectile/Missile flight & expiry".
-the locked target, nothing spawned (the shell class 0xBD1 is dead code; the visible round is
+The same audit killed the pool's stale-point auto-hit (unguided contact was tested against
-the 0xBCD renderer tracer).  So: an AFC50 landing hits beyond its authored range is a PORT
+the fire-time pick, so a locked AC shell could not miss a moving target — fixed 2026-08-11:
-DEVIATION, not arcade behavior; "a locked AC could not miss a mover" IS the arcade behavior
+the contact test now tracks the target's live position on every round kind).
 (#168's "stale-point auto-hit" kill removed an authentic design while fixing its stale-point
 detail).  Missiles: genuinely flying + seeker-led (the machine leads for the player), 4.0u
 proximity fuze, swept contact, salvo randomization — the #168 flight direction stands THERE.
 Gauss CORRECTED (2026-08-13 audit): the FIELDED Gauss record is CLASS 0xBCD (ProjectileWeapon)
 — full AC hitscan WITH damage; port agrees. GaussRifle 0xBCE = dead code both sides (zero
 shipped records). The earlier no-damage claim was class-level truth mis-applied to the field.
 #112 (AFC multi-zone shotgun) is the port's per-round scatter vs the binary's one batched
 message.  Full chain + evidence: [[decomp-reference]] §5 (corrected); port fix pending the
 #171 group decision (Oracle's HOLD).
 ⚠ **`hostZoneDamage` is `Subsystem::damageZone->damageLevel` (weapon `@0xE0 → +0x158`), NOT
 heatLoad** — the port originally computed `effectiveRange = (1 − heatLoad) × weaponRange`
 (mechweap.cpp), the SAME `@0xE0`-DamageZone-vs-heat misattribution corrected in
@@ -253,56 +253,20 @@ From the weapon `.SUB` records + the charge-curve `.data` constants (PE-parsed a
  @004b0b18) + @004b0abc (auto/manual, modeAlarm@0x2CC); attach/detach @004b0dd8/@004b0e30
  (generator client capacity @0x1e4/count @0x1e8).
-### Projectile/Missile flight & expiry (#168 audit 2026-08-11; ⚠ CORRECTED by the #171 hunt 2026-08-13) [T1]
+### Projectile/Missile flight & expiry (#168 audit, 2026-08-11) [T1]
- ⚠ **#168's "no range cap / WeaponRange is fire-control only" was MISSILE-ONLY truth,
+- **No range cap exists anywhere in the flight chain.** WeaponRange feeds fire-control only
-  mis-generalized.** The AC's own FireWeapon @004bc104 (recovered by the 2026-08-06 re-export;
+  (effectiveRange @0x328 = (1−hostZoneDamage)×WeaponRange — targeting/reticle "in range" call).
-  the old projweap.cpp "body NOT recovered" banner was stale) IS range-gated AND hitscan:
+  FireWeapon @4bcc60 reads no +0x328: the trigger fires at any target distance.
-  batch counter +0x408, every TracerInterval-th (+0x438) round → dist = |aimPoint(+0x37c) −
+- **Plain Projectile (AFC shell) live integrator @4bddec** (master); expiry: age since
  muzzle| vs LIVE effectiveRange +0x328 (dynamic: WeaponRange × (K − owningZone structure),
  recomputed @004b9bdc; K @_DAT_004b9c98 presumed 1.0 [T4-unread]) → ONE single-panel Damage
  (batchedCount × DamageAmount) to the LOCKED target *(mech+0x388) via @004b9728 AT FIRE TIME.
  Out of range = batch silently discarded (tracer still renders).  The visible AC round is the
  0xBCD renderer TRACER (tracerModel+0x41C, TOF @004bc06c) — cosmetic.  MissileLauncher
  @004bcc60 (no fire-time damage, no range read) is the genuinely-flying family.
  **GAUSS CORRECTED (weapons audit 2026-08-13): the FIELDED Gauss is a 0xBCD ProjectileWeapon
  record** (BTL4.RES @0xf16e6, bytes cd0b verified raw — 20 pts Ballistic, 900 range, 8s cycle,
  16 rounds, yellow pip) — it runs the FULL AC hitscan chain WITH damage; port and binary agree.
  The GaussRifle class 0xBCE (@004bdca4 charge-dump-only FireWeapon) + shell 0xBD1 are real code
  with ZERO shipped content records — dead on both sides. (The earlier "port Gauss damage is an
  invention" claim was wrong at the fielded level and is retracted.)
  **Every FireWeapon (all classes) requires a LOCKED target**: both sims gate the trigger on
  *(mech+0x388) != 0 — no lock, no fire [T1].
 - **Plain Projectile (shell) class 0xBD1 is DEAD CODE in the shipped binary**: 0xbd1 appears
  exactly once in the whole export (the DefaultData lookup @004c1590); Make @004be384 has zero
  code callers; the fully-recovered AC fire path spawns nothing.  Its flight body @4bddec
  (below) is a complete never-instantiated Model-B implementation — its 5.0s clock is real
  code that nothing shipped ever runs.  [T1; 6.5% export-gap caveat]
 - **Plain Projectile (dead class) integrator @4bddec** (master); expiry: age since
  spawn(+0x19c) > `_DAT_004be054` = **5.0 s** → FUN_0042061c kill.  Dead-reckon divergence²
  vs predicted pos(+0x260, predictor FUN_00422060 via +0x254) > `_DAT_004be050` = 0.1 →
-  updateModel(+0x18)|=1 (ForceUpdate, replication-dirty — NOT a kill).  Contact: swept test
+  updateModel(+0x18)|=1 (ForceUpdate, replication-dirty — NOT a kill).
  @0042291c per tick; on hit → explosion msg 0x5C at live pos + direct TakeDamage (zone −1)
  to the struck entity via @004be078; at EXPIRY silent removal, no damage.
 - **Missile live integrator @4bef78** (master); expiry: age > burnTime(+0x340, model rec
  +0x44) + `_DAT_004bf5ac` = **10.0 s** coast margin, OR pos.y < `_DAT_004bf5b0` = **−1.0**
  kill-plane → FUN_0042061c.  Keepalive: no update for `_DAT_004bf594`=2.0 s → dirty bit.
  **Proximity fuse**: seeker rangeToTarget(+0x10C) < `_DAT_004bf5a4` = **4.0** → detonate on
  targetEntity without a geometry hit.  Seeker drops a destroyed target (movementMode 2|9,
  FUN_0049fb54).
 - **Missile FLIGHT MODEL (velocity/dt audit 2026-08-13) [T1]:** per frame:
  `MissileThrusterSimulation` **@004be474** (the standalone Performance body — recovered;
  misthrst.cpp's "folded" note was stale) slerps the missile toward the seeker aim
  ≤ MaxThrusterRotationRate·dt and, while burning, adds `(0,0,−ThrusterAccel)` to the
  frame-zeroed accumulator; @4bef78 adds quadratic drag `−COD·ρ·sign(v)·v²` per body axis
  (burn AND coast; the dead plain-Projectile @4bddec uses power **1.0** via
  ApplyAirResistanceAndGravity(1.0)), zeroes LATERAL local vel/accel while burning (velocity
  slaved to the nose; slaving STOPS at burnout), applies gravity (env, default **6.5**) ONLY
  when coasting, then integrates ×dt (@00421bac).  Net: `dv/dt = T − COD·v²`.  **Authored
  CODs = 0.001 ALL axes, all four models** (BTL4.RES type-15 rid 315 srm / 319 lrm /
  323 strk / 327 nrk, raw floats); ctor @4bf5b4 (part_013.c:18416) precomputes **terminal
  speed `sqrt(T/negCOD.z)` → missile+0x348** (the seeker lead-time base): LRM/Streak/NRK
  **547.7 u/s**, SRM **774.6**.  Port pool carries the same ODE since the 2026-08-13 drag
  fix (mech4.cpp; drag was previously omitted as "negligible" — it is the speed governor).
 - **Performance split** (Projectile ctor @4be1bc / Missile ctor @4bf5b4): instance
  (flags&0xC)==ReplicantInstance(4) → FUN_004221c0 dead-reckoning smoother (no physics);
  master → 4bddec/4bef78.  (CLASSMAP's old authoritative/ghost labels were swapped —
@@ -436,19 +400,18 @@ From the weapon `.SUB` records + the charge-curve `.data` constants (PE-parsed a
  the Execute machine, but this handler overwrites it).  The earlier "jams are
  mission-permanent / no unjam" claim was WRONG (read only the hold path) — corrected +
  swept 2026-07-23 (projweap.cpp case-5/jam-log comments, players/README.txt).
-  **#21 RETRACTED 2026-08-13 (era testimony -> adversarial re-audit; witnesses 6-0):** the
+  **LATENT ARCADE BUG found + diverged (issue #21, 2026-07-21) [T1 byte-verified + T2 repro]:**
-"LATENT ARCADE BUG / permanently bricked / the arcade shipped this" claim that stood here was
+  the Emitter charge integrates toward the GENERATOR voltage (@004ba838 `fld [src+0x1dc]`), full
-a CONSTANT-WIDTH MISREAD.  `_DAT_004ba830` is a **DOUBLE 1.0** (bytes 00..00 f0 3f @0x4ba830 --
+  is detected only inside the +-1% snap window around `seekVoltage[idx]` (ComputeOutputVoltage
-the old "0.0 byte-verified" read the low DWORD of a QWORD: 1.0's mantissa zeros; Ghidra's own
+  @004ba738: `|rl-1|<=0.01 -> 1.0`), and OVERCHARGE reads ZERO (`rl>1.01 -> 0`,
-"globals overlap smaller symbols" warning applies).  ComputeOutputVoltage @004ba738 truth [T1
+  `_DAT_004ba830 = 0.0` byte-verified).  Toggle the seek gear while a charge is in flight and
-raw disasm]: ratio = charge/seekV[idx]; snap |rl-1|<=0.01 -> 1.0f; else CLAMP to [0,1] --
+  the level lands ABOVE the new gear's window -> rechargeLevel pinned 0, the `==1.0` Loaded test
-overcharge stores EXACTLY 0x3f800000, the `==1.0` Loaded test passes next tick, the weapon
+  (`_DAT_004bac04`) never fires, charging continues to the generator ceiling where EVERY gear
-FIRES the stored charge (damage divisor is the RECOMMENDED gear, so a wrap-fired charge deals
+  reads overcharged -> the weapon is PERMANENTLY bricked (arc dark, ready-dot dark, no fire; gen
-normal top-gear damage).  ToggleSeekVoltage @004ba478 genuinely skips ResetFiringState on wrap
+  re-select cannot help).  The arcade shipped this (locked 60 fps + rare seek use hid it); the
-[T1] -- but nothing downstream ever bricks.  The machine never had a dark laser; the port's
+  port's clickable seek button triggers it in seconds.  **Deliberate divergence** (emitter.cpp
-"rescue" was a patch over OUR misread (removed with the clamp fix, same date).  Constants
+  Loading tick): overcharge (`currentLevel > seekVoltage[idx]`) counts as fully charged ->
-extracted: 0x4ba818=0.0f 0x4ba81c=1e-4f 0x4ba820=1.0f 0x4ba824=0.01f 0x4ba828=dbl 0.0
+  Loaded ("full == the gear's seek voltage" is the arcade's own discharge algebra).  Repro +
 0x4ba830=dbl 1.0 0x4bac04=1.0f.  Gotcha family 30/31 (width/scope misreads).  Repro +
  verify: BT_SEEKTEST seek-abuse -- pre-fix deadlocks at pct=0/alarm=3; post-fix 38 rescues,
  ends Loaded/pct=1.  `[seek]` log prints the per-gear table + rescue events.
 - **EJECT/PANIC cluster DECODED + WIRED 2026-08-02 [T1 raw disasm]:** `@0049f854`
@@ -618,14 +581,9 @@ extracted: 0x4ba818=0.0f 0x4ba81c=1e-4f 0x4ba820=1.0f 0x4ba824=0.01f 0x4ba828=db
  (`FUN_0041bd98`) — **VARIABLE-STEP, slice in real SECONDS**, line-for-line the
  WinTesla SIMULATE.cpp body. Corollaries: every Performance's `time_slice` is
  seconds in BOTH binary and port (no hidden tick/second unit gap); a dt-LESS
-  per-frame term into PERSISTENT state (myomer kinetic @4b8d18's middle term)
+  per-frame term (myomer kinetic @4b8d18's middle term; the Mover gravity
-  fires at the FRAME cadence = the tick rate, 28 Hz nominal, sagging under load.
+  `-= **(mover+0x250)` @0x421e77) fires at the FRAME cadence = the tick rate,
-  ⚠ The Mover gravity `-= **(mover+0x250)` @0x421e77 is NOT in that class
+  28 Hz nominal, sagging under load. Mover velocity is u/s (the replicant
  (corrected 2026-08-13, velocity/dt audit): `Mover::PerformAndWatch` =
  **`FUN_00422360`** zeroes `localAcceleration` every frame (== T0
  MOVER.cpp:672), so gravity/drag/thrust adds define a fresh per-frame
  ACCELERATION integrated ×dt by `ApplyWorldAccelerations` @00421bac —
  frame-rate-independent, no 28 Hz rescale. Mover velocity is u/s (the replicant
  dead-reckoner @0x421f7c does `pos += vel × ticksΔ/DAT_0052140c`). The myomer
  heat term's ONE caller is the **unexported MyomersSimulation body
  @0x4b8b9a-0x4b8d0d** (E8-scan + raw disasm; the export gap hid it): calls
@@ -774,12 +732,7 @@ default-ON (`'0'` disables).
 | `BT_WARP_*` | translocation-warp visual tuning — all default to the verified values (see [[translocation-warp]]): `EYE_UP=8.25` (on-axis), `SPIN=4`, `TWIST=0`, `CONTRAST=1.0`, `LO_*/HI_*` (lavender ramp), `BLUR=0`, `TESS=3`, `MIP=1`, `CULL=cw`, `ANISO=0` |
 | `BT_WARP_SELFTEST` / `BT_WARP_SELFSHOT=<prefix>` | DIAG (off by default): force a held warp in a solo game / dump backbuffer frames to disk (visual-verification harness) |
 | `BT_GYRO_LOG` | gyro bring-up log (gyro.cpp) |
-| `BT_GYRO_TRACE` | per-frame hit-bounce integrator trace (`[gtrace]`, carries wall-ms `t=`) |
+| `BT_GYRO_TRACE` | per-frame hit-bounce integrator trace |
 | `BT_GYRO_SPRING_HZ=<hz>` | gyro spring-damper integrator clock override (default 28 = the pod tick, census item 2; `=0` restores render-cadence stepping) |
 | `BT_GYRO_KICK=<frame>[,amt]` | one-shot deterministic gyro impulse via the authentic GyroApplyDamage fan-out (impulse-response bench; default frame 900, amt 25; `[gyro-kick]` receipt) |
 | `BT_AUD_SMOOTH_HZ=<hz>` | footstep AudioControlSmoother intake clock override (default 28, census item 3; `=0` restores per-poll intake) |
 | `BT_AUD_DOPPLER_HZ=<hz>` | doppler velocity-operand sample-and-hold clock override (default 28, census item 4; `=0` restores per-call) |
 | `BT_AUD_DOPPLER_LOG` | `[doppler]` receipt (cents/v/d/h/hm2, capped 6000; h=1 fresh / 0 held / -1 legacy) |
 | `BT_CRIT_PROBE=<zone>` | hammer one own-mech zone every 4s (crit-propagation diag, task #2; mech4.cpp) |
 | `BT_REPL_LOG` | replicant/MP replication log (mech4.cpp) |
 | `BT_GOTO="enemy"\|"x z"` / `BT_GOTO_LOG` | self-driving beeline (to the enemy or a map coordinate) + its log |
@@ -194,8 +194,7 @@ register.  ⚠ The audit also flags the damage-economy item as SELF-CONTRADICTOR
 - **Environment gravity unwired — the myomer CLIMB-heat term is inert (found 2026-07-31, #85).**
  `Mover::localEnvironment` (MOVER.h:271) is declared and never populated anywhere in the port;
  `GetEnvironment()->gravityConstant` is the named analog of the original Mover's gravity POINTER at
-  `+0x250` = `localEnvironment` (MOVER.h; `FUN_00421e2c` = ApplyAirResistanceAndGravity does
+  `+0x250` (`FUN_00421e2c` does `vy -= **(+0x250)` per tick), which the myomer drive-heat
  `worldAccel.y -= **(+0x250)` each frame onto the frame-zeroed accumulator), which the myomer drive-heat
  integrator uses for its `m·g·|vy|·dt` climb-work term.  The bridge (`BTMechMyomerMotionSample`)
  null-guards it, so climbing currently generates no myomer heat (flat maps: moot; cavern slopes:
  a missing cost).  EnvironmentZone resources (RES type 23) exist in BTL4.RES — wiring them (or at
@@ -716,19 +715,12 @@ register.  ⚠ The audit also flags the damage-economy item as SELF-CONTRADICTOR
  the Mover holds to the live gravity cell). Harmless on flat ground (|vy|≈0 anyway) but the
  climb-work heat never accrues on slopes. Find where our Environment gravity actually lives
  and re-point the bridge. [T2 measured]
- **✅ Engine Mover gravity needs NO 28Hz rescale — CLOSED 2026-08-13 (velocity/dt audit) [T1/T0].**
+- **Engine Mover gravity is dt-LESS per-frame in the binary** (`@0x421e77: vy -= **(+0x250)`,
-  The worry was that the per-frame `worldAccel.y -= **(+0x250)` (@0x421e77, no dt) would scale
+  raw, once per Perform — same class as the myomer kinetic term). If our linked WinTesla
-  with our frame rate.  It cannot: **`Mover::PerformAndWatch` = `FUN_00422360` ZEROES
+  MOVER.cpp kept that form, gravity acceleration in the port scales with OUR ~59Hz frame rate
-  `localAcceleration` at the top of every frame** (`FUN_0040a7f4(+0x1dc, &DAT_004e0fd4)` ==
+  (≈2.1× the pod's 28) — jump arcs, falls, missile droop all stiffer than the pod. CHECK the
-  T0 MOVER.cpp:672 `localAcceleration = Motion::Identity`), so gravity/drag/thrust adds are
+  WinTesla source's gravity line; if per-frame, it needs the same 28Hz reference-rate
-  per-frame *acceleration definitions* on a fresh accumulator — a constant `a` integrated ×dt
+  normalization the myomer kinetic term got. [T1 binary side / T4 port impact until read]
  by `ApplyWorldAccelerations` @00421bac (`v += a·dt`), correct at ANY frame rate.  The port's
  linked MOVER.cpp has the identical structure.  (Distinct from the myomer per-tick accumulates,
  which land in PERSISTENT state and did need the ×(dt·28) fix.)  The separate gap — the port's
  `GetEnvironment()->gravityConstant` sampling 0 (unwired Environment, entry above) — is still
  open; note the target constant is **6.5** (LATTICE.cpp:413 default), not 9.8 (9.8 appears only
  in the AC tracer fire-control math @0x4bc67c).  Full audit:
  `docs/WEAPONS_DRIFT_AUDIT.md` §VELOCITY/DT INTEGRATION.
 - **Gait-noise phantom acceleration feeds the myomer accel term** (`[myoheat] a=4..33` at
  steady cruise — our per-frame velocity re-derivation jitters; the binary read the same
  `AccelerationLastFrame` cell off ITS OWN gait, noise level unknown). Bounded contributor
@@ -1233,18 +1225,7 @@ consistent with internals-only, see #103). Logs: `scratchpad/night16/` + Oracle
  (x2 spawns, AFC50/Zanin Neko); once ABSENT for an entire fatal leak (5-6 voice repetitions,
  no flash on any MFD, died ~9:38PM CT -- the match's only death, receipt findable in night16).
  Distinct from the closed #135: this is the SET-edge / first-flash arming path.
- **#173 -- chicken-walker seek-4 myomer overheat: ⚑ SETTLED AUTHENTIC (2026-08-13, three
+- **#173 -- chicken-walker seek-4 myomer overheat** (see [[subsystems]] night-16 note).
  lanes, see [[subsystems]] #173 blocks).** Bench matrix reproduced Elengil exactly (vulture
  seek-4: degT in 13.9 s, governor-pinned eq ~1801, loop-5 bar full-scale ~15-20 s, sheds
  −10/s only at standstill); the equilibrium is a chassis-blind governor thermostat — the
  port's avatar (same 143-kph speed class) cooks identically, and the manual prints the big
  walkers even FASTER (MadCat 175/220) than the port runs them. Counter-play measured: one
  loop-5 valve press sustains drive 0.68 vs 0.54. TWO residuals stay open: (1) [T3] the
  accel-term |a| ring is frame-cadence-sensitive (|a|≈14-20 at 59 fps on a straight flat
  cruise = gait surge; pod 28 Hz would read ~2-3x smaller → pod eq likely ~1200-1500, softer
  governor — if calibration is wanted, fix the |a| feed cadence, never the byte-exact
  constants); (2) locomotion-lane: port per-chassis top speeds undershoot the manual prints
  (thor 87 vs 143 N print) — correcting them raises heat further.
 - **#165 ROOT-CAUSED + FIXED + BENCHED (2026-08-13, the 4-agent hunt):** gotcha SS30 -- the
  binary's ONE destroyed cell was split into four port stores and the crit-lottery kill path
  (MechWeapon::TakeDamage) wrote none the gate reads. Fixed (chain to MechSubsystem::TakeDamage,
@@ -1264,49 +1245,3 @@ consistent with internals-only, see #103). Logs: `scratchpad/night16/` + Oracle
  wanted for the pair-leak confirmation.
 - Nanook lost the coolant MFD glass window, recovered only by config edit -> #76 comment
  (in-game layout-reset recovery path).
 ## Cadence census -- the shared 28Hz sim-clock migration checklist (2026-08-13) [T1/T2]
 Census of sampled-history-at-render-cadence sites (gotcha 32; full table in the census
 workflow journal wf_d875938f + classifier verdicts).  Night-17 queue receipts:
 `scratchpad/night17/CADENCE_STATE.md` (benches + curves for every row below); commits
 f811c64/567ffac/b355844/f9546d2.
 DONE (each a per-instance 28Hz accumulator/hold, heat.cpp #119 pattern, env-override to
 bracket, receipts armed):
 - velocity ring default-28Hz (mech4.cpp, HIGH, the #173 fix -- myomer heat inherits);
  heatLoad already pod-clocked (#119, the reference pattern).
 - instability/sway operand: VERIFIED inherited from the ring fix (no code change) -- same-
  build A/B via BT_MYO_RING_HZ=0: cruise unstablePct median 0.490 -> 0.109 (4.5x), |a| p90
  48 -> 14.5.  [T2]
 - gyro eye/body spring-damper integrators (gyro.cpp, was MED): stepped on a 28Hz
  accumulator, integrator bodies byte-untouched; impulse bench (BT_GYRO_KICK one-shot)
  BEFORE/AFTER: overshoot 23%->7%, period 1.43->2.4s, settle equal (per-SECOND damping is
  cadence-invariant); shape verdict = timescale-only.  BT_GYRO_SPRING_HZ.  [T2]
 - footstep AudioControlSmoother intake (AUDCMP.cpp, was LOW): only the Add clocked; the
  poll path measured at ~141-147/s (worse than assumed -- the wall window was 0.2s, not
  0.5s); ramp90 0.55->1.23s (the authored ~1.07s window restored); N untouched.
  BT_AUD_SMOOTH_HZ.  [T2]
 - doppler operand (AUDLOC.cpp, was LOW): relative WORLD velocity sample-and-held at 28Hz,
  consumer-side only, head-frame geometry live; operand total-variation -20%, stride std
  preserved.  BT_AUD_DOPPLER_HZ + [doppler] receipt (BT_AUD_DOPPLER_LOG).  [T2 mechanism /
  T3 PERCEPTUAL -- solo-rig cents jitter moved only -11% (near-field geometry dominates at
  d~1); the nearby-walking-mech warble claim still needs an ear A/B or a 2-node rig.]
 - ring dt-accumulator registry productionized: 16-slot static -> heat.cpp map pattern
  (17th-mech silent revert now impossible); smoke identical within run noise.  [T2]
 REMAINING for the BTSimClock migration (one shared 28Hz tick source; tick-native sites
 subscribe; dt-native physics -- Mover/missile ODE -- stays on real dt):
 - LOW published localVelocity 0.25s exponential smoothing (mech4.cpp:5578-5613): re-base to
  the binary's own 15-sample@28Hz filter once the shared clock exists.
 - NEW (night-17, receipt-exposed, pre-existing, UNTOUCHED): doppler sources linked to CHILD
  entities of a moving mech read relative v = full ground speed (child worldLinearVelocity
  = 0) -> dopplerCents ~ -200 on self-sounds at d~1 (the [doppler] receipt shows it
  directly).  Not a cadence bug -- a producer-frame question; determine whether the
  binary's located-audio path behaved the same (the pod cab may simply have masked it)
  before touching anything.  [T3]
 - DEFERRED dead-reckon divergence (@_DAT_004be050) -- only matters if the Missile entity
  revives.  ANNOTATE-ONLY: Simulate telemetry FilteredScalars (dormant).
 - Consolidation: five per-site clocks now duplicate the same accumulator idiom (heat, ring,
  gyro, smoother intake, doppler hold) -- fold onto the one BTSimClock when it lands, one
  site per commit, using the existing env-overrides as the bracketing levers.
@@ -635,14 +635,6 @@ CONCLUSION: the coolant-loop reconstruction is faithful; every difference is 4.0
 (loadout rework on Loki, small-laser redistribution elsewhere), NOT a bug. `BT_SPEC_LOG` (mech4)
 re-dumps on demand.
 ### Weapon-range cross-check RESULT (2026-08-13) [T1] — manual prints only 4 range figures
 The manual has NO weapons table and prints NO per-weapon damage number anywhere; the per-mech
 stat sheets carry no range/damage column. The only printed ranges are MFD screenshot mockups:
 PPC 900M (p20) and AFC100 400M (p21) MATCH the shipped 4.10 content; SRM4/SRM6 450M (p22/p23)
 vs the shipped 800 is the same 4.0→4.10 drift class as the Loki rework (and missile WeaponRange
 is display-only — no physics change either way). Full table + the rangefinder-scale verdict:
 `docs/WEAPONS_DRIFT_AUDIT.md` §RANGE AUTHENTICITY.
 ## Key Relationships
 - Renders: [[gauges-hud]] (the MFD surfaces), [[rendering]] (the main 3D view).
 - Input: [[locomotion]] (the mapper).
@@ -1064,26 +1064,3 @@ passes every reader-side bench. **Bench rule (the #165 lesson, second applicatio
 week):** prove the bench detects the bug BEFORE trusting its pass -- the negative control
 here (fix stashed) scored 147 cascades where the fixed build scores 1.
 Bench: BT_ZONE_HAMMER=<zone>[,<amt>] (mech4.cpp) + scratchpad/night16/cascade_bench.sh.
 ## 32. SAMPLED-HISTORY OPERANDS AT RENDER CADENCE: rate-scaling amounts cannot un-noise a signal (#173 flip, 2026-08-13)
 The port's #137 pattern (term x dt x 28) correctly rate-normalizes per-tick heat AMOUNTS -- but
 OPERANDS COMPUTED FROM SAMPLED HISTORY (rings, last-frame deltas, moving averages, per-tick
 integrators carrying state) still sampled at RENDER cadence where the pod's board sampled at
 28 Hz.  The killer case: the AccelerationLastFrame velocity ring pushed per render frame read
 |a| 17-26 of pure gait-aliasing noise on a flat straight cruise -- 64-66% of myomer drive-heat
 generation was PHANTOM; at 28 Hz sampling the noise drops to 24-32% and generation HALVES at
 equal speed.  Consequence: every fast-heavy chassis was over-punished ~2x; the authored loop-5
 valve boost genuinely sustains FULL supercharge seek-4 at 28 Hz (bench m173v2 leg G: 54-57 u/s
 held, Tm stable ~1500) -- the machine's lever worked and the manual's Super Charged print was
 honest.  Witnesses (Draco/Oracle/Elengil + the user's own eyes) beat the validated-solver
 extrapolation AND the first bench: 7-0 this week.
 **Rules:** (1) any binary value derived from per-tick sampled history must be sampled on the
 BINARY'S clock -- accumulate render dt, take one sample per 1/28 s measuring over the full
 elapsed window (the heat.cpp #119 heatLoad clock is the reference implementation; the ring
 default-28Hz in mech4.cpp is the second).  (2) Rate-scaling the consuming term is NOT
 sufficient -- it scales amounts, not signal quality.  (3) A validated model EXTRAPOLATES; only
 the bench INTERPOLATES -- never print "sustains indefinitely" from a solver row the bench
 never flew (the #173 verdict flipped twice on exactly this).  Census of the whole class (11
 rows, 2 HIGH fixed, migration checklist for the shared 28 Hz sim-clock): open-questions.md
 SSCadence-census + the wf_d875938f journal.
@@ -122,10 +122,7 @@ Making a base byte-exact GROWS every subclass — they must be re-based TOGETHER
  (mech4.cpp): `+0x1C4` vec = `localVelocity.linearMotion`; `+0x82C` vec =
  `localAcceleration.linearMotion` (the authentic 15-ring smoothed AccelerationLastFrame);
  `+0x20C` = `moverMass` (the collision divert's cell — the old "motion gain" label was wrong);
-  `*(+0x250)` = the environment **gravity pointer** (= `Mover::localEnvironment`, MOVER.h;
+  `*(+0x250)` = the environment **gravity pointer** (`FUN_00421e2c` does `vy -= **(+0x250)`/tick).
  `FUN_00421e2c` = ApplyAirResistanceAndGravity does `worldAccel.y -= **(+0x250)` each frame —
  onto an accumulator `Mover::PerformAndWatch` @00422360 zeroes every frame, so it is a constant
  acceleration integrated ×dt, NOT an accumulating per-tick term; velocity/dt audit 2026-08-13).
  Physics: `heat += ratio²·(1+dmg)·[0.005·½mv² + 0.2·m·g·|vy|·dt + 0.2·m|v||a|·dt]` — kinetic work
  + climb power + acceleration power.  The binary `fabs`es v.y (the port's old draft didn't).
  **Authored tuning extracted from BTL4.RES** (all 18 mech variants share one record):
@@ -342,178 +339,17 @@ speed in the spike frames.  The calibration facts above (constants byte-exact, d
 kinetic term) all STAND; the in-life governor (running hot at sustained top speed derates the
 myomers) is authentic and remains.
-## Spawn-discharge loop heat + per-chassis generator bias — ✅ CONFIRMED (2026-08-13, #173 bench lane) [T2 live receipts + T1 authored data]
+## Spawn-discharge loop heat + per-chassis generator bias (night-16 field, 2026-08-12) [T3 field-observed, mechanism plausible-authentic]
 Oracle's explanation for the long-standing "missiles cook off on one side first" reports
 (Lynx): at spawn ALL weapons start discharged, and the initial CHARGE of the energy weapons
 dumps heat into their loops; generator assignments are biased per chassis (Avatar: BOTH large
 lasers on GeneratorA), so that side's loops start hot and a missile launcher sharing the side
-cooks off first even if the player fires only missiles.  **All three links verified:**
+cooks off first even if the player fires only missiles. Not yet verified against the authored
-(1) *Weapons spawn discharged* — Emitter ctor @004bb120 AND ResetToInitialState @004ba4d0 both
+data -- verify via BTL4.RES generator/loop assignments before treating as established. [T3]
 seed `currentLevel=0` + weaponAlarm Loading [T1]; (2) *charge heat lands in the powering
 generator* — `TrackSeekVoltage` @004ba838: `src->pendingHeat += seekRate² · dtScale · dt`,
 and the generator conducts to its authored condenser (GenA→C1, GenB→C2, GenC→C3, GenD→C5,
 uniform all 18 variants) [T1]; (3) *the Avatar bias is authored* — BTL4.RES avatar chain:
 ERLLaser_1 AND ERLLaser_2 both `VoltageSource=GeneratorA` (plus ERMLaser_1) [T1].
 **Live receipt** (solo avatar, BT_HEAT_LOG, ZERO shots fired, first 5 s census,
 `m173_ava_s4.log`): GeneratorA T=559 load pegged 1.0 vs GeneratorC T=172 / GeneratorD T=79;
 Condenser1 (GenA's loop) 390→459 while C4/C6 sit at 75; and **LRM10_1 (authored to loop 1)
 reads 144 vs LRM10_2 (loop 3) at 94** — the asymmetric missile-side heating with no trigger
 pulled.  Loops also BACK-HEAT their weapons (ConductHeat is two-way): vulture's ERMLasers on
 hot loop 2 climbed 76→182 in 10 s, zero firing (`m173_vul_s1.log`).  [T2]
 Related open defect/question: **#173** -- Vulture (and per Oracle ALL large chicken walkers)
 cannot run seek-4 myomers without immediate overheat, loop 5 pegged, heat only sheds near
-standstill (Elengil; Oracle repro). ⚑ SETTLED 2026-08-13 by the three lanes below (authored
+standstill (Elengil; Oracle repro). Investigate myomer heat rate + loop-5 capacity vs authored
-data + cooling-side audit + live bench matrix): AUTHENTIC authored balance, reproduced and
+data for those chassis; authored-balance vs port-artifact undecided. The myomer drive-heat
-quantified; one bounded [T3] severity residual (accel-term ring cadence) and the locomotion
+FORMULA is verified faithful (#137, above) -- if the rate is right, the question moves to the
-top-speed cross-check remain flagged. See the BENCH lane block for the numbers.
+authored loop capacity / condenser assignments.
 ## #173 COOLING-SIDE AUDIT (2026-08-13) -- the binary heat chain leans AUTHENTIC [T1 data/formulas, T2 equilibrium]
 Full authored heat-family dump (content/BTL4.RES type-17 walk, scratchpad heatscan3.py) + the
 dissipation chain from the decomp. The chain: Myomers -> Condenser5 -> Bank -> ambient-300.
 - **Cooling is authored CHASSIS-FLAT** [T1]: every one of the 18 variants streams IDENTICAL
  condensers (k=315000, mass=420000, refrig=3, degT=failT=2000), myomers (k=190000, mass=250000,
  velEff .995/accEff .8, gears 0.3/0.5/0.7/0.9999 rec 2 at res+0x198..0x1AC), generators (rated
  10000, GeneratorD AND Myomers both sinkIdx=8 = Condenser5 on ALL chassis), reservoir, and bank
  (k=231000, mass=1.39e6) -- the ONLY per-chassis field is bank HeatSinkCount (owens 5, blkhawk 6,
  mad2 10, mad1/thr1/sunder 13, madcat 14, avatar/loki 15, vulture/vul1/lok1/lok2 18, thor 28,
  snd1 30), and at balanced valves it moves max shed by <2% (bottleneck is condenser->bank).
  moverMass (type-15 GameModel +0x0): owens 35k, blkhawk/VULTURE 60k, loki 65k, avatar/thor 70k,
  madcat 75k, sunder 90k. **No per-chassis gear cap exists** (shared 4-gear table + sentinel).
 - **NO speed-dependent cooling anywhere in the binary** [T1]: full operand walks of
  ComputeHeatFlow @004ad9ec (dt, k, coolantLvl/cap, flowScale, masses only), the bank radiator
  @004ae73c (adds zone damage + frozen ambient-300 target), UpdateCoolant @004adbf8 (damage-driven
  leak). No airflow term. "Cools only near standstill" = generation (~0.07·m·v²·ratio²/s kinetic
  + 0.2·m|v||a| gait term) vs a FIXED-capacity chain. Authentic behavior, not a port break.
 - **Seek-4 heat multiplier is 2.04x, NOT 16x** [T1]: ratio = gears[cur]/gears[rec] CLAMPED >= 1
  (@004b8d18), so gears 1-3 are heat-identical and gear 4 = (0.9999/0.7)^2 = 2.04x -- times the
  supercharge speed feedback (v x1.4284 -> v^2 x2.04) = ~4.2x total at unrestricted cruise.
 - **Equilibrium math validated against live data** [T2]: linearized per-second stage rates
  (massScale=3 refrigeration included): myo->cond 44.4k/K-of-dT, cond->bank 28.8k x valveShare x
  (2.73·Tc - 0.91·Tb), bank->ambient 23.1k x hsCount x (Tb-300). Solver (scratchpad equilib.py)
  reproduces the measured BLH gear-4 governed equilibrium (predicted 1783 @ v=44 vs measured
  1718-1770, seek4.log) and the radiator shed line. Balanced-valve chain capacity ~= 1.9e7/s at
  Tm=2000 on EVERY chassis; **loop-5 boost (detent 50) = 2.7x capacity (5.07e7/s)** -- the
  authored lever that sustains unrestricted gear-4 on a vulture (0.67-0.83x of capacity).
 - **Why the big chicken walkers cook**: gear-4 unrestricted generation / failure-line capacity =
  1.77x for blkhawk (hovers at ~1770, keeps moving) vs 2.2-2.8x for madcat/thor/vulture-if-faster
  (blows THROUGH failT 2000 -> myomers fail -> speed+turn freeze; ~6 s to degT, ~12 s to failT
  from cold; Condenser5's MFD bar pegs in ~3 s since load full-scale ~= T 300). GeneratorD shares
  loop 5 everywhere, so a pegged loop back-heats it toward its own trip -- the freeze amplifier.
 - **Residual for the verdict** [T3]: vulture mass = blkhawk mass (60k), so the field split
  vulture-vs-blkhawk requires the vulture's CLIP-DERIVED base speed (mech+0x34C, run-cycle avg
  root speed @0x4a80d4) and/or gait |a| bob to exceed the BLH's measured 43.6 u/s -- at equal
  speed the two compute identically. Type-15 gamedata splits 33-36/43-44 into exactly two motion
  groups (avatar/madcat*3/sunder*2/thor*2/vulture*2 vs blkhawk*2/loki*3/owens*2) [T4 field
  meaning]. Bench lane: measure vulture v/|a| at gear 3/4 vs BLH.
 **#173 AUTHORED-DATA lane result (2026-08-13) [T1]: the heat/cooling plumbing is
 BYTE-IDENTICAL across ALL 18 mech variants** (parsed from `content/BTL4.RES` type-17
 SubsystemModelStream, parser `scratchpad/night17/myoaudit.py`; walk verified byte-exact --
 every stream consumed to the byte via the self-describing `subsystemModelSize@+0x24`).
 Every chassis gets the SAME: Myomers record (velEff 0.995 / accEff 0.8 / tmass 2.5e5 /
 cond 1.9e5 / seekV 0.3/0.5/0.7/0.9999 rec 2, hsi=8→Condenser5, vsi=13→GeneratorD), six
 identical Condensers (tmass 4.2e5, cond 3.15e5, refrig 3.0, degT=failT=2000), identical
 Generators (rated 10000 V, tap 6, GenD always hsi=8→Condenser5), identical Reservoir
 (cap 20, squirt 0.5).  Loop 5 = GeneratorD + Myomers ONLY on madcat/vulture/vul1/thor/
 thr1/loki/lok1/lok2 (the fielded chicken walkers carry the MINIMUM possible loop-5 load);
 the variants with extra loop-5 weapons are mad1 (2 MLaser), mad2 (ERPPC), ava1 (2 ERMLaser),
 own1 (AFC25), sunder (2 MLaser) -- no walker bias.  Manual COOLANT LOOPS pages print the
 same GenD+Myomers loop-5 topology for every mech [T1 primary].  The ONLY chassis-varying
 heat inputs are moverMass (type-15 GameModel +0x00: owens 35t, vulture/blkhawk 60t, loki
 65t, thor/avatar 70t, madcat 75t, sunder 90t) and top speed/accel (manual stat sheets:
 Loki/Thor 143 N / 182 SC kph, Vulture ~175 N [OCR-fuzzy T3], Owens 154, Blackhawk ~190
 [OCR-suspect]; maxAccel authored: madcat 20, loki 15, vulture/thor/avatar/sunder 10,
 blkhawk 30, owens 50).  m·v² through the verified formula puts the 60-75 t / 143-175 kph
 walkers at ~1.6-2.2x the light-humanoid kinetic heat rate into the SAME fixed-size loop --
 seek-4 zero-cooling degradation bound ~7-10 s (vulture worst) vs ~16 s (owens).  Verdict
 lean: AUTHENTIC authored balance (fast-heavy chassis cook by design); residual port-defect
 risk sits ONLY in the runtime velocity magnitude (locomotion lane -- cross-check the port's
 per-chassis top speed vs the manual print figures; bench lane #9).
 **#173 BENCH lane result (2026-08-13) [T2 measured, live receipts]: Elengil's report
 REPRODUCED and quantified; the equilibrium is a GOVERNOR THERMOSTAT, and it is chassis-blind
 above a speed threshold.**  Rig: solo grass/day expert, full forced throttle + constant 0.12
 turn, BT_FORCE_SEEK, 150 s/run (`scratchpad/night16/myo173.sh` + `myo173b.sh`, digest
 `digest173.py`, logs `content/m173_*.log`).  Streaming receipts on every chassis matched the
 authored data exactly ([heat-link] Myomers sinkIdx=8→Condenser5 k=1.9e5 m=2.5e5; [spec]
 hsCount 18/14/28/13/15/90t...; C5 mScale=3 flowScale=0.1667 balanced).  Matrix (peak v /
 steady v u/s, time to degT 1000, equilibrium T, C5 end):
 - vulture seek-1: 24.5/21.1, never, **eq 423**, C5 333 — gears 1-3 are harmless (ratio floors 1).
 - vulture seek-4: 56.9/31.9, **13.9 s**, **eq 1801**, C5 1410 (loop-load bar full-scale ~15-20 s).
 - madcat seek-4: 56.8/28.4, 12.1 s, eq 1820, C5 1420.  avatar seek-4 (HUMANOID, same 143-kph
  speed class in the port): 56.2/29.5, 14.3 s, eq 1819, C5 1424 — **identical cook**.
 - thor seek-4: 34.6/34.3, 79.9 s, 1311 still climbing; sunder (90 t): 34.6/34.0, 49.4 s, 1592
  climbing — the port's slow (87-kph-class) chassis merely take longer.
 - STRAIGHT-line control (no turn): eq 1802, |a|≈20 — same as turning, so the accel term is
  GAIT-SURGE noise in the velocity ring, not commanded motion (see residual below).
 - **Valve boost, the authored counter-play** (one MoveValve press on Condenser5, detent 1→5 =
  share 0.5): myomers still governor-pinned ~1715 but **sustained drive 0.68 vs 0.54 balanced
  (v 39-41 vs ~32) and C5 968 vs 1410** — boost buys back speed, not myomer temp.  The run's
  accidental map-edge stop measured the standstill shed: 1723→586 in ~35 s (−10/s) — "heat
  only sheds near standstill" is the fixed-capacity chain being honest (slope +0.02/s while
  cruising).
 **Mechanics of the thermostat**: T rises past degT → AvailableOutput derates → v falls →
 m·v² generation falls → equilibrium lands ~1700-1820, just UNDER failT 2000, on EVERY chassis
 fast enough to out-run the loop — better cooling raises the sustained SPEED at the same
 pinned T.  Steady seek-4 cruising never freezes (failT) solo; the field freezes need adders
 (damage dmgGain ≤2x, combat weapon heat, respawn transients, a GenD trip on the shared loop).
 **Manual cross-check settles the speed residual in favor of AUTHENTIC** [T1 print, re-read
 via fitz p24/25]: MadCat prints **175 N / 220 SC**, Vulture **175 N**, Thor/Loki 143/182 —
 the machine's big chicken walkers were FASTER than the port's measured 143 N / 205 SC (thor
 port 87/124 is far under its print), so authored m·v² generation on the pod was ≥ what these
 benches produced; the phenomenon (manual p21 warns of it explicitly) is authored.  The port's
 per-chassis top-speed shortfalls remain a LOCOMOTION-lane item but cannot rescue a "port too
 hot" theory — correcting them makes it hotter.
 **Verdict: AUTHENTIC mechanism + authored balance [T2]; ONE bounded port-side severity
 residual [T3]** — the accel term's |a| operand (15-sample velocity-ring derivative) is
 frame-cadence-sensitive: at ~59 fps a straight flat cruise reads |a|≈14-20 (thor/sunder's
 smoother gait ≈4) and the accel term carries 50-65% of steady seek-4 generation; the same
 gait bob sampled through the pod's 28 Hz ring would read ~2-3x smaller, so the pod's
 equilibrium may have sat ~1200-1500 (shallow governor, near-full supercharge speed held)
 instead of the port's ~1800 (deep governor, speed sags to ~0.54).  Same phenomenon, softer
 edge — this is the gap that would reconcile Oracle's "ran 182 kph sustained" pod memory.
 If field calibration is ever wanted, the knob is the |a| feed's cadence (sample the ring at
 28 Hz), NOT the formula constants (byte-exact, #137).  Also note the CLIMB term is inert in
 the port (g=0, EnvironmentZone unwired) — the pod added heat on slopes that we do not.
 **#173 DECISIVE VALVE-BOOST + CADENCE EXPERIMENT (2026-08-13 night17;
 `scratchpad/night17/m173v_bench.sh`/`m173v2_bench.sh`/`m173v_check.py`, logs
 `content/m173_vul_[ABCEFG]*.log`, captures `content/vlv_*.png`) — the residual above is now
 MEASURED and the verdict sharpens:**
 - **Valve delivery PORT-FAITHFUL, receipt-PROVEN [T2 live / T1 operands]**: the `BT_VALVE5=N`
  hook (mech4.cpp, env-gated) presses Condenser5's MoveValve N times; every landed detent is
  receipted (`[valve] Condenser5 valve -> detent 5/50/0` + the zero-sum redistribute lines:
  share 0.5 → 0.909091 → 0, others 0.1 → 0.0182 → 0.2, exactly @0049f788's flow=valve/total).
  C5 responds as flow dictates (cruise Tc5: balanced 1396 → detent-50 712 → CLOSED 1902 pegged).
 - **The bottleneck SHIFTS to the valve-independent stage [T1+T2]**: myomer→Condenser5
  conduction measures 4.1–5.0e4 heat-units/K of dT in EVERY run (balanced/boosted/closed,
  both cadences) — constant because @004ad9ec's exponent reads the CALLER's own
  flowScale@+0x15C (the myomer's, pinned 1.0), never the condenser's valve share [T1 raw
  pseudocode]. The solver's 2.7x boosted whole-chain throughput is CONFIRMED live (4.8e7/s
  through the chain at 60fps detent 50) but its "sustains under degT 1000" corollary is
  WRONG at any cadence: sub-1000 Tm caps stage-1 flux at ~3e7/s < supercharge generation, so
  the governor always engages; the boost buys SPEED at a lower pinned Tm, never sub-1000 Tm.
 - **The cadence residual is REAL and ~2x [T2 measured]** (BT_MYO_RING_HZ=28 knob landed in
  mech4.cpp — cadence only, constants untouched): at ~59fps the |a| gait-surge noise carries
  64–66% of steady seek-4 generation (|a|≈17–26 flat grass); at 28Hz it drops to 24–32%
  (|a|≈6–7) and generation at equal speed HALVES. The six curves (solo vulture seek-4, t1000 /
  equilibrium / held v): 60fps bal 12.3s/1800/28 · 60fps d50 12.3s/1675/44 · 60fps CLOSED
  13.9s/NO-EQ-1936-still-climbing/11.7 (drive 0.21, C5 pegged 1902 — the wrap-trap signature)
  · 28Hz bal 18.3s/~1710/39–41 · **28Hz d50 22.5s/~1500 STABLE/54–57 = FULL supercharge speed
  sustained indefinitely** (equilibrium sits on the soft governor shoulder, zero speed cost;
  500 under failT). **The machine's loop-5 boost genuinely sustained unrestricted seek-4 — the
  manual's Super Charged print is vindicated at pod cadence; the port at ~59fps over-punishes
  by ~2x. Severity verdict flips to PORT CADENCE DEFECT (fix = 28Hz ring sampling; default
  promotion = owners' decision).** Balance verdict unchanged: balanced-valve seek-4 cooks at
  BOTH cadences (authentic, manual-warned).
 - **Field "3–4s to governor" ≠ spawn preheat [T2]**: field-shaped run (autofire at scenery
  from frame 0, no pre-settle) still takes 15.9s to degT from cold; GenD/loop-5 preheat is
  real but slow (GenD 666 by 29s). The 3–4s feel is a WARM-START baseline (T≥800 → <3s) plus
  the deep 59fps governor making the derate obvious.
 - **Heat-MFD valve slider IN SYNC [T2 pixel-verified]**: the vertNormalSlider renders
  coolantFlowScale@0x15C — the SAME cell the flow model consumes (single-cell design, no
  split possible); measured bar fractions 0.20/0.51/0.90/0.04 for receipt shares
  0.167/0.5/0.909/0.0 (perfectly linear; `vlv_*.png`). Detent 50 (≈90% up) and detent 0
  (bottom) are clearly distinguishable; the residual legibility trap is CLOSED-vs-STARVED
  (share 0 vs 0.018–0.1 ≈ 1–2px apart) and the silent 4th-press wrap — "maxed and couldn't
  shed" field reports match the run-C wrap signature, not a display desync.
@@ -234,29 +234,3 @@ and asserts [crit] DESTROYED + [techstat] X + zero named FIRED after + REFUSED>0
 fires; P2 partial-crits to 0.5 and asserts the weapon KEEPS firing (freeze regression, the #164
 upstream writer). LESSON (gotcha SS30): BT_KILL_SUBSYS writes the cells directly and can only
 validate READERS -- state benches must drive the real writer path.
 ## weapons_sweep (night16) -- the all-families weapons bench (#171/#174/lane-12, PASS 2026-08-13)
 `scratchpad/night16/weapons_sweep.sh` + `weapons_check.py`: ONE 2-node bench proving every
 weapons family on one tree.  Rig: **A=mad2 "Zanin Neko"** (AFC50 + ERPPC + LRM15 + SRM6 x2 +
 ERMLaser x2 -- all three families on one chassis, the AC carrier per L4GAUGE.CFG) vs **B=madcat**
 (AFC100 + LRM15 x2 + ER lasers), grass/day, `BT_GOTO=enemy` + throttled BT_AUTOFIRE +
 BT_AF_MISSILE (periods 4/7), 300s, env `BT_DMG_LOG BT_PROJ_LOG BT_DEATH_LOG BT_AMMO_LOG`.
 Asserts per family: energy named [emitter] FIRED + type-3/4 dmghits; AC type=1 dmghits ALL
 burst=1 + ballistic (guided=0) PUSHes ALL dmg=0 (#171 hitscan; damage DETs reconciled to guided
 pushes per log); missiles [ammo] FIRED + IMPACT dmg>0 with burst>1 + type=2 dmghits (contact
 through the 4.0u fuze at drag-governed speeds); #174 cascade-per-life<=1; bystander receipts
 individually classified; hard rig floors.  `--selftest` feeds the detector pre-#171/pre-#174
 shapes + bystander false-positive shapes and requires each to FAIL (run first, gated).
 Rig lessons [T2]:
 - **avatar-vs-avatar forms NO locks** (three prior runs, 0 dmghits even from lasers) while the
  madcat family fights fine -- dodge with madcat-family chassis; `mad2` is the AC carrier.
 - Node->pilot-page mapping: `-net 1501` claims pilot `127.0.0.1:1502`, `-net 1601` claims
  `:1602` (net port + 1) -- sed each pilot page's vehicle= line EXACTLY for mixed pairings.
 - **BYSTANDER receipts legitimately fire in a 2-mech fight**: the production aim ray designates
  SCENERY constantly (the ac_bench BT_FIRE_AT_ICON lesson generalizes to the real pick), and a
  mech crossing a scenery-locked round's path detonates it -- the binary world-sweep shape
  (FUN_0042291c via @004bef78 [T1]).  All 11 receipts of the PASS run were true positives
  (struck = the other live mech, aim 146-1016u away, one paired IMPACT each;
  `scratchpad/night16/bys_forensics.py`).  A zero-receipt assertion is miscalibrated; assert
  the false-positive conditions instead (self-strike / phantom id / det-at-lock-aim / missing
  paired delivery / print-cap 24 reached).
@@ -428,14 +428,6 @@ in the mech reset (part_012.c:9446-9447) — a genuine regression, not dead cont
 inbound lock/launch on the player exists; DistanceToMissile = range to nearest inbound missile
 (far default when none). The authored watchers then reproduce the accelerating alarm unchanged.
 **ERA-WITNESS CONFIRMED (2026-08-13, Oracle, via the distributable soundboard):** Oracle found
 the sample himself on the board — `[1:74] IncomingAlarm01 n58` — and confirms it is the
 incoming-missile alarm, adding "**it was sped up in the pods**": independent field confirmation
 of both the sample identity AND the authored DistanceToMissile→tempo acceleration (the 100..800
 → tempo 600..10 mapping above). Gitea #83 (comments 1781/1783). The reconstruction target is now
 fully specified except the binary's exact IncomingLock/DistanceToMissile update rule (what sets
 +0x3FC/+0x400 — seeker-locked-only vs any-inbound, and nearest-missile selection).
 <a name="f5"></a>
 #### F5. FootStep is a contact LEVEL, not a clip-transition pulse (High, CONFIRMED)
@@ -625,60 +617,6 @@ reduce source-pool starvation.
 ---
 <a name="seqcensus"></a>
 ### The complete authored trigger-note census (2026-08-13, [T1]) — and the game-rate soundboard
 **The playback-rate model, end to end:** effective output rate = WAV declared rate x
 2^((note-60)/12) (`L4AUDIO.cpp:21-24` BTNotePitchFactor -> `alSourcef(AL_PITCH)` at :971
 per-SetupPatch; per-zone SoundFont tuning is baked into each extracted WAV's declared rate per
 F2, so note 60 == authored pitch). Sources default to note 60 (`AUDSRC.cpp:18 DEFAULT_NOTE`);
 the ONLY senders of NoteAudioControlID are AudioControlSequences (grep-verified engine-wide),
 so **every off-file-rate playback in the game comes from a sequence note**. Dynamic pitch-cents
 (doppler, pitch-scale watchers) modulate on top and are not part of the base rate.
 **BTL4.RES AudioControlSequence record format** (matches the ctor read order,
 `AUDSEQ.cpp:165-194`): `[classID=75 i4][objectID i4][dump i4][looped i4][targetObjID i4]
 [divisionsPerBeat i4][tempo i4][eventCount i4][count x (tick i4, ctlID i4, value f4)]`;
 ctl 8 = note, 1 = start, 2 = stop, 6 = attack-volume. Every object record in the audio
 streams is `[classID i4][objectID i4][payload]` with ClassID = the `VDATA.h` enum
 (AudioStateTrigger=28, sequence=75, splitter=76, DirectPatchSource=1001, PatchResource=1004,
 PatchLevelOfDetail=1011 — LOD payload `[voiceCount i4][renderType i4][suspendSecs f4]
 [bank u1][patch u1][maxFilter i4]`). Patch resources/LODs live in the shared
 `StaticAudioStream` (id 0); vehicle `*int.scp` streams reference them cross-stream.
 Sequence -> target -> source -> resource -> LOD resolution yields the patch for every one of
 the **324 sequences (15 unique patterns)** in the RES; cross-validated against live SetupPatch
 logs (mp4l/solo/genedge).
 | notes (authored order) | looped | patch | trigger (watcher census) |
 |---|---|---|---|
 | 16,19,23 | yes | 2:113 Warnings01 | coolant-leak voice loop (ReportLeak, 19 subsystems) |
 | 16,33,36 | no | 2:113 | ammo cook-off countdown voice (FireCountdownStarted) |
 | 29,16,26 | no | 2:113 | generator-out voice phrase (GeneratorState/GeneratorOn) |
 | 29,16,40 | no | 2:113 | Entity.SimulationState warning voice phrase |
 | 30,18 / 49,43 / 95 | no | 2:115 Death01 | death phrases — a **random 3-pick list** `[3][seqA][seqB][seqC]` at death |
 | 36,84 | no | 1:36/40/41/56/70/71/75 | weapon loaded/jam/misfire clunk-then-blip (WeaponState) |
 | 36,70 | no | 2:64 DestroyedInt06 | subsystem destroyed (SimulationState) |
 | 58 x6 | yes | 1:74 IncomingAlarm01 | missile-lock beeper (IncomingLock; DistanceToMissile -> tempo) |
 | 51,51 | yes | 1:83 ProgramButton01 | weapon-configure ticker (ConfigureActivePress) |
 | 30,48,46,96,44,63,66 | no | 2:83 MechExplosion01 | brnext.scp |
 | 45,38 | no | 2:97 MechFireLoop01 | brnext.scp |
 | 57,69,46,24 / 69,51 | no | 2:119+120 AuxExplosion01(+LOD2) | bigexp.scp / medexp+mbnexp.scp |
 Net: **92 (sample-zone, note) pairs play rate-shifted vs their WAV header** (worst: the
 Warnings01 voice zones at x0.079-x0.315 — the raw files render Yip's voice 3.2x-12.7x too
 fast, which is why the old soundboard's voice sounded wrong; weapon ready blips x4; death
 phrase C zones x7.55). Bonus census finding: **Death01 z8-11, AuxExplosion01 z6-7 and
 AuxExplosion01LOD2 z3 are reachable by NO authored note** — dead zones.
 **Tools:** `tools/soundboard.py` now plays every sample at its game rate (header-rewritten
 temp copies, per-note buttons, raw-rate toggle); `tools/mksoundboard.py` builds
 `dist/SOUNDBOARD_BT411.zip` — a self-contained tester soundboard page (file://-safe `<audio>`
 elements, `preservesPitch=false`, playbackRate = the game shift) whose COPY buttons emit the
 canonical `[bank:patch] Name n##` reference string for bug reports. Both share the census
 table (`GAME_TRIGGERS` in soundboard.py).
 ---
 <a name="validations"></a>
 ## 4. Validations and KB corrections (no fidelity gap)
@@ -745,5 +683,3 @@ table (`GAME_TRIGGERS` in soundboard.py).
 **Regression harness:** re-run the audit capture after each block and assert with
 `scratchpad/audio_coverage.py` that the dead-watcher count monotonically drops; A/B pitch/volume
 spot checks against the recovered-rate table for the bank regeneration.
 **EAR-VERIFIED 2026-08-13 (epilectrik): Yip's Warnings01 voice zones at GAME rate (x0.079-x0.315) confirmed correct by ear via the distributable soundboard — the raw-file 3.2-12.7x overspeed is settled as a storage artifact, the note-transposed rate is the pod's voice. SOUNDBOARD_BT411.zip approved for tester distribution.**
@@ -217,7 +217,7 @@ yet reconstructed (parse-skipped). The separate window is now the **live viewer*
 - **Pod MFD port-name reconcile (pod-only):** a ~5-line positional dual-name fallback in `SVGA16::Update`
  (`UL = GetGraphicsPort("auxUL2"); if(!UL) UL = GetGraphicsPort("Heat"); …`) so BT MFDs reach the pod's real
  monitors. NOT a CFG rename (the CFG uses BT names pervasively + `MUNGA_L4` is shared with RP's aux names).
- **Polish:** overlay-plane compositing over `sec`; a pod-accurate RGB-packing toggle; texture atlas
+- **Polish:** ~~overlay-plane compositing over `sec`~~ (DONE 2026-08-13, row 28); a pod-accurate RGB-packing toggle; texture atlas
  (one lock/upload); device-reset recreate hardening for the additional swap chain.
 ## Widget reconstruction — the real fix (in progress; mapped by `bt-gauge-widget-recon-map`)
@@ -691,7 +691,7 @@ reconfigure/externalConfigure); parse-skip list EMPTY ([gskip]=0), all 50 attr b
 | 25 | sec: headingPointer needle + numeric | YawPitchRoll decomposition (engine-convention port of euler[0]) | LIVE: 089→120 while turning under BT_GOTO | T2 | CORRECT |
 | 26 | sec: numericSpeed | LinearSpeed (abs(adv)/dt) | LIVE: 0 parked → 176-182 walking | T2 | CORRECT (units question stays open) |
 | 27 | sec: digitalClock MISSION TIME | engine mission clock (format enum arg) | LIVE: 09:19→06:04 counting | T0/T2 | CORRECT |
-| 28 | sec: sectorDisplay (overlay) | localOrigin → Round(−Z·0.01)+500 / Round(X·0.01)+500 | commit 4a4ec68 [BT_SECTOR_LOG live]; NOTE: overlay plane not composited into the dev sec cell (polish item) — data verified by log | T2 | CORRECT |
+| 28 | sec: sectorDisplay (overlay) | localOrigin → Round(−Z·0.01)+500 / Round(X·0.01)+500 | commit 4a4ec68 [BT_SECTOR_LOG live]; **overlay plane COMPOSITED 2026-08-13** (branch sec-overlay-composite): every sec expand (BTDrawGaugeSurfaces, BTDrawCockpitPanels, SVGA16::Update case 0, glass BlitSurface) now uses sec\|overlay = 0xFF — clut0 already held the arcade DAC's combined table (BuildSecondaryPalette spreads each port over the other's bit combos), so the fix is mask-only; the glass dirty token gained the overlay port so the numerals repaint. Verified live: SECTOR ticks + SCALE:1000 at spawn (1000 = radarRange init; the DOS shot's 4000 is max zoom) + the btsec1ov graticule now draws | T2 | CORRECT |
 | 29 | sec: schematic ARMOR view (cmArmor/multiArmor, dama.pcc) | zone damageLevel ×100 → adpal ramp | incr.3/4; all-green pristine LIVE | T2 | CORRECT (AUTH-STATIC all-green solo) |
 | 30 | sec: schematic CRITICAL view (cmCrit) | subsystem simulationState/damage | LIVE this audit: N-cycle shows the full subsystem list (GEN A-D, LOOP 1-6, HUD, SENSORS, GYRO, TORSO, weapons) | T2 | CORRECT |
 | 31 | sec: schematic HEAT view (cmHeat) | subsystem currentTemperature tint | #6 pixel-verified; re-cycled this audit (mask 0x450421→0x490421→0x510421) | T2 | CORRECT |
@@ -1,604 +0,0 @@
 # Weapons Drift Audit — binary vs port, per weapon lane (2026-08-13)
 **What this is:** the AI-facing ledger of the per-weapon-class physics audit
 (the #171 batch + this session's raw-byte verifications). The tester-facing
 twin is `docs/WEAPONS_FIELD_GUIDE.txt` — keep the two in sync; the field
 guide translates every address below into behavior and carries the era
 questions to the group.
 **Evidence tiers:** T1 = decomp/raw-byte verified · T2 = runtime receipts ·
 T3 = port approximation · T4 = guess. Addresses are `@ADDR` in the rebuilt
 2026-08-06 export (`reference/decomp/all/part_*.c`).
 **HOLD notice:** corrections in the "Corrections owed" section are HELD for
 the #171 group decision per Oracle's HOLD — do not sweep them into the KB or
 the queued ticket comment until the group signs off.
 ---
 ## Verdict summary
 | # | Lane | Verdict |
 |---|------|---------|
 | 1 | Lasers (Emitter 0xBC8, 6 marks) | FAITHFUL |
 | 2 | PPC/ERPPC fire chain (0xBD4) | FAITHFUL |
 | 3 | PPC rider (a) victim display scramble | KNOWN-DIVERGENCE [T3] (unimplemented; visual OPEN) |
 | 4 | PPC rider (b) victim generator short | FAITHFUL (tag-label scope OPEN) |
 | 5 | Seek-voltage gears (#21) | FAITHFUL (brick claim RETRACTED 2026-08-13 -- constant-width misread; binary clamps overcharge to 1.0 and fires; witnesses 6-0) |
 | 6 | Autocannons AFC25/50/100 (0xBCD) | FIXED-THIS-BUILD (#171; T2 field receipts pending) |
 | 7 | AC cosmetic tracer visual | FIXED-THIS-BUILD (recovered TOF constants 2.0f + 1/9.8 applied, b94636b) |
 | 8 | GAUSS (fielded, 0xBCD record) | FAITHFUL (correction sweep owed, HELD) |
 | 9 | GaussRifle class 0xBCE (+ shell 0xBD1) | FAITHFUL (dead code both sides) |
 | 10 | MissileLauncher core (0xBD0, all racks) | FAITHFUL |
 | 11 | Missile presentation (salvo visuals/gains) | KNOWN-DIVERGENCE [T3] (presentation-only) |
 | 12 | Missile bystander sweep | FIXED-THIS-BUILD (capsule sweep [T3 radius]; T2 field receipts pending) |
 | 13 | Missile splash (@0042fad0) | FAITHFUL under stock content (gate wiring [T3] on custom eggs) |
 | 14 | LRM flight + loft | FAITHFUL |
 | 15 | Streak racks (strk model) | FAITHFUL |
 | 16 | NRK5 / nrk-fed racks | FAITHFUL (KB model-table wording sweep owed) |
 | 17 | Shared reload/ammo/eject chain | FAITHFUL |
 | 18 | Jam roll | KNOWN-DIVERGENCE [T3] (extra gate — remove) |
 | 19 | Incoming-missile warning | OPEN (consumer unlocated; era testimony wanted) |
 ---
 ## Per-lane entries
 ### 1. Lasers — Emitter 0xBC8 (SLaser/ERSLaser/MLaser/ERMLaser/LLaser/ERLLaser)
 One C++ class, data-differentiated; **no pulse variants exist** in shipped
 content.
 - **Binary [T1]:** trigger-time hitscan beam @004bace8: charge-gear state
  machine (4 states, no jam/ammo), lock required (`*(mech+0x388)!=0`),
  viewFireEnable look-direction arming, `dist<=effectiveRange` gate AT FIRE.
  K byte-confirmed **1.0f @004b9c98** → effRange = (1.0 − zoneDamage) ×
  WeaponRange. Damage = authored × (charge/seekV)². ONE aimed single-panel
  message. Out-of-range: beam draws, damage silently discarded. 0.2s
  target-tracking afterglow. Hard heat-FAILURE stop (no gradual
  degradation). Heat/gear-toggle vet-expert only.
 - **Port [T2]:** `emitter.cpp` live-verified same machine end-to-end; 60Hz
  sub-stepped Loading tick (±1% snap-window robustness, port mechanism);
  beam from real `ermlaser.bgf` tube via `BTPushBeamKind` ([T3] dpl layer
  unported, #164 stuck-beam guards).
 - **Verdict:** FAITHFUL.
 ### 2. PPC / ERPPC — 0xBD4 fire chain
 - **Binary [T1]:** one-line tail-call into `Emitter::FireWeapon`
  (@004bb878); vtable/ctor/CSS otherwise identical — everything distinctive
  is authored data (EnergyDamage(4), 12/16 pts, 900/950 range, 5.0s
  recharge, blue pip).
 - **Port [T2]:** same tail-call structure; full charge = authored 12/16
  verified live; blue beam under its own renderer case.
 - **Verdict:** FAITHFUL.
 ### 3. PPC EnergyDamage rider (a): victim display scramble
 - **Binary [T1 call, T4 visual]:** @004a0230 (~0x4a03f3): damageType==4 →
  no-arg virtual on app-global +0x4c — identified via the port's own app
  decode as THE GAUGE RENDERER (`DAT_004efc94+0x4c` = GetGaugeRenderer,
  `btl4mppr.cpp:249`, cf. `btl4galm.cpp:404`) — a cockpit gauge/MFD scramble
  per Energy hit. Exact visual/duration unread (which vtable slot +0x4c does
  visually: OPEN).
 - **Port:** NOT implemented — grep of the port `TakeDamageMessageHandler`
  body finds no damageType==4 gauge call. Candidate for the #156/
  weapon-state polish list.
 - **Verdict:** KNOWN-DIVERGENCE [T3] (unimplemented rider; visual detail
  OPEN — era testimony wanted BEFORE building; question shipped in the
  field guide).
 ### 4. PPC EnergyDamage rider (b): victim generator short
 - **Binary [T1]:** @0049c690: damageType==4 + critical subsystems in struck
  zone → one random critical rolls `ForceShortRecovery` @004b11bc
  (brownout, charge stall, cockpit electronics dip). Tag 0x50f4bc
  Generator-vs-PoweredSubsystem label conflict still open.
 - **Port [T2]:** implemented — `mechdmg.cpp:471-477` (damageType==4 &&
  criticalSubsystemCount>0 → ForceShortRecovery).
 - **Verdict:** FAITHFUL (tag-label scope OPEN). NOTE: the two riders'
  port status differs — never lump them in vet-facing docs.
 ### 5. Seek-voltage gears (#21) -- brick claim RETRACTED 2026-08-13
 - **Binary [T1]:** `ToggleSeekVoltage` @004ba478 (vet/expert only): wrap
  from gear 4→1 skips `ResetFiringState` → charge above snap window →
  ~~overcharge reads ZERO → PERMANENTLY BRICKED~~ **RETRACTED: _DAT_004ba830
  is a DOUBLE 1.0 (width misread); the binary CLAMPS overcharge to 1.0 and
  the weapon fires -- the machine never bricked. Era witnesses (6-0) triggered
  the re-audit; full detail in context/decomp-reference.md.**
 - **Port [T3 deliberate, documented]:** overcharge counts as Loaded (issue
  #21, 38 benched rescues) — the port's ONE deliberate energy-lane
  divergence; lives in the deviations appendix.
 - **Verdict:** KNOWN-DIVERGENCE (deliberate rescue of a shipped latent
  brick).
 ### 6. Autocannons — AFC25/50/100 (ProjectileWeapon 0xBCD)
 - **Binary [T1]:** trigger-time hitscan @004bc104: lock required, batched
  per TracerInterval (all authored 1), fire-time `dist<=live
  effectiveRange` gate, ONE single-panel Ballistic message to the locked
  target, out-of-range batch silently discarded (round consumed, tracer
  flies), gyro recoil kick dmg/16 dir (0, 0.6, −1.5) (constants
  3.0f/0.0625 byte-confirmed), jam p=0.41·T/Tfail clamped [0.05,1.0]
  vet/expert-only, EJECT tap=one round / hold 3.0s=dump bay, shell 0xBD1
  dead code.
 - **Port [T2 bench]:** `projweap.cpp` at bbb3b4e implements the hitscan
  line-by-line (13-round batch + DET dmg=0 receipts); closes #112
  single-panel and the #168 mis-fix. Field verify armed for Friday (local
  rig can't form avatar locks).
 - **Verdict:** FIXED-THIS-BUILD (#171 batch; T2 field receipts pending
  Friday). The field guide carries the #27 "rounds can genuinely miss"
  RETRACTION.
 ### 7. AC cosmetic tracer visual
 - **Binary [T1 weapon-side, T4 renderer-side]:** `launchVelocity.z =
  −effRange/sqrt(2.0·(1/9.8)·muzzleHeight)` — both constants BYTE-CONFIRMED
  this audit: @0x4bc678 = 2.0f (float32); @0x4bc67c = 1/9.8 stored as an
  **80-bit x87 EXTENDED** (bytes `8a 6c 37 99 43 c6 fa d0 fb 3f` — does NOT
  decode as float32, likely why it stayed "unresolved") → ~1s to cross full
  range for a ~5u muzzle. Drawn by the unported 'Tracer' dpl renderable;
  renderer feed unrecovered.
 - **Port [T3]:** `projweap.cpp` still ships 1.0f stand-ins for both
  constants (marked 'unresolved' — now false) and drops the muzzleHeight
  term: tracer speed = effRange/1.0. Numerically close for ~5u muzzles by
  accident.
 - **Verdict:** KNOWN-DIVERGENCE [T3] (cosmetic-only; constants recovered —
  CLOSE IT next build; stale 'unresolved' comments must go in the same
  change).
 ### 8. GAUSS — the fielded weapon (lok2 'GAUSS' record)
 - **Binary [T1 raw bytes]:** BTL4.RES @0xf16e6: classID **0xBCD
  ProjectileWeapon** (bytes `cd 0b` / `d0 01` size 0x1D0, verified this
  session) — the pod's Gauss ran the FULL AC hitscan chain WITH damage:
  20 pts Ballistic single-panel out to 900, 8s cycle, 16 rounds
  (AmmoBinGAUSS 0xBCB), near-cold (1e6 heat), yellow pip, 1.25 gyro kick.
  Class histogram across all 18 streams: 78×0xBC8 / 16×0xBCD / 30×0xBD0 /
  14×0xBD4, **ZERO 0xBCE**.
 - **Port [T1]:** factory dispatches 0xBCD →
  `CreateProjectileWeaponSubsystem` (`mech.cpp:1939-1942`) — the port's
  fielded GAUSS ALSO deals AC-hitscan damage. The "port damage is an
  invention" framing (commit 80037a7, queued #171 comment, projweap
  banner) is WRONG at the fielded level — port and binary agree.
 - **Verdict:** FAITHFUL (mechanics match). Correction sweep owed — HELD
  (see below). Era question reframed and shipped in the field guide
  (headline). Chassis attribution (lok2 vs the weapsub scan's nearest-name
  'blh') rides the flagged heuristic — the field guide asks the vets which
  mech carried it rather than asserting.
 ### 9. GaussRifle class 0xBCE (+ shell 0xBD1)
 - **Binary [T1]:** real code, dead content: `FireWeapon` @004bdca4 = one
  store (charge=0), no damage/beam/heat; zero shipped records instantiate
  it. Shell 0xBD1 `Make` @004be384 has zero callers; flight body
  unreachable.
 - **Port [T1]:** same classes exist (GAUSS.CPP shipped source), same
  factory case, never constructed by shipped content; equally dead.
 - **Verdict:** FAITHFUL (dead code both sides). The CLASS-level "@004bdca4
  is a no-op" claim stands even after the fielded-Gauss correction.
 ### 10. Missile family core — MissileLauncher 0xBD0 (SRM/LRM/Streak/NRK)
 One class, data-differentiated.
 - **Binary [T1]:** genuinely flying, the ONE ungated family: lock-to-fire
  via the shared Loaded case (no dumb-fire trigger path, no Streak-special
  code), one entity per salvo, per-missile dmg = authored/count, seeker
  homes on the fire-time picked panel point with constant-bearing lead,
  4.0u fuze + swept contact (bystanders hittable), lock-drop on target
  death (@0049fb54 movementMode 2||9), WeaponRange is fire-control display
  only — reach is physics (burn + 10.0s coast, y<−1.0 kill plane, expiry
  fizzles), impact cluster roll `min(Random(n)+n/4, n)` then zone −1
  lottery with per-burst re-roll on the victim.
 - **Port [T2]:** full chain live-verified (382-impact night-9 corpus
  matches the roll exactly; Streak 277 u/s; fuze/lock-drop benched);
  direct Dispatch instead of the SubsystemMessageManager (deliberate,
  preserves burstCount, #95).
 - **Verdict:** FAITHFUL.
 ### 11. Missile presentation (salvo visuals + steering gains)
 - **Binary [T1]:** ONE Missile entity per salvo (single spawn call
  @004bcc60 — the pod drew one flying round + one explosion); body-turn
  slerp at authored MaxThrusterRotationRate; velocity slaved to nose while
  burning.
 - **Port [T3]:** N-round visual ripple (lead round carries damage; N−1
  damage-0 tracers), ±2.5° deterministic tube spread, steering gains
  4.0/8.0 port-tuned (the old 'MissileTurnGain=4.0' attribution was the
  proximity fuze), muzzle safety clamp.
 - **Verdict:** KNOWN-DIVERGENCE [T3] (presentation-only; damage economy
  identical).
 ### 12. Missile bystander sweep
 - **Binary [T1]:** FUN_0042291c swept real world geometry including movers
  each tick — a third mech in the flight path could be struck by someone
  else's missile.
 - **Port [T3 radius / port-shape, landed 2026-08-13]:** `mech4.cpp
  BTUpdateProjectiles` — per frame, damage-carrying GUIDED rounds (the one
  salvo lead) run a nearest-approach test of the flight segment against a
  VERTICAL CAPSULE per other registered mech (BTGetTargetCandidates walk;
  shooter + locked target excluded — the target keeps its own 4.0u fuze).
  Capsule derived from the victim's own collision template
  (Mover::GetCollisionTemplate BoxedSolid: axis = origin up minY..maxY,
  radius = larger horizontal half-extent; fallback 3.5u × 14u when the
  template is unresolved [T3]).  Earliest contact along the segment wins
  (bystander vs target fuze compared by segment param); the struck mech
  becomes the DIRECT victim through the unchanged dispatch path (cluster
  roll + splash semantics identical, splash excludes the new direct
  victim).  Wrecks stay in the sweep — binary detonates on any solid; the
  #174 victim-side guard keeps dead zones from cascading/scoring.  Capped
  `[projectile] BYSTANDER` receipt (24 prints) for field forensics.
 - **Verdict:** FIXED-THIS-BUILD (radius is a [T3] capsule approximation of
  the real solid sweep; T2 field receipts pending — Friday watch item
  unchanged: "hit by missiles aimed at someone else" now expected INSIDE
  splash range too).
 ### 13. Missile splash — Explosion::SplashDamage @0042fad0
 - **Binary [T1]:** gated on the FIRING player's advancedDamage (+0x264;
  every stock egg stamps 1 → splash WAS live on the pods), radius 30
  authored on all missiles, baseBurst = FULL rack count (post-roll restore
  @part_013.c:18283-18284, read + confirmed this session), falloff bursts
  = round(base/dist^1.25) floored 1 (1.25 = shipped double; divide
  grounded in T0 EXPLODE.cpp), excludes the explosion + direct victim;
  shooter self-exclusion **unprovable from the export**.
 - **Port [T2 mechanics / T3 gate]:** same math, full-count baseBurst (now
  [T1]-pinned correct), excludes shooter + direct victim; enable =
  SplashRadius>0 instead of the advancedDamage gate — coincides under all
  shipped eggs, diverges only on custom eggs.
 - **Verdict:** FAITHFUL under stock content (gate wiring KNOWN-DIVERGENCE
  [T3] on custom eggs — site option; shooter self-splash OPEN — era
  question shipped).
 ### 14. LRM family flight (lrm model) + loft
 - **Binary [T1]:** lrm rid=319: turn 60°/s, burn 10s @ 300 u/s², climb 50,
  splash 30 — climb>threshold triggers the seeker loft (aim.y +=
  0.1×min(range−200, 300), up to +30u): the signature climb-then-plunge
  arc; soft 30 u/s eject, some mounts +5 up-tilted.
 - **Port [T2]:** #84 fixed the thrust resolution (pre-fix 10× slow WAS the
  'missiles are slow' field report); loft constants exact
  (@004bec18/24/28).
 - **Verdict:** FAITHFUL.
 ### 15. Streak racks (strk model — 'SRM6' bins on at least one chassis)
 - **Binary [T1]:** strk rid=323: turn 360°/s (3× SRM — 6× the lrm), burn
  3s @ 300, climb 0 — the round that out-turns evasion. NO Streak class or
  special lock code exists; the ammo model alone is the identity.
 - **Port [T2]:** same records resolved; 277 u/s impact speed verified
  live.
 - **Verdict:** FAITHFUL.
 ### 16. NRK5 'Narc' + every nrk-fed rack
 Several 'LRM5/10/15' bins author nrk ammo.
 - **Binary [T1]:** no beacon mechanism exists anywhere in the binary
  (single launcher class, no beacon input on the seeker). nrk rid=327
  byte-dumped this session: turn 360°/s, burn 10s @ 300, **CLIMB 50**,
  splash 30 — the shipped NRK5 is a light LRM-profile rack WITH
  Streak-grade tracking AND the loft (missileB's 'no loft' corrected;
  dataA right). Field-visible agility difference between visually
  identical 'LRM' racks is authentic.
 - **Port [T2]:** same data-driven resolution; nothing Narc-special to
  port.
 - **Verdict:** FAITHFUL (KB model-table wording needs the nrk climb/turn
  correction — sweep owed).
 ### 17. Shared reload/ammo/eject chain (0xBCD + 0xBD0)
 - **Binary [T1]:** recoil countdown only while generator route powered,
  heat-scaled (hot generator reloads slower), MinVolt% 0.3 freeze; EJECT
  (@004bb9b8, read this session): TAP = one round out + weaponAlarm→3
  Loading UNCONDITIONALLY if rounds remain — the in-mission UNJAM (clears
  Jammed(5)); HOLD 3.0s = DumpAmmo whole bay → NoAmmo(7) roach motel (only
  re-arm/respawn Reset clears); novice locked out of eject.
 - **Port [T2]:** same machine (`projweap.cpp:751-968`); tap-unjam
  semantics match (the #166 advice rides on it).
 - **Verdict:** FAITHFUL.
 ### 18. Jam roll (ballistic + missile racks)
 - **Binary [T1]:** @004bbfcc: rolled after each shot whenever simLive AND
  the weapon's own heatAlarm != 0; p = 0.41·T/Tfail (0.41 byte-confirmed)
  clamped [0.05, 1.0]. Practically vet/expert-only (std/novice heat model
  off → never heat-alarms).
 - **Port [T3 marked bring-up gate]:** rolls only at heatAlarm >=
  DegradationHeat (`projweap.cpp:692`) — added when temps were static;
  now that the heat economy is live it masks authentic warm-weapon jams.
  Binary rolls at ANY nonzero heat alarm.
 - **Verdict:** KNOWN-DIVERGENCE [T3] (minor; REMOVE the extra gate next
  build — announced in the field guide).
 ### 19. Incoming-missile warning
 - **Binary [T1 registration, T4 consumer]:** Missile ctor registers with a
  MECH target via a virtual on target+0x418; the consumer (threat lamp?
  beeper tempo?) is unlocated in the decomp.
 - **Port [T3]:** structurally dead: the authored beeper/tempo attributes
  latch (`mech4.cpp:8536`) but the only producer lives in the dormant
  Missile TU — the pool never reports; the beeper never fires (#83).
 - **Verdict:** OPEN (needs the consumer decomp + era testimony — question
  shipped in the field guide).
 ---
 ## Contradictions resolved this audit (evidence log)
 1. **GAUSS (the headline):** BTL4.RES @0xf16e6 GAUSS record = classID
   0xBCD (bytes `cd 0b`, size 0x1D0); class histogram 78×0xBC8 / 16×0xBCD
   / 30×0xBD0 / 14×0xBD4, zero 0xBCE; port factory 0xBCD →
   CreateProjectileWeaponSubsystem (`mech.cpp:1939-1942`). The fielded
   Gauss is FAITHFUL on both sides; the earlier "port damage is an
   invention" audit examined the content-dead 0xBCE class. CLASS-level
   claim (@004bdca4 no-op) stands.
 2. **K constant:** `_DAT_004b9c98` dumped: CODE file+0xb9a98 = `0000803f`
   = 1.0f. [T4] closed → [T1]; undamaged effRange = exactly authored
   WeaponRange; port's 1.0 (`mechweap.cpp:558`) correct.
 3. **AC launch-speed constants:** @0x4bc678 = 2.0f (float32); @0x4bc67c =
   0.10204081 = 1/9.8 as a 10-byte x87 EXTENDED (`8a6c379943c6fad0fb3f`) —
   not float32-decodable, hence the stale "unresolved". projweap.cpp 1.0f
   stand-ins + dropped muzzleHeight term now closable.
 4. **Jam recovery:** @004bb9b8 (part_013.c:15709-15723): eject RELEASE
   path calls FeedAmmo and unconditionally SetLevel(weaponAlarm, 3
   Loading) when rounds remain — a tap clears Jammed(5). missileB's "jam
   permanent until re-arm" refuted; in-sortie unjam = YES (novice locked
   out).
 5. **nrk flight model:** BTL4.RES type-15 rid=327: turn 360°/s, burn 10s,
   accel 300, climb 50, splash 30. dataA right; missileB's "no loft"
   corrected. KB sweep owed (combat-damage.md "LRM … the only climber").
 6. **SRM6 count-4 variant:** the record authors MissileCount=4, Damage=20
   → 5.0/missile (dataA's 3.33 assumed count 6). ALL per-chassis
   attributions ride the nearest-name heuristic — re-verify against the
   egg/VehicleTable before any chassis-specific claim ships to the vets
   (the field guide deliberately names no chassis in Appendix A).
 7. **Splash baseBurst:** part_013.c:18271-18290: cluster roll stored to
   +0x338 for direct dispatch, then IF splash enabled the ORIGINAL full
   count is RESTORED (+0x338 = uVar3 at :18284) before SplashDamage — the
   port's full-missileCount baseBurst exactly faithful.
 8. **PPC display-scramble callee:** DAT_004efc94+0x4c = the GAUGE RENDERER
   (`btl4mppr.cpp:249`, `btl4galm.cpp:404`) — the Energy-hit rider is a
   gauge/MFD-side scramble virtual. Still open: what slot +0x4c does
   visually. Port does not implement the call; the generator-short rider
   IS ported — statuses differ.
 9. **SRM2 flight profile — UNRESOLVED:** dataA says the ava1 SRM2 bin
   feeds 'strk'; missileB groups SRM2 under 'srm'. The weapsub scan
   printed weapons only (no bin ammoModelFile column). Needs the bins
   pass before any SRM2 homing claim ships (field guide explicitly
   declines to state it).
 10. **@004bb9b8 relabel:** it is the EJECT handler, not fire-consume;
    FeedAmmo @004bd4f4 is the consume. Carry into any KB address table
    that copied the old tasking label.
 ---
 ## Corrections owed — HELD for the #171 group decision (Oracle HOLD)
 - Commit 80037a7's "Gauss deals NO damage in the shipped binary / port
  damage is an invention" framing.
 - `scratchpad/night16/queued_171_comment.md` — Gauss section + the queued
  era question (the reframed question now lives in the field guide). The
  #112/#168 sections are correct as written.
 - `context/decomp-reference.md` §SS5 and `context/combat-damage.md` Gauss
  claims.
 - Any `projweap.cpp` / gauge banner repeating the no-damage framing.
 ## Stale banners confirmed present (sweep with the next projweap change)
 - `projweap.cpp:990` still reads "the #168 flight-contact model" above the
  implemented hitscan body — same failure genus as the stale banner that
  caused #168 (gotcha: verify banners when the body changes).
 - The "unresolved" comments on `_DAT_004bc678`/`67c` are now false
  (constants byte-confirmed, see entry 7).
 ## Actionable closures queued (next build)
 1. AC tracer launch speed: land 2.0f + 1/9.8 + the muzzleHeight term in
   `projweap.cpp`; delete the stale comments and the :990 banner. Cosmetic;
   era question 14 (round pace) tunes the final feel.
 2. Jam roll: remove the `heatAlarm >= DegradationHeat` gate
   (`projweap.cpp:692`) — binary rolls at any nonzero heat alarm.
   Announced to the vets in the field guide §9.
 3. PPC gauge-scramble rider: BLOCKED on era testimony (field guide §3) —
   build the visual only after the group answers.
 4. Missile bystander sweep: LANDED 2026-08-13 (lane 12 entry) together
   with coast gravity (dt-table row); field guide §6's Friday watch item
   now doubles as the T2 verification channel.
 ## Era questions shipped to the vets
 The 14 questions live in `docs/WEAPONS_FIELD_GUIDE.txt`, distributed into
 the per-section "WE NEED YOUR MEMORY" blocks: Gauss killed + carrier
 chassis (headline, §5); PPC hit visuals + generator short (§3); splash
 proximity + shooter self-splash (§7); ballistic recoil, AFC25 eaten
 trigger, out-of-range round fate, round pace (§4); vet-only jams +
 tap-eject unjam, NoAmmo latch with ammo remaining (§9); gear-wrap brick
 (§2); incoming-missile warning (§10); generator switching mid-charge,
 out-of-range beam (§1); rack-name quirks, homing agility + Narc-did-
 nothing (§8).
 ---
 ## RANGE AUTHENTICITY — printed manual cross-check + the rangefinder scale (2026-08-13)
 **Why:** playtesters report ranges "feel wrong" and hits landing "beyond
 the maximum range on the range scale." Two legs: (1) diff every weapon
 range/damage number the original player manual prints against the
 authored content values (Appendix A of the field guide — this audit's
 numbers table); (2) pin down how the HUD rangefinder's displayed maximum
 relates to authored WeaponRange.
 ### Leg 1 — the manual cross-check
 Source: `reference/manual/Tesla40_BT_manual.pdf` (34 pp). Finding about
 the source itself [T1 — full-text sweep + page renders, this audit]: the
 manual prints **no weapons table**. The per-mech stat sheets (pp. 25-29)
 carry loadouts, coolant loops, tonnage/armor/reservoir/speeds — no
 per-weapon range or damage column. The ONLY printed range figures are
 MFD screenshot mockups (eng-data panels), four values total; **no
 per-weapon damage number appears anywhere in the manual** (the p. 4
 scoring page says damage is delivered "the same way their BattleTech
 boardgame counterparts do" — a fidelity claim with no numbers; the
 authored 4.10 values do NOT numerically track boardgame values, e.g.
 PPC 12 vs the boardgame's 10 — treat the sentence as blurb, not spec).
 | Weapon | MANUAL prints | CONTENT authored | Verdict |
 |---|---|---|---|
 | PPC | RANGE 900M (p. 20, PPC eng-data screen) | 900 | **MATCH** |
 | AFC100 | RANGE 400M (p. 21, AFC eng-data screen) | 400 | **MATCH** |
 | SRM4 | RANGE 450M (p. 22, Engineering MFD mockup) | 800 (display-only) | **MISMATCH** |
 | SRM6 | RANGE 450M (p. 22 AND p. 23, two independent screens) | 800 (display-only) | **MISMATCH** |
 | SLaser | — | 1.5 dmg / 150 | manual-silent |
 | ERSLaser | — | 2 / 225 | manual-silent |
 | MLaser | — | 2.5 / 350 | manual-silent |
 | ERMLaser | — | 3.5 / 500 | manual-silent |
 | LLaser | — | 5 / 600 | manual-silent |
 | ERLLaser | — | 6 / 750 | manual-silent |
 | ERPPC | — | 16 / 950 | manual-silent |
 | AFC25 | — | 7 / 900 | manual-silent |
 | AFC50 | — | 13 / 750 | manual-silent |
 | GAUSS | — | 20 / 900 | manual-silent (pip-color note below) |
 | LRM5/10/15/20, NRK5 | — | 6000 (display-only) | manual-silent |
 | ALL weapons, damage | — (zero numbers printed) | Appendix A | manual-silent |
 All four manual figures verified by page RENDER, not OCR alone (the
 450M reads twice on p. 22 and once on p. 23) [T1].
 **The SRM mismatch read:** 450 (manual) vs 800 (shipped 4.10 content,
 BLH live pip dump [T2] + field-guide Appendix A). Two mitigations:
 (a) the manual's panels are mockups from the earlier software revision —
 the same 4.0→4.10 drift already proven on the Loki loadout rework,
 the small-laser loop redistribution, and supercharge
 (`context/pod-hardware.md` §Manual); (b) for MISSILE racks,
 WeaponRange is fire-control display ONLY (lane 10 [T1]) — real reach is
 flight physics on both values, so the 450→800 change altered the pip
 position and panel text, never a trajectory. Verdict: authoring drift
 between revisions, NOT a port defect; the port displays the shipped
 4.10 content values.
 **Pip-color side note:** manual p. 10 codes the readiness pips PPC=Blue,
 Laser=Red, LRM=Green, SRM=Brown, AFC=Orange, **Gauss=Grey** — blue/red
 match our dumps exactly, SRM "brown" is the authored amber (0.6,0.4,0),
 but the shipped GAUSS record authors a YELLOW pip (lane 8 [T1]). Same
 drift class as the SRM range; cosmetic.
 ### Leg 2 — the rangefinder scale (what the ladder max actually is)
 The tester model behind the complaint — "the scale maxes at the selected
 weapon's range" — is FALSE in both the binary and the port. What the
 HUD right ladder actually is (full detail `context/gauges-hud.md`
 §Cockpit HUD reticle):
 - **The ladder is a FIXED 0–1200 m scale** — the ctor's hardcoded
  calibration (@004cc40c, ctor param 11 = 0x44960000) [T1]; port
  identical (`btl4vid.cpp:2382` kRetMaxRange = 1200.0f). It is never
  derived from any weapon's WeaponRange or live effectiveRange.
 - **Each weapon marks its own reach as a PIP** at its authored
  WeaponRange, CLAMPED into [0..1200] (`AddWeapon` @004cdac0; port
  transcription `btl4vid.cpp:3139-3147`) [T1]. Caret below a pip = that
  weapon reaches the target. An LRM/NRK rack authored 6000 therefore
  pins its pip AT THE LADDER TOP — authentic on both sides.
 - **The caret is the displayed target range**, sliding at 500 m/s toward
  the true pick range (HudSimulation part_013.c:5652 [T1]; port
  mech4.cpp targeting step [T2]); pegs at 1200 with no target. During
  depth-discontinuity walks the displayed value legitimately lags true
  range by hundreds of meters (the Gitea #4 measurement).
 - **The ONE displayed-range divergence:** the binary subtracts
  `_DAT_004b7ecc` = 100.0f from RangeToTarget (@0x1EC) every frame the
  timed flag @0x22C is set (timer @0x21C to limit @0x1D8) — the port
  does NOT implement this state (`hud.cpp:79-82` tracks it;
  `context/open-questions.md` §HUD range-bias) [T1 read, unimplemented].
  Direction: the shipped machine UNDER-reads range by up to 100 m in
  that state; the port never does. So when the state is active a
  true-850 m missile hit against an 800 pip DISPLAYED as in-range on
  the pod but displays as beyond-the-pip in the port — a bounded
  (≤100 m) port-side contributor to "hit beyond the scale" reports.
  Whether the state ever triggers in the field is unknown (what sets
  @0x22C is the open question) [T4 for field prevalence].
 **Rangefinder verdict:** the displayed scale is AUTHENTIC — no port
 miscalibration. "Hits beyond the maximum range on the range scale" is
 expected, authentic behavior with three drivers, largest first:
 (1) missile reach is flight physics, not WeaponRange — SRMs genuinely
 hit past their 800 pip and LRMs past the 1200 ladder top [T1];
 (2) energy/AC pips mark AUTHORED range while the live fire gate is
 effectiveRange = (1 − zoneDamage) × authored — a damaged weapon's real
 reach sits BELOW its pip, never above, so those lanes cannot produce
 beyond-pip hits; (3) the 500 m/s caret slide lags true range during
 fast picks. The single true divergence is the unimplemented −100 m
 bias state (above) — worth reconstructing if Friday's reports cluster
 in the 0–100 m band past a pip; anything farther past the pip is
 missiles being missiles.
 **Guide action:** MISMATCHES exist (SRM display ranges + the Gauss pip
 color) → per the audit rule the field guide was NOT touched; no
 reassurance paragraph shipped.
 ---
 ## VELOCITY/DT INTEGRATION — term-by-term audit (2026-08-13)
 **Scope:** the projectile velocity integration, binary vs port pool
 (`mech4.cpp BTUpdateProjectiles` / `BTPushProjectile`), plus the AC tracer
 launch-speed constants.  Binary side read end-to-end this audit:
 Missile::MoveAndCollide @004bef78, Seeker::FindTarget @004be9a0 /
 LeadTarget @004beae4, **MissileThrusterSimulation @004be474 — the
 standalone Performance body, RECOVERED** (misthrst.cpp's "folded into
 MoveAndCollide, no distinct @ADDR survives" note predated the 2026-08-06
 re-export and is corrected), Missile ctor @004bf5b4, launch composer
@004bcc60, and the engine Mover lane, decomp↔T0-source matched
 line-for-line: **FUN_00422360 = Mover::PerformAndWatch**,
 FUN_00421bac = ApplyWorldAccelerations, FUN_00421e2c =
 ApplyAirResistanceAndGravity(power), FUN_00421ca8 = CalculateDrag,
 FUN_00421b2c/FUN_00421b6c = UpdateWorldMotion/UpdateLocalMotion;
 mover+0x250 = localEnvironment, +0x254 = deadReckoner (MOVER.h/.cpp).
 **The dt model [T1/T0]:** every Performance slice is variable-step REAL
 SECONDS (the #96 sim-time result), and `Mover::PerformAndWatch`
 **ZEROES `localAcceleration` at the top of every frame**
 (@00422360: `FUN_0040a7f4(+0x1dc, &DAT_004e0fd4)` == T0 MOVER.cpp:672
 `localAcceleration = Motion::Identity`).  Thrust, drag and gravity are
 re-added FRESH each frame and then integrated ×dt (@00421bac:
 `p += v·dt + ½a·dt²; v += a·dt`).  Net missile physics is the clean ODE
 `dv/dt = ThrusterAccel − COD·ρ·v²` — **there is NO myomer-style 28 Hz
 per-tick accumulate anywhere in this lane**, so the port's
 `speed += accel·dt` needs no ×(dt·28) rescale.  (The myomer idiom applies
 to PERSISTENT accumulators; the Mover acceleration is not one.)
 **The headline: drag was wrongly waved off.**  All four missile models
 author linear drag CODs **0.001 on every axis** (BTL4.RES type-15 raw
 floats read this audit: srm rid=315 / lrm 319 / strk 323 / nrk 327;
 airDensity = 1.0, engine default).  0.001 multiplies **v²**: it is the
 flight model's GOVERNOR, and the binary says so itself — the Missile
 ctor @004bf5b4 (part_013.c:18416) **precomputes terminal speed
 `sqrt(ThrusterAccel / negCOD.z)` into missile+0x348** and the seeker
 uses it as the lead time base.  Terminal speeds: LRM/Streak/NRK
 **547.7 u/s**, SRM **774.6 u/s**.  The port pool (per #84's "drag
 ~0.001 = negligible") flew undamped: an LRM reached **3030 u/s** by
 burnout — 5.5× the binary cap.  Simulated head-to-head (28 Hz binary
 loop vs the port step): 800u arrival LRM 2.54 s @ 492 u/s (binary) vs
 2.22 s @ 695 (undamped port); at t=5 s the undamped LRM had covered
 3912u @ 1530 u/s vs the binary's 2108u @ 544.  **Fixed this build**:
 the pool integrates `speed += (accel_burn − 0.001·speed²)·dt` for
 thruster rounds, burn AND coast (a burned-out round decelerates,
 `v(t) = v0/(1+0.001·v0·t)` — Streak at t=5 s: 256 u/s, was 1005
 frozen).  AC tracers (p.accel==0) untouched — the binary's visible AC
 round is a renderer cosmetic with no Mover physics.
 | Term | Binary [T1] | Port (pre-audit) | Verdict | Player-visible consequence |
 |---|---|---|---|---|
 | dt model | variable slice, SECONDS; accel accumulator zeroed/frame (@00422360); integrate ×dt (@00421bac) | `speed += accel·dt`, `pos += vel·dt` | **FAITHFUL** (no per-tick accumulate here) | none |
 | Launch velocity | authored MuzzleVelocity (z negated) + shooter velocity (@004bcc60 / FUN_004b9cbc) | same (#67) | **FAITHFUL** | none |
 | Launch acceleration | ZERO in the make message (@004bcc60 authors the DAT_004e0fd4 zero Motion) | thrust applies from frame 1 (binary thruster adds the same frame) | **FAITHFUL** | none |
 | Thrust | @004be474: while burning, `burnLeft −= dt`, `localAccel += (0,0,−ThrusterAccel)` onto the fresh accumulator ⇒ constant accel | `speed += accel·burn_dt` | **FAITHFUL** | none |
 | **Drag** | `−COD·ρ·sign(v)·v²` per body axis, EVERY frame, burn + coast (@004bef78, power-2 inline; plain-Projectile dead class uses power 1.0 @004bddec); CODs 0.001 [T1 bytes]; terminal `sqrt(T/c)` precomputed @004bf5b4→+0x348 | **OMITTED** ("negligible") | **WAS DIVERGENT → FIXED-THIS-BUILD** | was: LRM 3030 u/s (5.5× cap), −13% time-to-800u, +40% impact speeds, map-crossing runaways past 2.5 s; now governed at 547.7/774.6 |
 | Speed clamp | none — drag IS the clamp | none — drag now is | **FAITHFUL** (post-fix) | — |
 | Velocity↔seeker slaving | burning: lateral local vel/accel zeroed each frame ⇒ velocity ≡ speed×nose; nose slerps ≤ MaxThrusterRotationRate·dt (@004be474); at burnout slaving STOPS (ballistic coast) | rotate-toward-aim at port gains 4/8·dt, speed-normalized; steers through coast | KNOWN-DIVERGENCE [T3] (lane 11) | turn dynamics shape; coast homing (real impacts are in-burn) |
 | Gravity | none during burn; coast: `worldAccel.y −= 6.5` fresh each frame (env default) + y<−1 kill plane | absent (loft arc + ttl only) | **WAS DIVERGENT → FIXED 2026-08-13** (`vel.y −= 6.5·coast_dt` on thruster rounds after burnout, exact at the burnout boundary; speed re-synced to |vel|) | post-burnout flights only; missed rounds now drop instead of flying level |
 | Position step | `p += v_old·dt + ½a·dt²` | `p += v_new·dt` (semi-implicit) | FAITHFUL-approx | ≤ ~12u lead over a full SRM burn @60fps; imperceptible |
 | Expiry | burn+10 s, y<−1 (@004bef78 tail) | same (#168) | **FAITHFUL** | none |
 | Seeker lead | `aim += targetVel × range/terminal(+0x348)`; loft climb-gated (thruster+0xF0 > _DAT_004bec1c) | live-position re-lead; loft constants exact | KNOWN-DIVERGENCE [T3] (lane 11) | small lead-shape difference |
 | AC tracer speed | `launchVelocity.z = −effRange/sqrt(2.0·(1/9.8)·muzzleHeight)` @004bc3fc:16157-16164 (bytes @0x4bc678 = `00000040` 2.0f; @0x4bc67c = `8a6c379943c6fad0fb3f` 1/9.8 x87 EXTENDED) | 1.0f stand-ins, no muzzleHeight | **WAS DIVERGENT (cosmetic) → FIXED-THIS-BUILD** | tracer pace now the authentic fall-time scale (~1 s to effRange at a ~5u mount) |
 **Port changes landed (this audit, build clean):**
 1. `mech4.cpp BTUpdateProjectiles` — quadratic drag `0.001·v²` on thruster
   rounds, burn and coast; stale "drag negligible / range-capped" comment
   replaced with the byte-cited model.
 2. `projweap.cpp` — `_DAT_004bc678 = 2.0f`, `_DAT_004bc67c = 1/9.8`
   (bytes above), muzzleHeight term restored via
   `ResolveLaunchVelocity()` (deferred to first use — the port ctor runs
   mid-roster-build; z==0 sentinel; complete-Mech-TU bridge
   `BTWeaponMountHeight` in mech4.cpp).  Stale "unresolved" constant
   comments, the :990 "#168 flight-contact model" banner, and the header
   "FireWeapon BODY NOT recovered" line all swept.
 3. `misthrst.cpp` — banner corrected: @004be474 IS the standalone
   MissileThrusterSimulation (steer + burn + thrust-add), recovered.
 **KB corollary (swept):** the "Mover gravity is dt-less per-frame → may
 need 28 Hz rescale at our frame rate" open question is CLOSED-NO: the
 per-frame gravity/drag adds land on an accumulator that PerformAndWatch
 zeroes every frame — they are per-frame *acceleration definitions*, not
 accumulating increments, and integrate ×dt correctly at ANY frame rate.
 (The port's separate "environment gravity reads 0" gap stays open.)
@@ -1,579 +0,0 @@
 =============================================================================
 BT411 -- WEAPONS FIELD GUIDE: EVERY WEAPON, AS THE ORIGINAL BUILT IT
 Updated 2026-08-13 -- reconstruction review edition, for the whole group.
 =============================================================================
 WHAT THIS IS: we have now read the original machine's firing code for
 every weapon family and compared it, behavior by behavior, against what
 our port does. This document is the result. For each family you get:
 what the shipped machine really did, what our port does today, what
 changes in the next build, and -- where the code alone cannot settle
 it -- what we need from your memory.
 HOW TO REVIEW: you flew the real pods; we only read their code. Read
 each section and mark anything that CONTRADICTS your memory of the
 machine -- and anything that confirms it. Silence is not confirmation:
 if a section simply matches what you remember, a one-word "matches" is
 genuinely useful data.
 HOW TO REPLY: EDIT THIS FILE. Every section that needs your input
 ends with a YOUR NOTES block -- write directly in it (callsign first,
 then MATCHES / CONTRADICTS / CAN'T RECALL, then your detail) and post
 the edited file back when you're done. Which floor, which revision,
 which mech you flew -- machine-room details turn weak memories into
 strong evidence.
 ONE RETRACTION UP FRONT: the last issues handout (#27) told you
 ballistic rounds could now "genuinely MISS a moving target". The
 deeper dig behind this document proved that framing wrong: the
 original decides a cannon hit AT THE TRIGGER, and the round you watch
 fly is a tracer drawn for show. We were wrong; details in the
 AUTOCANNONS section. Your review of that section is part of the
 group's sign-off on the correction.
 =============================================================================
 -----------------------------------------------------------------------------
 1. LASERS -- SMALL / MEDIUM / LARGE, STANDARD AND ER
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - A charge-and-release energy weapon. No ammo, no jam. Six marks, one
    mechanism -- only the numbers differ (Appendix A). No pulse lasers
    exist anywhere in the shipped content.
  - Firing needs three things at the trigger: a LOCKED target, your
    view actually on the target (the weapon only arms while you are
    looking where it points), and the target inside effective range.
  - The beam is instant. The hit is decided the moment you fire -- no
    travel time, no leading. What you see is the discharge drawing
    itself, plus a 0.2-second afterglow that tracks the target.
  - Damage scales with the SQUARE of the stored charge. You cannot
    fire early -- the trigger does nothing until the charge completes.
    The only partial-power shot is a LOWER CHARGE GEAR (section 2,
    veteran/expert only): a gear at half voltage is ready sooner but
    hits for a quarter. Distance never matters: full damage anywhere
    inside effective range, zero damage past it.
  - Out of range: the trigger works and the beam draws, but the damage
    is silently thrown away. No hit, no denied tone we can find.
  - Effective range shrinks in proportion to damage on the body zone
    carrying the weapon: a half-wrecked arm fires at half reach. An
    undamaged weapon reaches exactly its listed range -- confirmed
    down to the byte this week.
  - Overheat is a hard stop: a heat-failed laser quits outright. There
    is no gradual weakening on the way down.
  - Each hit lands on ONE panel of the target.
 OUR PORT TODAY
  Same machine, verified live end to end. The beam is drawn from the
  real laser-tube art. Charge timing is smoothed for modern frame
  rates -- a robustness measure, not a behavior change. (The #164
  stuck-beam guards are already in.)
 CHANGING IN THE NEXT BUILD
  Nothing mechanical.
 WE NEED YOUR MEMORY
  - Firing past range: do you remember shots that visibly fired at a
    distant target and just did nothing? That is what the machine did.
  - Generator switching mid-charge: did switching or re-routing
    generators while a laser was charging COST you the in-flight
    charge (with an electrical alarm)? A yes pins down a control path
    we have not been able to identify from the code alone.
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 2. THE CHARGE GEAR (ALL ENERGY WEAPONS) -- SETTLED BY YOUR MEMORY
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  Veteran/expert pilots could cycle an energy weapon's charge gear
  (four settings). Cycling past the top wraps to the bottom and the
  weapon keeps working: a stored charge above the new gear's level
  simply counts as FULL and fires (a wrap-fired top-gear charge deals
  its normal top-gear damage -- no secret super-shot, no dead weapon).
  FOR THE RECORD: an earlier version of this document claimed the
  machine shipped a flaw here -- that wrapping the gear left the
  weapon permanently dark. Several of you said flatly that never
  happened. You were right. The re-audit you triggered found our
  reading of one number in the original's code was wrong by a single
  byte-width, and the "brick" existed only in our port, never on the
  machine. This section is your correction, adopted.
 OUR PORT TODAY
  Fixed to match: cycling behaves exactly as you remember. (Before
  the fix our gauges could briefly read EMPTY on an over-charge where
  the machine read FULL -- if you saw that, it dies in the next
  build.)
 CHANGING IN THE NEXT BUILD
  The correction above ships. The old "deviation" entry is withdrawn
  -- the energy lane now has ZERO deviations from the original.
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 2B. SEEK-4 ON THE HEAVY WALKERS (VULTURE / MADCAT / THOR / LOKI)
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR (audited to the byte this week)
  - Drive heat scales with weight times speed squared, and no amount
    of running ever cools you -- there is no airflow cooling anywhere
    in this machine. The fast-heavy Clan walkers pay the most.
  - Gears 1 through 3 make IDENTICAL heat: seek-3 is free speed.
    Seek-4 roughly doubles drive heat AND raises your top speed --
    a sprint gear. Cruising it on a heavy walker without the boost
    crosses myomer degradation in under twenty seconds and the
    governor bites. A same-weight humanoid at the same speed cooks the same:
    it is physics, not a chicken-walker curse.
  - The machine gave you the tool: BOOST COOLANT LOOP 5 (Generator D /
    Myomers). One press visibly buys your speed back; a full boost
    sustains seek-4 on a Vulture indefinitely.
  - THE TRAP: one press past maximum WRAPS THE VALVE CLOSED. If you
    ever boosted a loop and it cooked anyway -- you probably hit the
    wrap. Two presses, never three.
  - Standing still sheds heat fast (~10 degrees a second).
 OUR PORT TODAY -- AND A CONFESSION YOUR REPORTS FORCED
  You were right that something was off. Our build measured your
  mech's acceleration at modern frame rates, and at those rates the
  measurement picks up walking-gait jitter as phantom acceleration --
  nearly TWO-THIRDS of your drive heat on a flat cruise was noise
  that the original's electronics (sampling at their own slower
  clock) never saw. The port was punishing heavy walkers roughly
  TWICE as hard as the machine did.
 CHANGING IN THE NEXT BUILD (the fix, measured)
  Heat sensing now samples on the original's clock. Result on the
  bench: a Vulture at seek-4 WITH the loop-5 boost holds FULL
  supercharge speed indefinitely, running hot but stable -- the
  valve trick works exactly as the machine intended, and the
  manual's printed Super Charged speed is real. WITHOUT the boost,
  seek-4 still cooks you (that part is the machine's own balance,
  and it stays). The wrap trap also stays: two presses, never three.
 WE NEED YOUR MEMORY
  - Next build: boost loop 5, hold seek-4 on a big walker. Does the
    sustained supercharge match how the machine felt? Your yes
    closes the oldest heat complaint on the books.
  - Did the loop-5 valve boost trick live in era pilot lore, or are
    we rediscovering it?
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 3. PPC AND ER PPC
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - The firing machine is literally the laser's -- the same code, the
    same rules (lock, look, range gate, square-law charge). Everything
    distinctive is the numbers: hardest energy hit (12; ER 16),
    longest energy reach (900 / 950), a slow 5-second recharge, and
    the only BLUE pip.
  - What makes a PPC special is what it does to the VICTIM. Energy
    hits -- and only energy hits -- carry two riders:
    (a) DISPLAY SCRAMBLE: every energy hit pokes the victim's cockpit
        display system once -- some kind of gauge/MFD disturbance per
        hit. What it looked like and how long it lasted is NOT
        recoverable from the code.
    (b) GENERATOR SHORT: if the struck zone holds critical equipment,
        one random critical component takes a forced short --
        generator brownout, charge stall, cockpit electronics dip.
 OUR PORT TODAY
  The firing machine and the generator short are in and verified. The
  display scramble is NOT implemented -- right now your screens shrug
  off a PPC hit, and that is a known gap on the polish list.
 CHANGING IN THE NEXT BUILD
  Not yet. We will not invent the visual: the scramble gets built
  AFTER your testimony, not before.
 WE NEED YOUR MEMORY (this one really matters)
  - Taking a PPC hit, from inside the pod: what happened on your
    screens? Static or interference? A flash? Gauges and MFDs going
    briefly to garbage? How long did it last?
  - Separately: after taking PPC hits, do you remember the generator
    browning out or your weapons losing charge (the short rider)?
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 4. AUTOCANNONS -- AFC25 / AFC50 / AFC100
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - The cannons are point-and-hit, not fly-and-hit. At the trigger,
    with a lock, the machine checks the range; in range, the damage
    lands THEN -- one solid ballistic hit on ONE panel of the locked
    target. The shell you watch fly is a TRACER, drawn for show after
    the decision is already made.
  - No leading. The lock does the aiming, and a dodging target does
    not shake a round already fired.
  - Out of range: the round is consumed, the tracer flies anyway, and
    the damage is silently discarded. No hit, no warning.
  - Recoil: every shot with real punch (damage over 3) kicks your own
    pod up and back through the gyro, in proportion to the shot --
    the AFC100 kicks about 3.5 times as hard as the AFC25.
  - Jams roll after each shot only while the weapon itself is running
    a heat alarm -- in practice veteran/expert only, since the lower
    sim levels never heat-alarm a weapon. Chance grows with weapon
    temperature.
  - One panel per pull. The old "shotgun across zones" spread in our
    port (#112) was our bug, never the machine's.
 OUR PORT TODAY
  Until now our cannons flew a real shell that could miss --
  reasonable, and wrong. The trigger-time machine is now implemented
  against the original's code line by line and bench-verified: full
  batches land as one hit, out-of-range rounds are consumed and
  discarded. What it needs now is a real night of fire.
 CHANGING IN THE NEXT BUILD
  - The trigger-time cannon goes live in the build you test next.
    This is the big #171 correction -- please hammer the cannons.
  - The tracer's flight speed: the original derives it from the gun's
    range and muzzle height (roughly one second to cross full range).
    Our stand-in speed was close by accident; the recovered authentic
    numbers go in.
 WE NEED YOUR MEMORY
  - When YOU fired cannons, did the pod's view visibly kick up and
    back (distinct from being hit)? Was the AFC100's kick noticeably
    harder than the AFC25's?
  - The AFC25 specifically: did it occasionally eat a trigger pull
    silently -- a click, no shot, on an apparently ready gun? (Its
    ammo feed timer exactly equals its firing cycle -- the only gun
    where the two can race.)
  - Firing beyond range: did the visible round fly on forever, or
    vanish? (The code consumes the round and discards the damage; the
    tracer's visual fate is the one part we cannot recover.)
  - Round pace: roughly how fast did cannon rounds visibly fly?
    About a second to cross full range, or very different?
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 5. GAUSS                                        ** THE HEADLINE QUESTION **
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR (as best we can prove it)
  - The Gauss the pods actually fielded is built ON THE CANNON
    MACHINE. The shipped content authors it as an autocannon-class
    weapon: about 20 points of solid shot out to 900 (tied with the
    AFC25 for longest ballistic reach), a slow 8-second cycle, only
    16 rounds, runs nearly cold, and the only YELLOW ballistic pip.
    It hit, and it hurt.
  - The code ALSO contains a second, "true" Gauss-rifle mechanism
    that charges up and does NOTHING when fired -- no damage, no
    heat, nothing. No shipped content anywhere uses it. Dead code.
  - Full honesty: for a while our audit believed the pods fielded the
    do-nothing Gauss and that our port's Gauss damage was an
    invention. Reading the raw content bytes settled it the other
    way -- the fielded Gauss is the cannon-class one.
 OUR PORT TODAY
  Same content, same machine: our Gauss already deals cannon-class
  damage. Port and original agree. Nothing changes.
 WE NEED YOUR MEMORY (the headline)
  - Confirm the Gauss KILLED: heavy single hits at extreme range, a
    slow cycle, a small magazine, a gun that never warmed up, and a
    yellow pip on the display. We expect YES.
  - Which mech do you remember carrying it? (We believe the Loki;
    your memory checks our content map.)
  - If instead you remember a Gauss that charged up and did NOTHING
    when fired -- say so loudly. That would be evidence that some
    floor revision shipped content for the dead mechanism, content
    we do not have.
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 6. MISSILES -- HOW EVERY RACK REALLY FIRES
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - Missiles are the one family that genuinely FLIES. Everything else
    in the game decides its hit at the trigger; a missile earns its
    hit in the air.
  - Lock required, same as every weapon. There is no dumb-fire path
    anywhere in the machine -- the trigger will not take without a
    lock. (No special Streak lock code exists either; all racks fire
    through the same gate.)
  - ONE flying round was drawn per salvo, plus one explosion -- the
    pod drew a single missile even for a 15-tube volley. The damage
    bookkeeping rides that one round.
  - The seeker LEADS: it flies a constant-bearing intercept toward
    the exact spot on the target picked at the moment of fire, and
    detonates on a close-proximity fuze with a swept contact check.
  - If the target dies mid-flight, the lock drops and the round goes
    stupid.
  - A rack's listed "range" is fire-control display only. Real reach
    is physics: motor burn, then up to 10 seconds of coasting. Rounds
    that hit the floor die; rounds that time out fizzle.
  - On impact, the cluster roll: out of an n-missile salvo, between
    roughly a quarter and all of them count as hits, scattered across
    the victim's zones with a fresh scatter every burst.
 OUR PORT TODAY
  The whole chain is live-verified: a 382-impact night corpus matches
  the original's cluster roll exactly, and the fuze and lock-drop are
  benched. (Flight SPEED is the one thing that corpus measured against
  the old dragless model -- see CHANGING below: speeds drop to the
  original's governed values next build.)
  Presentation deliberately differs: we draw the full ripple of N
  rounds (the lead round carries the damage, the rest are tracers)
  with a small tube spread. The damage economy is identical.
 CHANGING IN THE NEXT BUILD -- AND WHAT TO WATCH
  MISSILES SLOW DOWN, AND THIS IS THE ORIGINAL'S SPEED. The audit
  found the original governs every missile with air drag that caps
  its top speed -- an LRM tops out near 550, an SRM near 775. Our
  rounds had no drag: an LRM was leaving its burn at over 3000,
  five times the machine's speed, and flying nearly twice as far
  downrange. Next build restores the governor. Missiles will feel
  heavier, arcs will read slower, and long shots will take
  noticeably longer to arrive -- that is the pod's real feel, not a
  nerf. If missile flight FINALLY looks like you remember, say so;
  that confirmation is worth as much as a bug report.
  The bystander gap. The original's rounds swept everything in the
  flight path every tick -- a third mech between shooter and target
  could eat someone else's missile. Our rounds currently contact only
  the locked target and the terrain: a mech in between gets flown
  through, and only splash can touch it. This is the largest
  remaining missile-physics gap; the fix is designed but not landed.
  WATCH FRIDAY: if you take missile hits aimed at someone else --
  beyond splash distance -- report it with the shooter's name.
 -----------------------------------------------------------------------------
 7. MISSILE SPLASH
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  Splash was LIVE on the pods -- every stock mission enables it. An
  impact splashes a radius of about 30 around the hit point; strength
  scales from the FULL rack size and falls off steeply with distance;
  the mech that took the direct hit is excluded from its own splash.
  One thing the code cannot prove either way: whether the SHOOTER was
  protected from splashing themselves at point-blank.
 OUR PORT TODAY
  Same math, verified -- including the full-rack-size splash strength,
  pinned to the original's code this week. We exclude both the direct
  victim and the shooter. The on/off wiring differs slightly from the
  original in a way that only matters for custom missions -- a site
  option question for later, not a Friday one.
 WE NEED YOUR MEMORY
  - Do you remember splash hurting mechs standing NEAR an impact?
  - Critically: could you hurt YOURSELF firing missiles point-blank
    into a wall or an adjacent mech? Our port says no; the original's
    code will not tell us. Your memory is the tiebreaker.
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 8. HOMING AGILITY, THE LRM ARC, AND THE NARC
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - Three flight profiles exist in the shipped content:
      LRM round:    sluggish 60 deg/sec turn, 10-second burn, and the
                    LOFT -- at long range the seeker aims high and
                    the round flies the signature climb-then-plunge
                    arc.
      Streak round: 360 deg/sec turn -- six times the LRM -- short
                    3-second burn, no loft. The round that out-turns
                    evasion.
      Narc round:   the surprise -- Streak-grade 360 deg/sec tracking
                    PLUS the loft PLUS the long 10-second burn. A
                    light rack that flies like a hawk.
  - There is NO beacon mechanism anywhere in the machine. The Narc
    never tagged anyone for anything; it is a light, hard-tracking
    rack and nothing more.
  - Several racks LABELED "LRM" actually feed Narc-profile rounds,
    and at least one "SRM6" feeds Streak rounds. Two visually
    identical racks homing very differently -- one sluggish, one
    snapping onto target -- is authentic, not a bug.
  - Rack names lie in other ways too, all authentic authored data:
    an "SRM6" that fires 4 tubes, an "SRM4" that fires 2, LRM10s that
    hit nearly twice as hard on one chassis as another.
 OUR PORT TODAY
  Same data-driven resolution -- the racks read the same content, so
  all of the above carries over for free. One detail we are NOT
  stating yet: which profile the 2-tube SRM racks fly -- our two
  content reads disagree and the pass that settles it has not run.
 CHANGING IN THE NEXT BUILD
  Nothing mechanical.
 WE NEED YOUR MEMORY
  - Do you remember visually identical LRM racks homing very
    differently on different mechs? And can you confirm the Narc did
    nothing special -- no tag, no beacon, just a fast light rack?
  - Do you remember SRM racks that fired FEWER missiles than their
    name -- a 4-tube "SRM6", a 2-tube "SRM4"? Or LRM10s/LRM15s that
    hit noticeably harder on one chassis than another?
  - THE SRM RANGE PIP: the original PRINTED MANUAL shows the SRM
    range mark at 450 in three separate display mockups; the game
    content we have authors it at 800. That smells like the range
    changed between software revisions. Which number do you remember
    on the ladder when you selected an SRM rack -- 450 or 800?
    (Range on a missile is display-only either way -- the rounds fly
    on physics -- but your answer dates our content revision.)
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 9. AMMO, EJECT, AND JAMS (CANNONS AND RACKS)
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  - The reload countdown runs only while the weapon's generator route
    is powered, and a HOT generator reloads slower. Below minimum
    voltage the countdown freezes entirely.
  - The EJECT control (novice pilots are locked out of it):
      TAP  = eject one round, and the weapon resets to loading if
             rounds remain. This is the in-mission UNJAM.
      HOLD (3 seconds) = dump the entire bay. The weapon is out of
             ammo for good until re-arm or respawn.
  - Jams roll after each shot while the weapon itself runs a heat
    alarm; the chance grows with weapon temperature. Standard and
    novice sims never heat-alarm a weapon -- so in practice only
    veteran/expert pilots ever jammed.
 OUR PORT TODAY
  Same machine, tap-unjam included -- it is the mechanism behind
  #166's "tap EJECT" advice. One divergence: our jam roll currently
  waits for a higher weapon temperature than the original -- a
  leftover safety from before the heat economy went live.
 CHANGING IN THE NEXT BUILD
  The leftover jam gate comes out. Warm cannons and racks on
  veteran/expert may jam a touch more often. That is the authentic
  rate.
 WE NEED YOUR MEMORY
  - Did weapons only ever jam for veteran/expert pilots? And could
    you CLEAR a jam mid-sortie by tapping EJECT to cycle one round
    out?
  - Did a rack or cannon ever go permanently dark WITH ammo remaining
    after sustained fire? At what firing intensity? (Field case from
    our own nights: an SRM6 latched out with 19 rounds left. Whether
    that could happen at authentic intensity is an open question.)
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 10. INCOMING-MISSILE WARNING
 -----------------------------------------------------------------------------
 THE ORIGINAL'S BEHAVIOR
  Every missile launched at a mech announces itself to its target --
  the machinery for a warning is there. What CONSUMED that
  announcement (a warning tone? a beeper tempo change? a threat lamp?)
  we have not located in the original's code.
 OUR PORT TODAY
  The beeper plumbing exists but nothing feeds it -- the warning is
  structurally dead (#83 on the known-gaps list).
 WE NEED YOUR MEMORY
  - Did the pod audibly warn you of inbound missiles? A tone, a tempo
    change, a lamp -- anything. If it did not, #83 closes for free.
    If it did, describe it and we build exactly that.
  YOUR NOTES (callsign first, one entry per line):
  >
  >
  >
 -----------------------------------------------------------------------------
 APPENDIX A -- THE AUTHORED NUMBERS (straight from the shipped content)
 -----------------------------------------------------------------------------
 Range is in the same units your fire-control shows. Cycle in seconds.
 ENERGY (damage = full-charge shot; scales with the square of charge)
  WEAPON      DMG     RANGE   CYCLE   HEAT/SHOT
  SLaser      1.5     150     1       0.75
  ERSLaser    2       225     1       1.5
  MLaser      2.5     350     2       2
  ERMLaser    3.5     500     2       3.5
  LLaser      5       600     4       6
  ERLLaser    6       750     4       8.5
  PPC         12      900     5       11
  ERPPC       16      950     5       16
  Reach shrinks with damage to the carrying zone: listed x (1 - zone
  damage). An undamaged weapon reaches exactly the listed range.
 BALLISTIC (one panel, decided at the trigger; recoil kick = dmg/16)
  WEAPON      DMG     RANGE   CYCLE   NOTES
  AFC25       7       900     2       feed timer = cycle (see sec. 4)
  AFC50       13      750     4
  AFC100      25      400     8       hardest hit, shortest reach
  GAUSS       20      900     8       16 rounds, yellow pip, near cold
  Cannons and racks run effectively cold next to the energy bank.
 MISSILE RACKS (DMG = whole-salvo total; per-round in parentheses)
  RACK      TUBES   DMG           CYCLE
  SRM2        2     10   (5.0)    2
  SRM4        4     35   (8.75)   3
  SRM4        4     20   (5.0)    3
  SRM4        2     10   (5.0)    2
  SRM6        6     50   (8.3)    5
  SRM6        6     35   (5.8)    5
  SRM6        4     20   (5.0)    3
  LRM5        5     10   (2.0)    2.5
  NRK5        5     10   (2.0)    2.5
  LRM10      10     35   (3.5)    4
  LRM10      10     20   (2.0)    4
  LRM15      15     50   (3.3)    6
  LRM15      15     35   (2.3)    6
  LRM20      20     65   (3.25)   9
  Same rack name, different numbers = different chassis fits, all
  authentic. We are re-verifying the per-chassis map before naming
  which mech carries which variant. SRM racks display range 800;
  LRM/NRK racks display 6000 -- display only; real reach is flight
  time. Cluster roll on impact: of n launched, between about n/4 and
  n count as hits.
 FLIGHT PROFILES (authored on the round, not the rack)
  LRM round:     turn  60 deg/s   burn 10 s   loft YES
  Streak round:  turn 360 deg/s   burn  3 s   loft no
  Narc round:    turn 360 deg/s   burn 10 s   loft YES
  All rounds splash radius 30 on impact. After burnout a round coasts
  up to 10 seconds before fizzling.
 =============================================================================
 WHEN YOU'RE DONE: post your edited copy of this file back to the
 weapons thread. "MATCHES" alone in a notes block is real data. The
 GAUSS and PPC-HIT questions are the two we cannot answer without you.
 =============================================================================
@@ -1,5 +1,4 @@
 #include <cstdlib>
 #include <map>				// (BT411) the 28Hz smoother-intake clock (cadence census item 3)
 #include "munga.h"
 #pragma hdrstop
@@ -959,54 +958,7 @@ void
   if (control_ID == controlID)
 	{
-      // (BT411 cadence census item 3, gotcha 32) SAMPLE INTAKE AT 28 Hz:
+      audioControlAverage.Add(control_value);
      // the authored olympic windows (N=30/15, fill 0) are SIZED for one
      // sample per POD frame (28 Hz -- AUDWTHR.h:457); the port's watcher
      // poll feeds one per RENDER frame (~59), which halves the smoothing
      // window in wall time and feeds a noisier operand.  Clock ONLY the
      // Add (the smoother's own intake) on a per-instance 28 Hz wall
      // clock; the forward below still runs EVERY poll with the held
      // average, so downstream watcher semantics are untouched.  During a
      // hitch the pod would have taken ~28 samples/s of the stalled
      // value -- catch-up repeats the current sample, capped at the
      // largest authored window (30), heat.cpp #119 pattern.
      // BT_AUD_SMOOTH_HZ=<hz> overrides (=0 restores per-poll intake).
      {
         static Scalar s_hz = -2.0f;
         if (s_hz < -1.0f)
         {
            const char *hz = getenv("BT_AUD_SMOOTH_HZ");
            s_hz = (hz != 0 && *hz != '\0') ? (Scalar)atof(hz) : 28.0f;
         }
         if (s_hz > 0.0f)
         {
            static std::map<const void *, DWORD> s_nextMs;
            const DWORD now  = GetTickCount();
            const DWORD tick = (DWORD)(1000.0f / s_hz);      // 35 ms at 28
            DWORD &next = s_nextMs[this];
            if (next == 0)
            {
               audioControlAverage.Add(control_value);
               next = now + tick;
            }
            else if ((long)(now - next) >= 0)
            {
               int added = 0;
               while ((long)(now - next) >= 0 && ++added <= 30)
               {
                  audioControlAverage.Add(control_value);
                  next += tick;
               }
               if ((long)(now - next) >= 0)
                  next = now + tick;                         // hitch: resync
            }
            // else: off-tick poll -- hold; the average is forwarded below
         }
         else
         {
            audioControlAverage.Add(control_value);          // legacy per-poll
         }
      }
      if (getenv("BT_AUDIO_SPATIAL")) { static int s_sa=0;
         if ((controlID == 100 || controlID == 101) && s_sa++ < 3000)
@@ -1,5 +1,3 @@
 #include <cstdlib>			// (BT411) getenv -- the 28Hz doppler-operand clock (cadence census item 4)
 #include <map>				// (BT411) per-instance sample-and-hold registry
 #include "munga.h"
 #pragma hdrstop
@@ -444,45 +442,6 @@ void
 	);
 	Check(&relative_velocity_temp);
 	//
 	// (BT411 cadence census item 4, gotcha 32) 28 Hz SAMPLE-AND-HOLD on the
 	// doppler velocity operand: worldLinearVelocity is a raw per-render-frame
 	// position derivative (the mech4 producer feeds other consumers and is
 	// NOT touched); at ~59 fps its gait-aliasing noise reaches this divisor
 	// directly and warbles dopplerCents per stride.  The pod computed this
 	// once per 28 Hz frame.  Hold the RELATIVE WORLD velocity per instance
 	// on a 28 Hz wall clock; the head-frame transform below stays live per
 	// call (geometry is smooth -- only the noisy operand is clocked).
 	// BT_AUD_DOPPLER_HZ=<hz> overrides (=0 restores per-call sampling).
 	//
 	int dopplerFresh = -1;			// receipt: -1 legacy / 1 sampled / 0 held
 	{
 		static Scalar s_dopHz = -2.0f;
 		if (s_dopHz < -1.0f)
 		{
 			const char *hz = getenv("BT_AUD_DOPPLER_HZ");
 			s_dopHz = (hz != 0 && *hz != '\0') ? (Scalar)atof(hz) : 28.0f;
 		}
 		if (s_dopHz > 0.0f)
 		{
 			struct DopplerHold { DWORD nextMs; Vector3D held; };
 			static std::map<const void *, DopplerHold> s_dopHold;
 			DopplerHold &h = s_dopHold[this];
 			const DWORD now = GetTickCount();
 			if (h.nextMs == 0 || (long)(now - h.nextMs) >= 0)
 			{
 				h.held   = relative_velocity_temp;
 				h.nextMs = now + (DWORD)(1000.0f / s_dopHz);
 				dopplerFresh = 1;
 			}
 			else
 			{
 				relative_velocity_temp = h.held;
 				dopplerFresh = 0;
 			}
 		}
 	}
 	relative_velocity.MultiplyByInverse(
 		relative_velocity_temp,
 		head_entity->localToWorld
@@ -514,19 +473,6 @@ void
 		{
 			dopplerCents = cents;
 		}
 		// (BT411) doppler receipt -- jitter A/B for the 28Hz operand clock.
 		// hm2 = |held relative WORLD velocity|^2 (the clocked operand itself,
 		// geometry-free); h = 1 fresh sample / 0 held / -1 legacy per-call.
 		if (getenv("BT_AUD_DOPPLER_LOG")) { static int s_dl = 0; if (s_dl++ < 6000)
 			DEBUG_STREAM << "[doppler] t=" << (GetTickCount() % 1000000)
 				<< " this=" << (void *)this
 				<< " cents=" << dopplerCents
 				<< " v=" << speed_of_source
 				<< " d=" << distanceToSource
 				<< " h=" << dopplerFresh
 				<< " hm2=" << relative_velocity_temp.LengthSquared()
 				<< "\n" << std::flush; }
 	}
 	else
 	{
@@ -1313,6 +1313,20 @@ static void
 	int mask = port->GetBitMask();
 	int palId = port->paletteID;
 	int tint = w->monoTint;
 	// OVERLAY COMPOSITE (2026-08-13): the radar's low byte is TWO ports -- sec
 	// (0x3F) + overlay (0xC0: the graticule art, the SCALE numeral and the
 	// SECTOR numerals).  The arcade DAC indexed the full low byte and clut0
 	// already holds the combined 256-entry table (BuildSecondaryPalette spreads
 	// each port over the other's bit combinations), so the palette-expanded
 	// window composites both planes with just the OR'd mask -- previously the
 	// 0x3F mask left SECTOR:/SCALE: blank while their writers were live.
 	if (tint < 0)
 	{
 		L4GraphicsPort *ov =
 			static_cast<L4GraphicsPort*>(gr->GetGraphicsPort("overlay"));
 		if (ov != NULL && ov->graphicsDisplay == port->graphicsDisplay)
 			mask |= ov->GetBitMask();
 	}
 	static int sMfdPal = -1;
 	if (sMfdPal < 0) sMfdPal = getenv("BT_GLASS_MFD_PAL") ? 1 : 0;
 	if (sMfdPal && tint >= 0)
@@ -2110,6 +2124,11 @@ static unsigned long
 		const char *ports[8];
 		int np = 0;
 		ports[np++] = w->portPrimary;
 		if (w->monoTint < 0)
 			ports[np++] = "overlay";	// composited into the palette-expanded
 										// radar (2026-08-13) -- the SECTOR/SCALE
 										// numerals tick on THIS plane, so their
 										// churn must mark the window dirty
 		if (w->portAlt != NULL) ports[np++] = w->portAlt;
 		for (int gi = 0; gi < w->groupCount && np < 7; ++gi)
 		{
@@ -860,6 +860,31 @@ static const float kBTPanelH = 480.0f;
 // channel-enabled and which is BlankColor; the draw loop below skips the
 // blanked plane, so the dev cell shows exactly what the pod monitor shows.
 // The Eng entries share their sibling's cell rect and FOLLOW it in the list.
 //
 // OVERLAY COMPOSITE (2026-08-13).  The secondary CRT's low byte is authored as
 // TWO ports -- sec (bits 0x3F, btspal) and overlay (bits 0xC0, btopal,
 // TransparentZero) carrying the graticule art, the SCALE numeral (RadarRange)
 // and the SECTOR numerals (sectorDisplay).  The arcade DAC indexed the FULL low
 // byte, blending both planes; every port expand used to mask to sec's 0x3F,
 // which is why SECTOR:/SCALE: read blank on desktop while their reconstructed
 // writers were live in the buffer the whole time.  clut0 already holds the
 // DAC's combined 256-entry table (BuildSecondaryPalette spreads each port's
 // colors across the other's bit combinations), so the composite is exactly
 // this mask -- no blending code.
 //
 static int BTSecCompositeMask(GaugeRenderer *gr, L4GraphicsPort *sec_port)
 {
 	int mask = sec_port->GetBitMask();
 	if (gr != NULL)
 	{
 		L4GraphicsPort *ov =
 			static_cast<L4GraphicsPort*>(gr->GetGraphicsPort("overlay"));
 		if (ov != NULL && ov->graphicsDisplay == sec_port->graphicsDisplay)
 			mask |= ov->GetBitMask();
 	}
 	return mask;
 }
 static const BTGaugeSurfaceDesc kBTGaugeSurfaces[9] =
 {
 	{ "Heat", 0xFFFF,    0.0f / kBTPanelW, 0.0f,   320.0f / kBTPanelW, 0.5f, 0 },
@@ -946,7 +971,9 @@ void BTDrawGaugeSurfaces(LPDIRECT3DDEVICE9 device, float px, float py, float pw,
 			rot = s_secRot;
 		}
 		svga->DrawDevSurface(
-			device, i, port->GetBitMask(), port->paletteID, d.monoTint,
+			device, i,
 			d.monoTint < 0 ? BTSecCompositeMask(gr, port) : port->GetBitMask(),
 			port->paletteID, d.monoTint,
 			px + d.cellX * pw, py + d.cellY * ph, d.cellW * pw, d.cellH * ph, rot);
 	}
@@ -1322,7 +1349,9 @@ void BTDrawCockpitPanels(LPDIRECT3DDEVICE9 device)
 				}
 				int surf     = ckSlotOf[i];
 				int useTint  = (d.monoTint < 0) ? -1 : tint;
-				svga->DrawDevSurface(device, i, port->GetBitMask(), port->paletteID, useTint,
+				int useMask  = (d.monoTint < 0) ? BTSecCompositeMask(gr, port)
 				                                : port->GetBitMask();
 				svga->DrawDevSurface(device, i, useMask, port->paletteID, useTint,
 					(float)L.surfX[surf], (float)L.surfY[surf], (float)L.surfW[surf], (float)L.surfH[surf], rot);
 			}
 		}
@@ -5833,7 +5862,7 @@ Logical SVGA16::Update(Logical forceAll)
 		//Drawing secondary, make sure we got secondary
 		if (secPort != NULL)
 		{
-			secMask = secPort->GetBitMask();
+			secMask = BTSecCompositeMask(renderer, secPort);	// sec | overlay (DAC parity)
 			secPalette = &((SVGA16 *) secPort->graphicsDisplay)->palette[secPort->paletteID];
 		} else
 		{
@@ -24,10 +24,7 @@
 //      _DAT_004ba820 = 00 00 80 3f =  1.0f   (full-charge reference / upper clamp)
 //      _DAT_004ba824 = 0a d7 23 3c =  0.01f  (snap-to-1.0 tolerance)
 //      _DAT_004ba828 = 00 00 00 00 =  0.0f   (below-range clamp result)
-//      _DAT_004ba830 = 00 00 00 00 00 00 f0 3f = DOUBLE 1.0  (above-range clamp result;
+//      _DAT_004ba830 = 00 00 00 00 =  0.0f   (above-range clamp result)
 //          #21 CORRECTION 2026-08-13: this operand is a QWORD -- the old
 //          "byte-verified 0.0f" dword-read only the LOW 4 bytes (1.0's mantissa
 //          zeros) of the 8-byte constant.  gotcha 30/31 literal-width family.)
 //      _DAT_004ba9a4 = 00 00 00 00 =  0.0f   (dischargeTimer expiry threshold)
 //      _DAT_004bac04 = 00 00 80 3f =  1.0f   (outputVoltage "fully charged" test)
 //      _DAT_004bac08 = 00 00 00 00 =  0.0f   (firing dischargeTimer expiry threshold)
@@ -501,16 +498,21 @@ void
 		// generator + calibrated voltageScale/EC; TrackSeekVoltage integrates
 		// the charge over the authored RechargeRate seconds).
 		//
-		// POD-FRAME SUB-STEPPING (task #11): run the binary's own Loading
+		// POD-FRAME SUB-STEPPING (task #11, THE WEAPON-BRICK FIX): the binary's
 		// Loading tick assumes the pod's LOCKED 60 fps -- the charge integrates
 		// toward the generator's 10000V and the Loaded transition only fires
 		// while rechargeLevel crosses the +-0.01 snap window around seekV
 		// (level in [7920, 8080]; ~15 pod frames wide).  Under the port's
 		// variable dt, one spike frame (dt >= ~0.25s) jumps the whole window:
 		// the level overshoots, ComputeOutputVoltage's byte-verified >1.0
 		// clamp (_DAT_004ba830 = 0.0) zeroes rechargeLevel, and the weapon is
 		// PERMANENTLY stuck Loading at level ~10000 (observed live:
 		// "level=9999.68 alarm=3 edge=1").  Fix: run the binary's own
 		// tick -- TrackSeekVoltage + ComputeOutputVoltage + the Loaded test --
-		// at the binary's locked 60 fps.  Justification: the I^2R generator
+		// at the binary's own frame rate.  This also keeps the I^2R generator
-		// feed integral -- TrackSeekVoltage's seekRate^2 heat evaluated over
+		// feed integral accurate (big steps evaluate seekRate at a stale level
-		// big variable-dt steps uses a stale level and OVER-heats the
+		// and OVER-heat the generators).  Sub-step count is bounded; leftover
-		// generators.  (The former "weapon-brick fix" rationale was FALSE --
+		// time under extreme throttling just resumes next frame.
 		// see the #21 correction in ComputeOutputVoltage: the >1.0 clamp
 		// stores 1.0f, so an overshot charge goes Loaded next tick at any dt.)
 		// Sub-step count is bounded; leftover time under extreme throttling
 		// just resumes next frame.
 		{
 			static const Scalar kPodFrame = 1.0f / 60.0f;
 			Scalar remaining = time_slice;
@@ -529,13 +531,32 @@ void
 					weaponAlarm.SetLevel(2);			// -> Loaded
 					break;
 				}
-				// The old OVERCHARGE RESCUE (issue #21, 2026-07-21) lived here.
+				// OVERCHARGE RESCUE (issue #21, 2026-07-21) -- a DELIBERATE
-				// REMOVED 2026-08-13: it patched an artifact of the
+				// divergence from the binary, which DEADLOCKS here: change the
-				// _DAT_004ba830 width misread (double 1.0 dword-read as float
+				// seek gear while a charge is in flight and the level can land
-				// 0.0).  With the faithful clamp in ComputeOutputVoltage an
+				// ABOVE the new gear's snap window (level/seekV > 1.01) -- the
-				// overcharged weapon snaps rechargeLevel to exactly 1.0f, so
+				// byte-verified ComputeOutputVoltage clamp then zeroes
-				// the ==1.0f test above takes it to Loaded naturally -- the
+				// rechargeLevel (_DAT_004ba830 = 0.0) and the ==1.0 Loaded test
-				// binary never deadlocked here and neither do we.
+				// can never fire again while TrackSeekVoltage keeps charging
 				// toward the generator voltage: the weapon is PERMANENTLY
 				// bricked (arc dark, dot dark, never fires -- reproduced
 				// headless, and trivial to trigger from the clickable seek
 				// button; the pod's locked 60 fps + rare seek use hid it).
 				// The arcade's own math says "full == the gear's seek voltage"
 				// (discharge energy computes from seekV[rec]), so an
 				// overcharged weapon IS fully charged -- treat it as Loaded.
 				if (currentLevel > seekVoltage[seekVoltageIndex]
 					&& seekVoltage[seekVoltageIndex] > 0.0f)
 				{
 					rechargeLevel = 1.0f;				// full (the display/damage ratio)
 					weaponAlarm.SetLevel(2);			// -> Loaded
 					if (getenv("BT_SEEK_LOG"))
 						DEBUG_STREAM << "[seek] " << GetName()
 							<< " overcharge rescue: level=" << currentLevel
 							<< " > seekV=" << seekVoltage[seekVoltageIndex]
 							<< " -> Loaded" << std::endl;
 					break;
 				}
 			}
 		}
 		break;
@@ -780,17 +801,7 @@ void
 	}
 	else if (rechargeLevel > 1.0f)						// _DAT_004ba820
 	{
-		// #21 CORRECTION (2026-08-13, adversarial re-audit, raw-byte disasm):
+		rechargeLevel = 0.0f;							// _DAT_004ba830 (NB: also 0.0f)
 		// _DAT_004ba830 is a DOUBLE 1.0 (bytes 00 00 00 00 00 00 f0 3f), NOT
 		// float 0.0 -- the old "byte-verified 0.0f" dword-read the LOW HALF of
 		// the qword (1.0's mantissa zeros; gotcha 30/31 literal-width family).
 		// Overcharge clamps to EXACTLY 1.0f (0x3f800000 stored), so the ==1.0f
 		// Loaded test (_DAT_004bac04) passes next tick and the weapon fires.
 		// The machine NEVER bricked on overcharge; era testimony that no weapon
 		// ever went permanently dark triggered the re-audit.  This faithful
 		// clamp supersedes the old "overcharge rescue" divergence (removed
 		// from EmitterSimulation's Loading case).
 		rechargeLevel = 1.0f;							// _DAT_004ba830 (double 1.0, stored as float)
 	}
 }
@@ -53,8 +53,6 @@
 #endif
 #include <JOINT.hpp>				// Joint, JointSubsystem (fwd shim)
 #include <ROTATION.hpp>			// EulerAngles, Radian (fwd shim)
 #include <time.h>					// clock() -- [gtrace] wall-ms stamps (cadence bench)
 #include <map>						// the 28Hz integrator clock (heat.cpp #119 pattern)
 #if !defined(APP_HPP)
 #	include <app.hpp>
 #endif
@@ -480,51 +478,8 @@ void
 	if (swayAngle > maxAnimationNoise)	swayAngle = maxAnimationNoise;
 	if (swayAngle < minAnimationNoise)	swayAngle = minAnimationNoise;
-	// (cadence census item 2, gotcha 32) THE INTEGRATOR CLOCK: the binary
+	IntegrateEyeJoint(time_slice);		// FUN_004b2ec0
-	// steps these two spring-damper integrators once per Perform = once per
+	IntegrateBody(time_slice);			// FUN_004b30ec
 	// pod frame (28 Hz), and the math is PER-TICK (no-dt position step,
 	// damping overwrite carrying last-tick state) -- at the port's ~60 fps
 	// the springs stepped 2.14x as often and the cockpit bounce ran a
 	// foreign timescale.  Step them on a 28 Hz accumulator (heat.cpp #119
 	// pattern: per-instance static map -- the factory size-locks the layout,
 	// no new members), passing the pod tick as the time_slice exactly as
 	// the machine's frame did; remainder carries.  The integrator BODIES
 	// stay byte-exact (including the no-dt position step and the damping
 	// carry -- per-tick semantics, untouched).  Impulses landing between
 	// ticks sit in the force accumulators until the next tick, same as a
 	// between-frame hit on the pod.  Catch-up is capped at 28 ticks (1 s);
 	// a longer hitch drops the excess like the heat clock does.
 	// BT_GYRO_SPRING_HZ=<hz> overrides for bracketing (=0 or negative
 	// restores raw render-cadence stepping).
 	{
 		static Scalar s_springHz = -2.0f;
 		if (s_springHz < -1.0f)
 		{
 			const char *hz = getenv("BT_GYRO_SPRING_HZ");
 			s_springHz = (hz != 0 && *hz != '\0') ? (Scalar)atof(hz) : 28.0f;
 		}
 		if (s_springHz > 0.0f)
 		{
 			static std::map<const void *, Scalar> s_springClock;
 			Scalar &clock = s_springClock[this];
 			clock += time_slice;
 			const Scalar kTick = 1.0f / s_springHz;
 			int catchUp = 0;
 			while (clock >= kTick && ++catchUp <= 28)
 			{
 				clock -= kTick;
 				IntegrateEyeJoint(kTick);	// FUN_004b2ec0, pod-tick slice
 				IntegrateBody(kTick);		// FUN_004b30ec
 			}
 			if (catchUp > 28)
 				clock = 0.0f;				// hitch: drop the excess
 		}
 		else
 		{
 			IntegrateEyeJoint(time_slice);	// FUN_004b2ec0 (render cadence)
 			IntegrateBody(time_slice);		// FUN_004b30ec
 		}
 	}
 	// NOTE (task #56, byte-verified): the binary Performance @004b275c ENDS here.
 	// WriteEyeJoint/WriteMechJoint are NOT called from the gyro -- they are called
@@ -704,8 +659,6 @@ void
 					   + bodyOrientation.z*bodyOrientation.z;
 		if (m2 > 1e-9f)
 			DEBUG_STREAM << "[gtrace] g=" << (void *)this << " f=" << s_traceFrame
 				<< " t=" << (long)clock()		// wall ms (MSVC CLOCKS_PER_SEC=1000) -- the
 												// bounce-timescale bench needs real time, not frames
 				<< " eye=" << (float)eyePosition.x << " " << (float)eyePosition.y
 				<< " " << (float)eyePosition.z
 				<< " body=" << (float)bodyOrientation.x << " " << (float)bodyOrientation.y
@@ -159,7 +159,6 @@
 #include <matchlog.hpp>		// MP match forensics (DEATH/PROJ/SPLASH/RAM event lines)
 #include "btinput.hpp"		// the CONTROLS.MAP + XInput binding engine
 #include <string.h>			// [aud-tail] strchr -- BT_FIRE_PULSE "on,off" parse (Gitea #5, instrumentation only)
 #include <map>				// the ring 28Hz dt-accumulator registry (heat.cpp #119 pattern, census item 5)
 #if !defined(PLAYER_HPP)
 #	include <player.hpp>	// Player::VehicleDeadMessage -- the death->respawn notification (task #52)
 #endif
@@ -818,28 +817,6 @@ void
 		out = m->localOrigin.linearPosition;			// safe non-garbage fallback (owner origin)
 }
 //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 // BTWeaponMountHeight -- the @004bc3fc ctor's muzzle-height operand (velocity/
 // dt integration audit 2026-08-13): the ProjectileWeapon ctor computes
 // launchVelocity.z = -effRange / sqrt((1/9.8) * 2.0 * muzzleTransform.y)
 // (@004bc3fc part_013.c:16157-16164 -- FUN_004b9948 muzzle transform, then
 // FUN_0040a968 extracts the translation and local_5c = its Y).  The binary
 // takes that Y at CTOR time = the bind-pose mount height; the port resolves
 // lazily at first use (ctor runs mid-roster-build), so return the mount's
 // height ABOVE THE MECH ORIGIN -- the same quantity for a standing mech,
 // and terrain-independent.  Complete-Mech-TU bridge per the databinding rule.
 //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 Scalar
 	BTWeaponMountHeight(void *ownerMech, int segIndex)
 {
 	Mech *m = (Mech *)ownerMech;
 	if (m == 0)
 		return 0.0f;
 	Point3D p;
 	BTResolveWeaponMuzzle(ownerMech, segIndex, p);
 	return p.y - m->localOrigin.linearPosition.y;
 }
 // First vital damage-zone index (Mech__DamageZone::vitalDamageZone is protected; Mech has access).
 int
 	Mech::FirstVitalZone() const
@@ -1769,29 +1746,11 @@ static void
 		// autocannon could not miss a moving target.  Track the entity for the
 		// contact test; only GUIDED rounds also steer onto it.
 		extern int BTIsRegisteredMech(Entity *e);
-		if (p.target != 0 && BTIsRegisteredMech(p.target))
+		if (p.target != 0 && BTIsRegisteredMech(p.target)
 			&& !((Mech *)p.target)->IsMechDestroyed())
 		{
-			// #171 LOCK DROP predicate = @0049fb54 EXACTLY: movementMode 2||9,
+			p.targetPos   = ((Mech *)p.target)->localOrigin.linearPosition;
-			// the latched death modes -- NOT IsMechDestroyed(), whose extra
+			p.targetPos.y += p.aimOffsetY;
 			// graphicAlarm>=9 term fires at the vital-kill TRIGGER, seconds
 			// before the collapse (modes 5-8) reaches 2/9.  The binary seeker
 			// keeps TRACKING (and its 4.0u fuze stays live) all through the
 			// fall; it releases only once the wreck latches disabled
 			// (@004bef78: IsKindOf(Mech) && FUN_0049fb54 -> seeker+0xfc = 0).
 			const int mm = ((Mech *)p.target)->MovementMode();
 			if (!(mm == 2 || mm == 9))
 			{
 				p.targetPos   = ((Mech *)p.target)->localOrigin.linearPosition;
 				p.targetPos.y += p.aimOffsetY;
 			}
 			else if (p.guided)
 			{
 				// Seeker released: the round flies on UNGUIDED; only geometry
 				// can detonate it now.  (The old port kept steering to the
 				// corpse's last position and could still sphere-contact it.)
 				p.guided = 0;
 				p.target = 0;					// fuze disarmed with the lock
 			}
 		}
 		if (p.guided)
 		{
@@ -1823,74 +1782,28 @@ static void
 			}
 		}
-		// #84 THRUSTER BURN + QUADRATIC DRAG (velocity/dt integration audit
+		// #84 THRUSTER BURN: the binary Missile hosts a MissileThruster whose
-		// 2026-08-13).  Binary model [T1]: Mover::PerformAndWatch @00422360
+		// authored acceleration drives the round while BurnTime remains
-		// ZEROES localAcceleration EVERY frame (`FUN_0040a7f4(+0x1dc,
+		// (MISTHRST: acceleration = (0,0,-thrusterAccel) in the missile frame,
-		// &DAT_004e0fd4)` = the T0 MOVER.cpp:672 `localAcceleration =
+		// integrated by Missile::MoveAndCollide @4bef78).  The pool flew at
-		// Motion::Identity`), the MissileThruster's own Performance @004be474
+		// CONSTANT |MuzzleVelocity| -- the rack-eject speed (SRM 100, LRM 30
-		// (the standalone MissileThrusterSimulation body -- recovered by the
+		// u/s) -- which is why field missiles crawled (Oracle/Rajel, #84).
-		// 2026-08-06 re-export) re-adds (0,0,-ThrusterAccel) while burning,
+		// Authored (BTL4.RES type-15 missile models, +0x44/+0x48): SRM
-		// and Missile::MoveAndCollide @4bef78 adds the Mover quadratic drag
+		// burn 2.5s @ 600 u/s^2, LRM 10s @ 300, Streak 3s @ 300.  Drag is
-		// -COD*airDensity*sign(v)*v^2 per body axis, then integrates with dt
+		// authored ~0.001 (negligible at combat ranges) -- integrate accel
-		// in SECONDS (@00421bac ApplyWorldAccelerations: v += a*dt).  Net:
+		// along the current heading, un-damped, range-capped as before. [T1
-		//     dv/dt = ThrusterAccel - COD*v^2     (burn; lateral vel zeroed)
+		// values / T2 integration]
-		// -- a clean variable-step ODE.  There is NO myomer-style 28Hz
+		if (p.burnLeft > 0.0f && p.accel > 0.0f)
 		// per-tick accumulate in this lane: `speed += accel*dt` is the
 		// faithful form and needs no (dt*28) rescale.
 		// AUTHORED [T1, raw floats read from BTL4.RES type-15 records
 		// 2026-08-13]: linear CODs = 0.001 on ALL axes of ALL four models
 		// (srm rid=315, lrm 319, strk 323, nrk 327); airDensity = 1.0
 		// (engine default, LATTICE.cpp:413).  Drag is NOT negligible -- it is
 		// the flight model's GOVERNOR: the Missile ctor @004bf5b4 precomputes
 		// terminal speed sqrt(ThrusterAccel/negCOD.z) into missile+0x348 for
 		// the seeker's lead time (part_013.c:18416).  Terminal: LRM/Streak/
 		// NRK 547.7 u/s, SRM 774.6.  The old "drag ~0.001 negligible" note
 		// (#84) let an LRM reach 3030 u/s by burnout -- 5.5x the binary cap
 		// (sim: binary 492 u/s at 800u vs undamped 695; port t+15% to 800u).
 		// Drag runs during burn AND coast (a burned-out round DECELERATES,
 		// v(t)=v0/(1+COD*v0*t)); thruster rounds only -- the AC round is the
 		// 0xBCD renderer tracer, a cosmetic with no Mover physics.  Coast
 		// gravity ported below (audit close 2026-08-13).  Still unported
 		// [T3 minor]: the binary's burnout end to velocity-slaving (lives
 		// beyond practical impact ranges -- see WEAPONS_DRIFT_AUDIT
 		// 'VELOCITY/DT INTEGRATION').
 		if (p.accel > 0.0f)
 		{
 			Scalar burn_dt = (dt < p.burnLeft) ? dt : p.burnLeft;
-			if (burn_dt > 0.0f)
+			p.burnLeft -= burn_dt;
-				p.burnLeft -= burn_dt;
+			Scalar ns = p.speed + p.accel * burn_dt;
 			Scalar ns = p.speed + p.accel * burn_dt
 					  - 0.001f * p.speed * p.speed * dt;	// authored COD x airDensity 1.0 [T1]
 			if (ns < 1.0f) ns = 1.0f;			// guard: drag alone never reverses flight
 			if (p.speed > 0.01f)
 			{
 				Scalar k = ns / p.speed;
 				p.vel.x *= k; p.vel.y *= k; p.vel.z *= k;
 			}
 			p.speed = ns;
 			// COAST GRAVITY (WEAPONS_DRIFT_AUDIT dt-table row closed
 			// 2026-08-13).  The binary applies the environment gravity
 			// (default 6.5, FUN_00421e2c `vy -= g`) ONLY while COASTING --
 			// the @4bef78 else-branch; during burn the thruster re-adds
 			// thrust with NO gravity term onto the frame-zeroed accumulator
 			// [T1, decomp-reference flight-model bullet].  It is a fresh
 			// per-frame ACCELERATION (not an accumulate), so `vy -= 6.5*dt`
 			// is the faithful integration at any frame rate.  coast_dt is
 			// this frame's non-burning remainder (exact at the burnout
 			// boundary).  Applies to EVERY thruster round, lock or no lock
 			// -- it is Mover physics, not seeker logic (a lock-dropped
 			// round coasts ballistic and now DROPS instead of flying level).
 			Scalar coast_dt = dt - burn_dt;
 			if (p.burnLeft <= 0.0f && coast_dt > 0.0f)
 			{
 				p.vel.y -= 6.5f * coast_dt;		// env gravityConstant default [T1]
 				Scalar sv = (Scalar)sqrtf((float)(p.vel.x*p.vel.x
 							+ p.vel.y*p.vel.y + p.vel.z*p.vel.z));
 				if (sv > 0.01f)
 					p.speed = sv;				// keep speed == |vel| (the step
 												// length + normalizations below
 												// divide by p.speed)
 			}
 		}
 		p.pos.x += p.vel.x*dt; p.pos.y += p.vel.y*dt; p.pos.z += p.vel.z*dt;
 		p.age += dt;
@@ -1979,154 +1892,25 @@ static void
 		// detonation being flung past the target at high speed.
 		Point3D hitPos = p.pos;
 		Scalar contactD2;
 		Scalar tTgt = 0.0f;					// target's nearest-approach segment param
 											// (the bystander sweep compares against it)
 		{
 			const Scalar sx = p.pos.x - prev.x, sy = p.pos.y - prev.y, sz = p.pos.z - prev.z;
 			const Scalar seg2 = sx*sx + sy*sy + sz*sz;
 			Scalar t = 0.0f;
 			if (seg2 > 1.0e-6f)
 			{
 				const Scalar wx = p.targetPos.x - prev.x,
 							 wy = p.targetPos.y - prev.y,
 							 wz = p.targetPos.z - prev.z;
-				tTgt = (wx*sx + wy*sy + wz*sz) / seg2;
+				t = (wx*sx + wy*sy + wz*sz) / seg2;
-				if (tTgt < 0.0f) tTgt = 0.0f; else if (tTgt > 1.0f) tTgt = 1.0f;
+				if (t < 0.0f) t = 0.0f; else if (t > 1.0f) t = 1.0f;
 			}
-			hitPos.x = prev.x + tTgt*sx; hitPos.y = prev.y + tTgt*sy; hitPos.z = prev.z + tTgt*sz;
+			hitPos.x = prev.x + t*sx; hitPos.y = prev.y + t*sy; hitPos.z = prev.z + t*sz;
 			const Scalar cx = p.targetPos.x - hitPos.x,
 						 cy = p.targetPos.y - hitPos.y,
 						 cz = p.targetPos.z - hitPos.z;
 			contactD2 = cx*cx + cy*cy + cz*cz;
 		}
-		// BYSTANDER SWEEP (WEAPONS_DRIFT_AUDIT lane 12, closed 2026-08-13).
+		const int contact = (contactD2 < (10.0f*10.0f));
 		// The binary's per-tick contact test is a WORLD sweep (FUN_0042291c,
 		// called from Missile::MoveAndCollide @004bef78): every solid in the
 		// flight path can detonate the round -- including a third mech between
 		// shooter and target [T1].  The port pool tested only the locked
 		// target sphere + the terrain ray, so a mech standing in someone
 		// else's crossfire was flown through (splash alone could touch it).
 		// Port shape [T3 -- radius approximation of the real solid sweep]:
 		// nearest-approach of this frame's flight segment vs a VERTICAL
 		// CAPSULE per other registered mech.  The capsule is derived from the
 		// mech's OWN collision template (Mover::GetCollisionTemplate, the same
 		// BoxedSolid whose maxY is CylinderReferenceHeight): axis = mech
 		// origin up template minY..maxY, radius = the template's larger
 		// horizontal half-extent (the axis ignores the template's near-zero
 		// horizontal center offset, so no yaw rotation is needed).  Fallback
 		// when the template is unresolved: 3.5u x 14u [T3 fixed constants,
 		// fielded-mech scale -- CylinderReferenceHeight ~14].  Wrecks stay in
 		// the sweep: the binary detonates on ANY solid, and the victim-side
 		// zone-state guard (#174) keeps a dead zone from cascading or scoring.
 		// Cheap by construction: damage-carrying GUIDED rounds only (= the one
 		// salvo lead; the N-1 visual tracers are damage-0), per-mech
 		// distance cull, shooter excluded by the registry helper, locked
 		// target excluded (its own 4.0u fuze above is the authentic test).
 		Entity *bysMech = 0;
 		Scalar  bysT    = 2.0f;
 		Point3D bysHit  = p.pos;
 		if (p.guided && p.damage > 0.0f)
 		{
 			extern int BTGetTargetCandidates(Entity *shooter, Entity **out, int maxOut);
 			Entity *cand[32];
 			const int nc = BTGetTargetCandidates(p.shooter, cand, 32);	// excludes the shooter
 			const Scalar sx = p.pos.x - prev.x, sy = p.pos.y - prev.y,
 						 sz = p.pos.z - prev.z;
 			const Scalar seg2 = sx*sx + sy*sy + sz*sz;
 			const Scalar step = p.speed * dt;
 			for (int ci = 0; ci < nc && seg2 > 1.0e-6f; ++ci)
 			{
 				Entity *e = cand[ci];
 				if (e == 0 || e == p.target || !BTIsRegisteredMech(e))
 					continue;
 				Mech *m = (Mech *)e;
 				Point3D mp = m->localOrigin.linearPosition;
 				// distance cull: farther than this frame's step + the largest
 				// plausible capsule reach -> untouchable this frame
 				const Scalar cdx = mp.x - prev.x, cdz = mp.z - prev.z;
 				const Scalar cull = step + 24.0f;
 				if (cdx*cdx + cdz*cdz > cull*cull)
 					continue;
 				Scalar rr, cy0, cy1;
 				BoxedSolid *tmpl = m->GetCollisionTemplate();
 				if (tmpl != 0 && tmpl->maxY > tmpl->minY)
 				{
 					const Scalar hx = 0.5f * (tmpl->maxX - tmpl->minX);
 					const Scalar hz = 0.5f * (tmpl->maxZ - tmpl->minZ);
 					rr  = (hx > hz) ? hx : hz;
 					cy0 = mp.y + tmpl->minY;
 					cy1 = mp.y + tmpl->maxY;
 				}
 				else
 				{
 					rr = 3.5f; cy0 = mp.y; cy1 = mp.y + 14.0f;	// [T3] fixed fallback
 				}
 				if (!(rr > 0.5f)) rr = 0.5f;		// degenerate/garbage template
 				if (rr > 10.0f)   rr = 10.0f;		// guards (NaN falls to 0.5)
 				// closest points between the flight segment prev+t*(sx,sy,sz)
 				// and the vertical axis segment (mp.x, cy0..cy1, mp.z) --
 				// standard segment-segment closest point (RTCD 5.1.9 with
 				// d2 = (0,h,0))
 				const Scalar h   = cy1 - cy0;
 				const Scalar rx  = prev.x - mp.x, ry = prev.y - cy0,
 							 rz  = prev.z - mp.z;
 				const Scalar ee  = h*h;
 				const Scalar f   = h*ry;
 				const Scalar c   = sx*rx + sy*ry + sz*rz;
 				const Scalar b   = sy*h;
 				const Scalar den = seg2*ee - b*b;
 				Scalar t = (den > 1.0e-6f) ? (b*f - c*ee) / den : 0.0f;
 				if (t < 0.0f) t = 0.0f; else if (t > 1.0f) t = 1.0f;
 				Scalar s = (ee > 1.0e-6f) ? (b*t + f) / ee : 0.0f;
 				if (s < 0.0f)      { s = 0.0f; t = -c / seg2; }
 				else if (s > 1.0f) { s = 1.0f; t = (b - c) / seg2; }
 				if (t < 0.0f) t = 0.0f; else if (t > 1.0f) t = 1.0f;
 				const Scalar qx = prev.x + t*sx - mp.x;
 				const Scalar qy = prev.y + t*sy - (cy0 + s*h);
 				const Scalar qz = prev.z + t*sz - mp.z;
 				if (qx*qx + qy*qy + qz*qz < rr*rr && t < bysT)
 				{
 					bysT    = t;
 					bysMech = e;
 					bysHit.x = prev.x + t*sx;
 					bysHit.y = prev.y + t*sy;
 					bysHit.z = prev.z + t*sz;
 				}
 			}
 		}
 		// #171 FUZE RADIUS: a GUIDED damage round detonates at the binary's
 		// proximity fuze _DAT_004bf5a4 = 4.0 (seeker rangeToTarget +0x10C <
 		// 4.0, @004bef78 -- the SAME constant the old steering comment
 		// misattributed as "MissileTurnGain"; the 4.0/8.0 turn rates above are
 		// PORT TUNING [T3], not that constant).  Cosmetic rounds (damage 0:
 		// AC tracers, cluster visuals) keep the wider 10u so their burst
 		// visuals still read at speed.
 		const Scalar fuzeR = (p.guided && p.damage > 0.0f) ? 4.0f : 10.0f;
 		int contact = (contactD2 < fuzeR * fuzeR);
 		Entity *victim = p.target;
 		if (bysMech != 0 && (!contact || bysT < tTgt))
 		{
 			// a third mech sits EARLIER on this frame's flight segment than
 			// the locked target's fuze point -- it takes the round (the
 			// binary's sweep detonates on the FIRST solid in the path).  The
 			// struck mech becomes the DIRECT victim; the salvo cluster roll
 			// and the splash package below run unchanged against it.
 			contact = 1;
 			victim  = bysMech;
 			hitPos  = bysHit;
 			static int s_bysPrints = 0;		// capped receipt (field forensics)
 			if (s_bysPrints < 24)
 			{
 				++s_bysPrints;
 				DEBUG_STREAM << "[projectile] BYSTANDER id="
 					<< (int)bysMech->GetEntityID()
 					<< " t=" << bysT
 					<< " at(" << bysHit.x << "," << bysHit.y << "," << bysHit.z
 					<< ") dmg=" << p.damage
 					<< " lockedTgt=" << (void *)p.target
 					<< ((s_bysPrints == 24) ? " (receipt cap reached)" : "")
 					<< "\n" << std::flush;
 			}
 		}
 		if (!contact && (p.age >= p.ttl || (p.guided && p.pos.y < -1.0f)))
 		{
 			if (getenv("BT_PROJ_LOG"))
@@ -2151,13 +1935,10 @@ static void
 					<< "," << p.targetPos.z << ")" << std::endl;
 			if (p.salvoLead)					// ONE Explosion per salvo (@004bcc60)
 				BTSpawnRoundDetonation(p.shooter, p.weaponSubsys, hitPos);
-			Entity *tgt = victim;
+			Entity *tgt = p.target;
-			// Deliver to the struck mech -- normally the launcher's locked
+			// Deliver to the projectile's target mech -- the launcher set p.target
-			// target (the shooter's 0x388 slot; any peer mech in MP, task #46),
+			// from the shooter's 0x388 slot (the picked victim; any peer mech in
-			// or the BYSTANDER the sweep above found first on the flight
+			// MP, task #46).  A replicant target reroutes cross-pod via Dispatch.
 			// segment (lane 12: it becomes the direct victim, exactly as if it
 			// had been the lock).  A replicant victim reroutes cross-pod via
 			// Dispatch.
 			extern int BTIsRegisteredMech(Entity *e);
 			if (tgt != 0 && BTIsRegisteredMech(tgt) && p.damage > 0.0f)
 			{
@@ -2197,22 +1978,9 @@ static void
 					int bursts = (p.splashBurst > 0) ? p.splashBurst : 1;
 					if (bursts > 1)
 					{
 						// #171: the binary roll (part_013.c:18272-18281) has NO
 						// floor-of-1 -- rolled = min(Random(n) + n/4, n), and
 						// Random(n) is 0..n-1, so n<4 can roll ZERO.  The
 						// VICTIM side is what floors it: the binary burst loop
 						// is a DO-WHILE (part_012.c:14660-14685 -- local_2c =
 						// burstCount, apply, decrement, `if (local_2c < 1)
 						// break` AFTER the body), so a 0-burst message still
 						// applies ONE burst; mech.cpp:1229's `burstsLeft < 1
 						// -> 1` guard is the faithful mirror of that shape.
 						// The old sender-side floor made the end-to-end count
 						// identical, but shipped a message field the binary
 						// never sends (burstCount 1 where the binary says 0);
 						// roll it exactly as the binary does and let the
 						// receiver's do-while supply the floor.
 						int rolled = Random(bursts) + (bursts >> 2);
 						if (rolled > bursts) rolled = bursts;
 						if (rolled < 1)      rolled = 1;
 						bursts = rolled;
 					}
 					dmg.burstCount   = bursts;
@@ -7836,98 +7604,6 @@ void
 			gBTValveKey = (s_vtFrame >= 600 && s_vtFrame < 610) ? 1 : 0;
 		}
 		// #173 decisive experiment (BT_VALVE5=N): press Condenser5's MoveValve
 		// exactly N times, spaced BT_VALVE5_SPACING frames (default 30) apart,
 		// starting at frame BT_VALVE5_AT (default 300) -- walks the detent
 		// cycle 1->5->50->0 (@4ae464) a CONTROLLED number of steps on the
 		// MYOMER loop.  Dispatches the same ReceiverDataMessageOf<ControlsButton>
 		// press payload the pod's engineering-screen aux button delivers; the
 		// handler's unconditional [valve] receipt proves each landed detent.
 		// Self-contained (does NOT ride gBTValveKey, so it cannot collide with
 		// the BT_VALVE_TEST edge detector).  Env-gated, off by default.
 		if ((Entity *)this == application->GetViewpointEntity()
 			&& getenv("BT_VALVE5"))
 		{
 			static int s_v5Frame = 0;
 			static int s_v5Sent  = 0;
 			++s_v5Frame;
 			int want = atoi(getenv("BT_VALVE5"));
 			const char *atEnv = getenv("BT_VALVE5_AT");
 			const char *spEnv = getenv("BT_VALVE5_SPACING");
 			int startAt = (atEnv != 0) ? atoi(atEnv) : 300;
 			int spacing = (spEnv != 0) ? atoi(spEnv) : 30;
 			if (spacing < 2) spacing = 2;					// edge needs a gap
 			if (s_v5Sent < want
 				&& s_v5Frame >= startAt + s_v5Sent * spacing)
 			{
 				Subsystem *condenser = 0;
 				for (int s = 1; s < GetSubsystemCount(); ++s)
 				{
 					Subsystem *sub = GetSubsystem(s);
 					if (sub != 0 && (int)sub->GetClassID() == 0xBBD	// Condenser
 						&& sub->GetName() != 0
 						&& stricmp(sub->GetName(), "Condenser5") == 0)
 					{
 						condenser = sub;
 						break;
 					}
 				}
 				if (condenser != 0)
 				{
 					++s_v5Sent;
 					DEBUG_STREAM << "[valve-tx] BT_VALVE5 press " << s_v5Sent
 						<< "/" << want << " -> " << condenser->GetName()
 						<< " (frame " << s_v5Frame << ")\n" << std::flush;
 					ReceiverDataMessageOf<ControlsButton> msg(
 						4 /*Condenser::MoveValveMessageID*/,
 						sizeof(ReceiverDataMessageOf<ControlsButton>),
 						(ControlsButton)1 /*press*/);
 					condenser->Dispatch(&msg);
 				}
 			}
 		}
 		// GYRO IMPULSE BENCH (cadence census item 2, night17):
 		// BT_GYRO_KICK=<frame>[,<amt>] injects EXACTLY ONE deterministic hit
 		// into the gyro's authentic damage fan-out (GyroApplyDamage ->
 		// ApplyDamageResponse @004b2980) at the given master-perf frame
 		// (default 900) -- a FIXED damageForce so the direction never rides
 		// the random fallback, ballistic type, amt default 25.  Pure gyro
 		// impulse (no zone damage, no armor change): the [gtrace] receipts
 		// then record the eye/body bounce trajectory for the 28-vs-render
 		// spring-timescale comparison.  Env-gated, off by default.
 		if ((Entity *)this == application->GetViewpointEntity()
 			&& getenv("BT_GYRO_KICK"))
 		{
 			static int s_gkFrame = 0;
 			static int s_gkDone  = 0;
 			++s_gkFrame;
 			int   gkAt  = 900;
 			float gkAmt = 25.0f;
 			{
 				char spec[64];
 				strncpy(spec, getenv("BT_GYRO_KICK"), sizeof(spec) - 1);
 				spec[sizeof(spec) - 1] = 0;
 				char *comma = strchr(spec, ',');
 				if (comma != 0) { *comma = 0; gkAmt = (float)atof(comma + 1); }
 				if (spec[0] && atoi(spec) > 0) gkAt = atoi(spec);
 			}
 			if (!s_gkDone && s_gkFrame >= gkAt)
 			{
 				s_gkDone = 1;
 				Damage dmg;
 				dmg.damageType   = Damage::BallisticDamageType;
 				dmg.damageAmount = gkAmt;
 				dmg.burstCount   = 1;
 				dmg.damageForce  = Vector3D(1.0f, 0.0f, 0.3f);	// fixed => deterministic dir
 				dmg.impactPoint  = localOrigin.linearPosition;
 				extern void GyroApplyDamage(Subsystem *, const Damage &);
 				GyroApplyDamage(gyroSubsystem, dmg);
 				DEBUG_STREAM << "[gyro-kick] frame=" << s_gkFrame
 					<< " amt=" << gkAmt << " (one-shot)\n" << std::flush;
 			}
 		}
 		// Gitea #6 scripted verify (BT_VIEWCYCLE_TEST=<frame>): pulse one
 		// secondary-schematic cycle (Damage -> Critical -> Heat -> Damage) at
 		// the given frame and every 300 frames after -- with
@@ -8788,60 +8464,15 @@ void
 	// footstep smoother fed (fed per-stride only, it took 10-20 s to warm
 	// from fill=0: the late-footsteps bug).
 	{
 		// #173 RING CADENCE -- DEFAULT 28 Hz (promoted 2026-08-13, user call,
 		// six-curve bench m173v2): push velocity-ring samples at the pod's
 		// BOARD cadence (28 Hz nominal [T1]) instead of once per render
 		// frame.  The ring / mean-derivative STRUCTURE is untouched
 		// (byte-faithful); the only change is the sampling clock, which on
 		// the pod WAS the frame clock (28 Hz) and here is ~60.  MEASURED
 		// [T2]: at ~59fps the |a| operand carries 64-66% ALIASING NOISE
 		// (|a| 17-26 on flat straight grass); at 28 Hz it drops to 24-32%
 		// and drive-heat generation HALVES at equal speed -- the port was
 		// over-punishing every fast-heavy chassis ~2x.  With the authored
 		// loop-5 boost, 28 Hz sustains FULL supercharge speed indefinitely
 		// at a stable Tm ~1500 (leg G: 54-57 u/s held) -- the machine's
 		// valve lever worked, and the manual's printed Super Charged stat
 		// was honest.  BT_MYO_RING_HZ=<hz> overrides for bracketing
 		// (=0 or negative restores raw render-cadence sampling).
 		int ringPush = 1;
 		Scalar ringDt = dt;
 		{
 			static Scalar s_ringHz = -2.0f;
 			if (s_ringHz < -1.0f)
 			{
 				const char *rh = getenv("BT_MYO_RING_HZ");
 				s_ringHz = (rh != 0 && *rh != '\0') ? (Scalar)atof(rh) : 28.0f;
 			}
 			if (s_ringHz > 0.0f)
 			{
 				// per-instance dt accumulator -- the heat.cpp #119 static-map
 				// pattern (unbounded, keyed per instance, no layout growth).
 				// Replaces the bench-era 16-slot fixed registry, whose silent
 				// overflow would have reverted a 17th mech to render-cadence
 				// sampling (cadence census item 5, night17).
 				static std::map<const void *, Scalar> s_ringClock;
 				Scalar &acc = s_ringClock[this];
 				acc += dt;
 				if (acc < 1.0f / s_ringHz)
 					ringPush = 0;
 				else
 				{
 					ringDt = acc;
 					acc = 0.0f;
 				}
 			}
 		}
 		if (ringPush)
 		{
 		Scalar rdx = localOrigin.linearPosition.x - accelPrevPos.x;
 		Scalar rdy = localOrigin.linearPosition.y - accelPrevPos.y;
 		Scalar rdz = localOrigin.linearPosition.z - accelPrevPos.z;
 		accelPrevPos = localOrigin.linearPosition;
-		if (ringDt > 1.0e-4f)
+		if (dt > 1.0e-4f)
 		{
-			velRingFwd[velRingCursor]  = (Scalar)sqrtf(rdx * rdx + rdz * rdz) / ringDt;
+			velRingFwd[velRingCursor]  = (Scalar)sqrtf(rdx * rdx + rdz * rdz) / dt;
-			velRingVert[velRingCursor] = rdy / ringDt;
+			velRingVert[velRingCursor] = rdy / dt;
-			velRingDt[velRingCursor]   = ringDt;
+			velRingDt[velRingCursor]   = dt;
 			velRingCursor = (velRingCursor + 1) % 15;
 		}
 		Scalar fwdMean = 0.0f, vertMean = 0.0f, dtMean = 0.0f;
@@ -8861,7 +8492,6 @@ void
 				-((fwdMean - accelPrevFwdMean) / dtMean));
 		accelPrevFwdMean  = fwdMean;
 		accelPrevVertMean = vertMean;
 		}
 	}
 	// (AUDIO_FIDELITY F7) latch the incoming-missile report accumulated by
@@ -14,21 +14,11 @@
 //
 // Coverage:
 //   confident : ctor @004be7c4, dtor thunk @004be8bc, TestInstance @004be8e8,
-//               CreateStreamedSubsystem @004bf8ec,
+//               CreateStreamedSubsystem @004bf8ec
-//               MissileThrusterSimulation @004be474 -- the STANDALONE
+//   best-effort: MissileThrusterSimulation -- the Performance method pointer is
-//               Performance body, RECOVERED by the 2026-08-06 re-export
+//               installed by the ctor (PTR_LAB_00512b20) but its standalone body
-//               (the old "folded into Missile::MoveAndCollide, no distinct
+//               is folded into the host Missile::MoveAndCollide @004bef78 (which
-//               @ADDR survives" note here was stale).  @004be474 does BOTH
+//               samples the thruster acceleration buffer at missile+0x234/+0x250).
 //               jobs each frame: (1) slerp the owner Missile's orientation
 //               toward the Seeker aim point (+0xE4), rate-capped at
 //               maxThrusterRotationRate*dt; (2) while ThrusterBurning:
 //               burnTimeRemaining -= dt and owner->localAcceleration +=
 //               this->acceleration (0,0,-thrusterAccel) -- onto the
 //               accumulator Mover::PerformAndWatch @00422360 zeroes every
 //               frame, i.e. a constant acceleration, dt-correct.  The
 //               simplified body below remains the dormant-TU stand-in;
 //               the LIVE port path is the mech4.cpp pool (see
 //               docs/WEAPONS_DRIFT_AUDIT.md 'VELOCITY/DT INTEGRATION').
 //   excluded   : the 0x41xxxx engine vtable slots (pure Subsystem base behaviour;
 //               vtable @00512bbc overrides only slot0 = destructor)
 //
@@ -28,14 +28,11 @@
 //       ProjectileWeaponSimulation @004bbd04 (Performance -- FULLY RECOVERED
 //       by capstone disasm, Gitea #12 2026-07-19; see context/decomp-reference.md
 //       s5 -- the old RivetGun-modeled body is retired).
-//       FireWeapon @004bc104 (slot 18) -- RECOVERED by the 2026-08-06
+//   best-effort (vtable slot proven, function prologue present, BODY NOT
-//       re-export (part_013.c:16023-16125; trigger-time hitscan, implemented
+//   recovered by the decompiler -- bodies below are reconstructed from context
-//       in the #171 batch.  The old "BODY NOT recovered" line here was the
+//   and clearly marked):
-//       stale banner behind the #168 mis-fix -- see the FireWeapon block).
+//       FireWeapon @004bc104 (slot 18), GetStatusFlags @004bbf88 (slot 12),
-//   best-effort (vtable slot proven, function prologue present, body still
+//       UpdateWeaponState @004bbc20 (slot 16).
 //   unrecovered -- reconstructed from context and clearly marked):
 //       GetStatusFlags @004bbf88 (slot 12), UpdateWeaponState @004bbc20
 //       (slot 16).
 //   excluded (belong to sibling/derived classes, NOT ProjectileWeapon):
 //       Emitter family @004bb120/@004bb888/@004ba4d0-@004bb478 (energy weapons),
 //       MissileLauncher @004bcff0/@004bd060/@004bd08c & FireWeapon @004bcc60,
@@ -48,9 +45,6 @@
 //      _DAT_004bc068 = 0000803f =  1.0f    (jam-chance clamp ceiling)
 //      _DAT_004bb3b0 = 000080bf = -1.0f    (resource "unset" sentinel)
 //      _DAT_004bb3b4 = 0000003f =  0.5f
 //      _DAT_004bc678 = 00000040 =  2.0f    (tracer fall-time factor; audit 2026-08-13)
 //      _DAT_004bc67c = 8a6c379943c6fad0fb3f = 1/9.8 as 80-bit x87 EXTENDED
 //                                          (tracer gravity term; audit 2026-08-13)
 //      DAT_004e0f74  = {0,0,0}             (zero vector)
 //      DAT_004e0fd4  = {0,0,0}             (zero point)
 //      0x40400000    =  3.0f               (TotalTimeToEject default)
@@ -151,18 +145,8 @@ namespace {
 static const Scalar	_DAT_004bc100	= 1.0f;		// time-of-flight bias
 static const Scalar	_DAT_004bc064	= 0.41f;	// heat -> jam-chance coefficient
 static const Scalar	_DAT_004bc068	= 1.0f;		// jam-chance clamp ceiling
-// Tracer launch-speed constants -- BYTE-CONFIRMED by the 2026-08-13 weapons
+static const Scalar	_DAT_004bc678	= 1.0f;		// muzzle-speed term (unresolved)
-// audit (WEAPONS_DRIFT_AUDIT.md entry 7 / evidence log 3); the old 1.0f
+static const Scalar	_DAT_004bc67c	= 1.0f;		// muzzle-speed term (unresolved)
 // "unresolved" stand-ins are RETIRED:
 //   @0x4bc678 = 40000000            = 2.0f  (float32)
 //   @0x4bc67c = 8a6c379943c6fad0fb3f = 0.10204081... = 1/9.8, stored as an
 //               80-bit x87 EXTENDED (not float32-decodable -- why it sat
 //               "unresolved").  Together with the ctor's muzzle-height term:
 //   launchVelocity.z = -effRange / sqrt((1/9.8) * 2.0 * muzzleHeight)
 // i.e. the tracer crosses effectiveRange in the time a shell would fall
 // from muzzle height under g=9.8 (~1s for a ~5u mount).
 static const Scalar	_DAT_004bc678	= 2.0f;			// fall-time factor (2h/g)
 static const Scalar	_DAT_004bc67c	= 0.10204081f;	// 1/9.8 (x87 extended in the image)
 // FUN_00408050 -- uniform [0,1) random (jam roll).  The old `return 0.0f` stub ALWAYS
 // jammed (0 < any positive jam chance), so a projectile weapon could never fire.  A real
@@ -327,16 +311,11 @@ ProjectileWeapon::ProjectileWeapon(
 	minVoltagePercentToFire	= subsystem_resource->minVoltagePercentToFire;	// +0x1C8 -> 0x404
 	tracerCounter			= 0;					// 0x408
-	// launchVelocity seed = (0,0,0) then z = -(effectiveRange / muzzleSpeed),
+	// launchVelocity seed = (0,0,0) then z = -(effectiveRange / muzzleSpeed)
 	// muzzleSpeed = sqrt((1/9.8) * 2.0 * muzzleHeight) -- @004bc3fc:16157-16164
 	// resolves the muzzle transform (FUN_004b9948) INSIDE the ctor and uses its
 	// translation Y.  The port DEFERS the z term to first use
 	// (ResolveLaunchVelocity below): our ctor runs mid-roster-build, where the
 	// segment resolve can fall back to the mech origin and would bake a garbage
 	// constant.  z == 0 is the unresolved sentinel (the binary's z is always
 	// negative after its ctor).
 	launchVelocity			= Vector3D(0,0,0);		// FUN_00408440(this+0x410, DAT_004e0f74)
 	Scalar muzzleSpeed		= (Scalar)Sqrt(_DAT_004bc67c * _DAT_004bc678);	// FUN_004dd138
 	tracerOrigin			= Vector3D(0,0,0);		// FUN_00408440(this+0x42c, DAT_004e0f74)
 	launchVelocity.z		= -(effectiveRange / muzzleSpeed);	// this[0x106] = -(this[0xca]/speed)
 	totalTimeToEject		= 3.0f;					// 0x3F0  (0x40400000)
 	timeToEject				= totalTimeToEject;		// 0x3F4
@@ -727,30 +706,6 @@ Logical
 	return (p > UniformRandom()) ? True : False;				// FUN_00408050
 }
 //
 // The @004bc3fc ctor launch-speed expression, deferred to first use (see the
 // ctor note).  Byte-grounded constants (audit 2026-08-13): @0x4bc678 = 2.0f,
 // @0x4bc67c = 1/9.8 (80-bit x87 extended `8a6c379943c6fad0fb3f`):
 //     launchVelocity.z = -effectiveRange / sqrt((1/9.8) * 2.0 * muzzleHeight)
 // The binary computes it once at ctor time (bind pose, effectiveRange still ==
 // authored WeaponRange); the port computes it once at first use with the mount
 // height above the mech origin -- the same quantity for a standing mech.
 //
 void
 	ProjectileWeapon::ResolveLaunchVelocity()
 {
 	if (launchVelocity.z != 0.0f)
 		return;													// already resolved
 	extern Scalar BTWeaponMountHeight(void *ownerMech, int segIndex);
 	Scalar muzzleHeight = BTWeaponMountHeight(owner, GetSegmentIndex());
 	if (muzzleHeight < 0.5f)
 		muzzleHeight = 0.5f;		// port guard: unresolved segment (muzzle==origin
 									// fallback) -- the binary's bind-pose height is
 									// always positive; keeps sqrt/divide sane
 	Scalar muzzleSpeed = (Scalar)Sqrt(_DAT_004bc67c * _DAT_004bc678 * muzzleHeight);	// FUN_004dd138
 	launchVelocity.z   = -(effectiveRange / muzzleSpeed);		// this[0x106] = -(this[0xca]/speed)
 }
 //
 // @004bc06c -- lead-solution time-of-flight.  Sample the current target
 // position, advance it by launchVelocity, and re-run targeting.  With no target
@@ -760,7 +715,6 @@ void
 Scalar
 	ProjectileWeapon::ComputeTimeOfFlight()
 {
 	ResolveLaunchVelocity();									// deferred @004bc3fc z term
 	GetTargetPosition(leadPosition);							// FUN_004b9cbc(this, this+0x420)
 	leadPosition.x += launchVelocity.x;							// this+0x420 += this+0x410
 	leadPosition.y += launchVelocity.y;
@@ -1015,27 +969,10 @@ void
 //
 // @004bc104 -- slot 18 FireWeapon (overrides MechWeapon's pure-virtual trap
-// @004ba45c).
+// @004ba45c).  Body NOT recovered by the decompiler (prologue confirmed at
-//
+// 0x4bc104, immediately following the 1.0f literal @0x4bc100).  Spawns the
-// STALE-BANNER CORRECTION (2026-08-13, the #171 hunt): this comment said
+// generic projectile/tracer.  MissileLauncher overrides this again at @004bcc60
-// "Body NOT recovered by the decompiler" -- TRUE of the 2025 export, FALSE
+// to launch Missile entities.
 // since the 2026-08-06 re-export (part_013.c:16023-16125), and the stale
 // banner is the ROOT CAUSE of the #168 mis-fix: the ballistic model was
 // generalized from the MISSILE launcher @004bcc60 instead of this body.
 // The RECOVERED truth [T1]: the AC FireWeapon is TRIGGER-TIME HITSCAN --
 // it spawns NOTHING (no shell entity; class 0xBD1 is dead code; the visible
 // round is the 0xBCD renderer tracer).  It batches the round counter
 // (+0x408) and every TracerInterval-th round (+0x438) computes
 // dist = |mech aimPoint(+0x37c) - muzzle| and, gated dist <= LIVE
 // effectiveRange (+0x328, the same gate shape as the Emitter), sends ONE
 // single-panel Damage (amount = batchedCount x DamageAmount) to the LOCKED
 // target *(mech+0x388) via SendDamageMessage @004b9728 -- at fire time.
 // A locked AC inside range cannot miss; out of range the batch silently
 // discards (tracer still flies).  The body below IMPLEMENTS that hitscan
 // (landed in the #171 batch, bbb3b4e; the earlier "still carries the #168
 // flight-contact model" sentence here outlived the change -- the exact
 // stale-banner genus this block documents).  MissileLauncher (@004bcc60)
 // is the genuinely-flying family (Model B) and overrides this slot.
 //
 // Gitea #12: the view/target gate, the ammo pull and the recoil set are
 // STRIPPED from this body -- they belong to the CALLER (the recovered
@@ -1049,9 +986,10 @@ void
 {
 	Check(this);
-	// (ComputeTimeOfFlight @004bc06c is called INSIDE the tracer branch below,
+	// Refresh the lead solution (leadPosition / targeting) for the spawn below
-	// exactly where @004bc104 calls it -- :16091, between the muzzle resolve and
+	// (@004bc06c; the recovered sim no longer calls it -- best-effort placement
-	// the range gate.  The old top-of-body placement predated the re-export.)
+	// inside the unrecovered @4bc104 body, which is where the lead data is used).
 	ComputeTimeOfFlight();								// @004bc06c
 	// THE FIRING HEAT (task #9; was "a separate pre-existing gap"): the
 	// binary adds heatCostToFire RAW to the weapon's own pendingHeat --
@@ -1093,82 +1031,27 @@ void
 		}
 	}
-	//
+	// WAVE 7 Phase B: launch one flying ballistic round toward the owner's target (port
-	// #171 (2026-08-13) -- THE AUTHENTIC AC: TRIGGER-TIME HITSCAN (@004bc104
+	// reconstruction; the byte-exact world-entity Projectile is blocked by the 2007 engine
-	// [T1], the body the stale banner hid; the prior flying-damage round was
+	// Entity base mismatch -- see BTPushProjectile / mech4.cpp).  Speed = |launchVelocity|.
 	// the #168 mis-generalization from the MISSILE launcher).  Binary shape:
 	// increment the round counter (+0x408); every TracerInterval-th (+0x438)
 	// round, dist = |owner aimPoint(+0x37c) - muzzle| gated against the LIVE
 	// effectiveRange (+0x328, the same gate the Emitter uses) -> ONE
 	// single-panel Damage (amount = batchedCount x DamageAmount) to the
 	// LOCKED target *(mech+0x388) via SendDamageMessage @004b9728, AT FIRE
 	// TIME.  Out of range: the batch silently discards.  A locked AC inside
 	// range cannot miss; nothing about the hit rides the visible round.
 	// (This also restores the binary's ONE-message-per-batch = single-panel
 	// damage -- closing #112's multi-zone shotgun, which was the port's
 	// per-round zone re-roll scatter.)
 	//
 	char *o = (char *)owner;							// inherited MechSubsystem::owner (the Mech)
 	// The owner's target slots (raw 0x388/0x37c -- the same raw offsets mech4's
 	// targeting block writes; a consistent producer/consumer pair on our
 	// compiled object).  Task #41: the slots hold the WORLD PICK -- the enemy
 	// mech under the boresight, or the terrain downrange (a missile fired at
 	// the scenery flies to the ground point and detonates without damage), or
 	// null at the sky (BTPushProjectile refuses; the weapon's own 0x388 gate
 	// wouldn't have fired either).
 	void *target = (o != 0) ? *(void **)(o + 0x388) : 0;
 	Point3D targetPos = (o != 0) ? *(Point3D *)(o + 0x37c) : muzzle;
 	++tracerCounter;									// @0x408 -- incremented BEFORE the test
 	// @4bc104:16079: bVar1 = (interval != 0) && (counter % interval == 0).
 	// interval==0 -> bVar1 never set: NO damage ever, counter never resets
 	// (and the short-circuit means no divide) -- the port matches exactly.
 	if (tracerInterval != 0 && (tracerCounter % tracerInterval) == 0)
 	{
 		// @4bc104:16088-16091: muzzle transform (@004b9948, resolved above as
 		// `muzzle`), then the lead/targeting refresh -- INSIDE the tracer
 		// branch, not per-round.  (Also re-derives effectiveRange via
 		// UpdateTargeting, so the gate below reads the LIVE value.)
 		ComputeTimeOfFlight();							// @004bc06c
 		Vector3D toAim;
 		toAim.Subtract(targetPos, muzzle);				// :16096 FUN_00408644(aimPoint(+0x37c), muzzle)
 		Scalar dist = (Scalar)sqrtf(toAim.x*toAim.x + toAim.y*toAim.y + toAim.z*toAim.z);
 		if (dist <= effectiveRange && target != 0)		// :16098 gate (+0x328); the null test is the
 		{												//   port's guard (binary trusts 0x388!=0 upstream)
 			// @4bc104 fills EXACTLY TWO fields of the batched record (:16100-16120):
 			//   damageAmount = counter x authored amount        (:16110)
 			//   damageForce  = rootSegRot . leadPosition(+0x420) x float(+0x41c)
 			// impactPoint is NOT set here -- SendDamageMessage @004b9728 sets it
 			// (FUN_00408440(dmg+0x20, mech+0x37c) = the aim point); the port's
 			// SendDamageMessage doesn't do that write, so it is FOLDED here as
 			// impactPoint = targetPos (net-identical).  burstCount RIDES the
 			// authored record untouched (Damage ctor = 1 for an AC; only
 			// MissileLauncher re-authors it, and that class overrides FireWeapon).
 			// damageForce divergence [T3, deliberate]: the binary's value is the
 			// lead point rotated by the mech root-segment matrix scaled by the
 			// float at +0x41c (a dword lifted from the tracer-model record --
 			// direction-equivalent junk; gyro rattle consumes DIRECTION only,
 			// and the Emitter's own path authors force = target - muzzle
 			// (emitter.cpp:360 @0x3B0)), so the port uses the same toAim shape.
 			Scalar batched = (Scalar)tracerCounter * damageData.damageAmount;	// :16110
 			Damage batchSave = damageData;				// binary builds a LOCAL copy (:16100);
 			damageData.damageAmount = batched;			//   the port mutates+restores because our
 			damageData.impactPoint  = targetPos;		//   SendDamageMessage reads the member
 			damageData.damageForce  = toAim;
 			SendDamageMessage((Entity *)target);		// @004b9728 -- at fire time
 			damageData = batchSave;
 		}
 		tracerCounter = 0;								// :16123 -- resets on HIT and MISS alike,
 	}													//   but ONLY on tracer rounds
 	// The VISIBLE round: the binary's 0xBCD renderer tracer, cosmetic only
 	// (the shell entity 0xBD1 is dead code -- nothing shipped ever spawns a
 	// damage-carrying ballistic round).  Fly the pool round as that visual:
 	// damage 0, straight at the pick, authentic muzzle speed
 	// (= effRange / sqrt(2*(1/9.8)*muzzleHeight), the recovered @004bc3fc
 	// constants -- see ResolveLaunchVelocity).
 	ResolveLaunchVelocity();
 	Scalar  speed = (Scalar)sqrtf(launchVelocity.x*launchVelocity.x
 								+ launchVelocity.y*launchVelocity.y
 								+ launchVelocity.z*launchVelocity.z);
-	BTPushProjectile(muzzle, o, target, targetPos, speed, 0.0f /*cosmetic tracer*/,
+	BTPushProjectile(muzzle, o, target, targetPos, speed, damageData.damageAmount,
 					 0 /*aim straight at the pick*/, 0 /*BALLISTIC: no seeker on a shell*/,
 					 subsystemID /*messmgr explosion bundling at impact (task #7)*/,
 					 0 /*splash_burst*/, -1 /*muzzle_seg*/,
-					 (int)damageData.damageType);
+					 (int)damageData.damageType /*issue: autocannon = Ballistic, was Explosive*/);
 	// task #61: mark this shot for REPLICATION so the peer sees the enemy's
 	// cannon.  ++fireCounter; the CALLER's Loaded case makes the two
@@ -1228,7 +1111,6 @@ void
 	// plus the DAFC muzzle flash.
 	Point3D mz;
 	GetMuzzlePoint(mz);										// @004b9948
 	ResolveLaunchVelocity();								// deferred @004bc3fc z term
 	Scalar spd = (Scalar)sqrtf(launchVelocity.x*launchVelocity.x
 							 + launchVelocity.y*launchVelocity.y
 							 + launchVelocity.z*launchVelocity.z);
@@ -283,14 +283,6 @@ class NotationFile;
 		Scalar
 			ComputeTimeOfFlight();
 		// The @004bc3fc ctor launch-speed expression
 		// (launchVelocity.z = -effRange / sqrt((1/9.8) * 2.0 * muzzleHeight);
 		// constants byte-confirmed @0x4bc678/@0x4bc67c, audit 2026-08-13),
 		// deferred to first use because the port ctor runs mid-roster-build.
 		// z == 0 is the unresolved sentinel; no data member added (layout locked).
 		void
 			ResolveLaunchVelocity();
 		// --- cross-family isolation helpers (owning Mech subsystem roster +
 		//     cockpit controls live in the mech/controls families).  Declared
 		//     here so the recovered bodies compile; the mech family wires the
@@ -6,11 +6,6 @@ rem ONE unified build (2026-07-21): the exe carries every layer; the GLASS
 rem platform env arms the desktop cockpit (clickable MFD buttons, trigger
 rem config, bindings.txt keymap).
 set BT_PLATFORM=glass
 rem Weapons-verification forensics (2026-08-13): named per-shot receipts in the
 rem session log -- makes every weapon incident attributable by name in Friday's
 rem logs (#165/#171 verification net).  Costs log verbosity only; SENDLOGS zips
 rem compress it away.  Remove after the weapons-verification cycle closes.
 set BT_DMG_LOG=1
 if not exist "build\Release\btl4.exe" goto badpath
 if not exist "content\OPERATOR.EGG" goto badpath
 set BT_RELAY=107.202.218.169:1500
@@ -9,11 +9,6 @@ rem ONE unified build (2026-07-21): the exe carries every layer; the GLASS
 rem platform env arms the desktop cockpit (clickable MFD buttons, trigger
 rem config, bindings.txt keymap).
 set BT_PLATFORM=glass
 rem Weapons-verification forensics (2026-08-13): named per-shot receipts in the
 rem session log -- makes every weapon incident attributable by name in Friday's
 rem logs (#165/#171 verification net).  Costs log verbosity only; SENDLOGS zips
 rem compress it away.  Remove after the weapons-verification cycle closes.
 set BT_DMG_LOG=1
 if not exist "build\Release\btl4.exe" goto badpath
 if not exist "content\OPERATOR.EGG" goto badpath
 set BT_RELAY=10.0.0.46:1500
@@ -8,11 +8,6 @@ rem ONE unified build (2026-07-21): the exe carries every layer; the GLASS
 rem platform env arms the desktop cockpit (clickable MFD buttons, trigger
 rem config, bindings.txt keymap).
 set BT_PLATFORM=glass
 rem Weapons-verification forensics (2026-08-13): named per-shot receipts in the
 rem session log -- makes every weapon incident attributable by name in Friday's
 rem logs (#165/#171 verification net).  Costs log verbosity only; SENDLOGS zips
 rem compress it away.  Remove after the weapons-verification cycle closes.
 set BT_DMG_LOG=1
 if not exist "build\Release\btl4.exe" goto badpath
 if not exist "content\OPERATOR.EGG" goto badpath
 set BT_START_INSIDE=1
@@ -8,11 +8,6 @@ cd /d %~dp0
 if not exist "build\Release\btl4.exe" goto badpath
 if not exist "content\BTL4.RES" goto badpath
 set BT_PLATFORM=glass
 rem Weapons-verification forensics (2026-08-13): named per-shot receipts in the
 rem session log -- makes every weapon incident attributable by name in Friday's
 rem logs (#165/#171 verification net).  Costs log verbosity only; SENDLOGS zips
 rem compress it away.  Remove after the weapons-verification cycle closes.
 set BT_DMG_LOG=1
 set BT_STEAM_NET=1
 rem Cockpit view + MFD gauges -- the same client flags the join bats set
 rem (field report: players booted 3rd-person without them).  Inherited by
@@ -1,88 +0,0 @@
 #!/usr/bin/env python
 # Forensics for the 11 [projectile] BYSTANDER receipts of the weapons_sweep
 # run: classify every DET lock pointer by aim-point motion (a mech lock TRACKS
 # the walking enemy; a scenery/wreck lock is stationary), then test each
 # receipt: struck id vs the shooter's own id, det-to-aim distance, and the
 # paired IMPACT delivery.
 import io
 import math
 import re
 RX_DET = re.compile(r"\[projectile\] DET at\(([^)]*)\) dmg=([\d.eE+\-]+) "
                    r"lead=(\d+) tgt=(\S+) aim\(([^)]*)\)")
 RX_BYS = re.compile(r"\[projectile\] BYSTANDER id=(\d+) t=(\S+) at\(([^)]*)\) "
                    r"dmg=([\d.eE+\-]+) lockedTgt=(\S+)")
 RX_RST = re.compile(r"\[respawn\] Mech::Reset (\d+):(\d+) ")
 def p3(s):
    a = [float(x) for x in s.split(",")]
    return a
 def dist(a, b):
    return math.sqrt(sum((a[i] - b[i]) ** 2 for i in range(3)))
 for fn in ("ws_a.log", "ws_b.log"):
    lines = io.open("/c/git/bt411/content/" + fn.replace("/", ""),
                    encoding="latin-1", errors="replace").read().splitlines() \
        if False else io.open("C:/git/bt411/content/" + fn,
                              encoding="latin-1", errors="replace").read().splitlines()
    own = None
    dets = {}          # tgt ptr -> list of (lineno, aim)
    bys = []           # (lineno, id, at, dmg, lockedTgt)
    for i, l in enumerate(lines):
        m = RX_DET.search(l)
        if m and m.group(4) != "0x0" and float(m.group(2)) > 0.0:
            dets.setdefault(m.group(4), []).append((i, p3(m.group(5))))
            continue
        m = RX_BYS.search(l)
        if m:
            bys.append((i, int(m.group(1)), p3(m.group(3)),
                        float(m.group(4)), m.group(5)))
            continue
        m = RX_RST.search(l)
        if m and own is None:
            own = int(m.group(2))
    print("=== %s (own mech entity=%s) ===" % (fn, own))
    # classify pointers: max pairwise aim distance within a 1500-line window
    cls = {}
    for ptr, lst in sorted(dets.items(), key=lambda kv: -len(kv[1])):
        move = 0.0
        for j in range(1, len(lst)):
            if lst[j][0] - lst[j - 1][0] <= 1500:
                d = dist(lst[j][1], lst[j - 1][1])
                if d > move:
                    move = d
        cls[ptr] = "MECH-TRACKING" if move > 60.0 else "STATIONARY"
        print("  lock %s  damage-DETs=%3d  max local aim motion=%7.1fu  -> %s"
              % (ptr, len(lst), move, cls[ptr]))
    ok = True
    for (i, mid, at, dmg, ptr) in bys:
        # find this receipt's own DET (next damage DET line)
        aim = None
        imp = False
        for j in range(i, min(i + 4, len(lines))):
            m = RX_DET.search(lines[j])
            if m and aim is None:
                aim = p3(m.group(5))
            if "[projectile] IMPACT damage=" in lines[j]:
                imp = True
        d_aim = dist(at, aim) if aim else -1.0
        verdict = []
        if mid == own:
            verdict.append("SELF-STRIKE (geometry bug)")
        if cls.get(ptr) == "MECH-TRACKING":
            verdict.append("LOCK WAS THE MECH (exclusion failed)")
        if aim is not None and d_aim < 30.0:
            verdict.append("det at the lock's own aim point")
        if not imp:
            verdict.append("no paired IMPACT delivery")
        tag = "TRUE-POSITIVE" if not verdict else "FALSE-POSITIVE: " + "; ".join(verdict)
        if verdict:
            ok = False
        print("  bys@%-6d struck=%d dmg=%g lock=%s(%s) det-to-aim=%.0fu imp=%d  %s"
              % (i + 1, mid, dmg, ptr, cls.get(ptr, "?"), d_aim, imp, tag))
    print("  %s: %s" % (fn, "all receipts TRUE positives" if ok
                        else "FALSE POSITIVES PRESENT"))
@@ -1,533 +0,0 @@
 #!/usr/bin/env python
 # weapons_sweep checker -- ONE comprehensive per-family PASS/FAIL adjudicator
 # for scratchpad/night16/weapons_sweep.sh (2-node mad2-vs-madcat fight).
 #
 # Families / assertions (env on both nodes: BT_DMG_LOG BT_PROJ_LOG BT_DEATH_LOG
 # BT_AMMO_LOG):
 #   A1 ENERGY : named [emitter] FIRED lines present AND laser/PPC [dmghit]
 #               (type 3/4) land on the victim.
 #   A2 AC     : type=1 [dmghit] present, ALL of them burst=1 (single panel,
 #               trigger-time hitscan); ZERO damage-carrying ballistic rounds in
 #               the pool ([projectile] PUSH guided=0 must all be dmg=0 -- the
 #               #171 restoration: the visible round is the cosmetic 0xBCD
 #               tracer); every damage-carrying DET is reconciled to a guided
 #               (missile) push.
 #   A3 MISSILE: [ammo] LRM/SRM FIRED receipts, [projectile] IMPACT dmg>0
 #               deliveries with burst>1 salvo bursts, type=2 [dmghit]s landing
 #               (burst>1 present) -- contact through the 4.0u fuze at
 #               drag-governed speeds; plus >=1 death or zone cascade (vitality).
 #   A4 NO-REGR: no zone cascades twice within one life per log (the #174
 #               guard); every [projectile] BYSTANDER receipt in the 2-mech
 #               fight is a TRUE positive of the lane-12 sweep.
 #               CALIBRATION NOTE (2026-08-13 run): the original "zero receipts
 #               in a 2-mech fight" operationalization assumed a lock is always
 #               the enemy mech.  The run disproved the ASSUMPTION, not the
 #               code: the production aim ray designates SCENERY constantly
 #               (the ac_bench lesson), and rounds locked onto a structure DO
 #               fly past the other mech -- the binary's world sweep detonates
 #               on ANY solid in the path (FUN_0042291c via @004bef78 [T1]), so
 #               a receipt there is the sweep WORKING.  Forensics on all 11
 #               receipts of the calibration run (bys_forensics.py): struck ==
 #               the other live mech every time, aim point 146-1016u away from
 #               the strike (the round was flying elsewhere), one paired IMPACT
 #               delivery each.  The check now asserts the actual false-positive
 #               conditions per receipt:
 #                 - struck id == the shooter's OWN mech  -> self-strike bug
 #                 - struck id not a known player mech    -> phantom strike
 #                 - no DET/IMPACT pairing                -> lost/duplicated delivery
 #                 - strike within 30u of the round's aim -> the sweep raced the
 #                   authentic 4.0u fuze at the lock itself (exclusion suspect)
 #               and flags a hit receipt-print cap (24) as unmeasurable.
 #   A5 FLOORS : rig-sanity minimums so a quiet rig cannot PASS.
 #
 # --selftest: prove the detector CAN fail (the week's hard rule) by feeding it
 # synthetic pre-fix log shapes (pre-#171 damage-carrying ballistic round,
 # pre-#174 double cascade, a bystander receipt, and an empty/quiet rig) and
 # requiring each to FAIL the matching assertion.
 import io
 import math
 import re
 import sys
 RX_DMGHIT = re.compile(r"\[dmghit\] mech=(\S+) zone=(-?\d+) vital=(\d+) "
                       r"type=(\d+) amt=([\d.eE+\-]+) burst=(\d+)")
 RX_PUSH   = re.compile(r"\[projectile\] PUSH target=(\S+) len=(\S+) speed=(\S+) "
                       r"dmg=([\d.eE+\-]+) guided=(\d+)")
 RX_DET    = re.compile(r"\[projectile\] DET .*?dmg=([\d.eE+\-]+) lead=(\d+)")
 RX_IMPACT = re.compile(r"\[projectile\] IMPACT damage=([\d.eE+\-]+) .*?burst=(\d+)")
 RX_EMIT   = re.compile(r"\[emitter\] FIRED '([^']+)'")
 RX_AMMO   = re.compile(r"\[ammo\] (\S+) FIRED, rounds left=")
 RX_CASC   = re.compile(r"\[cascade\] zone (\d+) DESTROYED")
 RX_BYS    = re.compile(r"\[projectile\] BYSTANDER id=(\d+) t=(\S+) at\(([^)]*)\) "
                       r"dmg=([\d.eE+\-]+) lockedTgt=(\S+)")
 RX_DETAIM = re.compile(r"\[projectile\] DET at\(([^)]*)\).*?aim\(([^)]*)\)")
 RX_RESET  = re.compile(r"Mech::Reset (\d+):(\d+)")
 RX_DEATH  = re.compile(r"\[death\] VehicleDead")
 # rig-sanity floors (A5) -- calibrated against the proven kd_bench/cascade_bench
 # engagement levels (worker's 180s single-shooter evidence run: 200 dmghits).
 FLOOR_DMGHITS_TOTAL  = 40
 FLOOR_ENERGY_HITS    = 5      # type 3+4 dmghits
 FLOOR_AC_HITS        = 3      # type=1 dmghits
 FLOOR_MISSILE_HITS   = 3      # type=2 dmghits
 FLOOR_MISSILE_BURSTS = 2      # IMPACT dmg>0 burst>1 deliveries
 FLOOR_EMITTER_FIRED  = 10
 FLOOR_VITALITY       = 1      # deaths + cascades
 def _p3(s):
    try:
        v = [float(x) for x in s.split(",")]
        return v if len(v) == 3 else None
    except ValueError:
        return None
 def _dist(a, b):
    return math.sqrt(sum((a[i] - b[i]) ** 2 for i in range(3)))
 def parse(lines):
    lines = list(lines)
    d = {
        "dmghit": [],      # (mech, zone, vital, type, amt, burst)
        "push": [],        # (dmg, guided)
        "det": [],         # (dmg, lead)
        "impact": [],      # (dmg, burst)
        "emitter": {},     # name -> count
        "ammo": {},        # name -> count
        "bystander": 0,
        "bys_recs": [],    # (line, struckid, at, dmg, lockptr, aim|None, paired)
        "bys_capped": 0,   # the 24-receipt print cap was reached
        "own": None,       # this node's own mech entity id (first Reset h:ID)
        "mechids": set(),  # every player-mech entity id seen in Reset lines
        "deaths": 0,
        "resets": 0,
        "casc_total": 0,
        "casc_worst": {},  # zone -> worst per-life repeat count
    }
    life = {}
    for i, l in enumerate(lines):
        m = RX_DMGHIT.search(l)
        if m:
            d["dmghit"].append((m.group(1), int(m.group(2)), int(m.group(3)),
                                int(m.group(4)), float(m.group(5)), int(m.group(6))))
            continue
        m = RX_PUSH.search(l)
        if m:
            d["push"].append((float(m.group(4)), int(m.group(5))))
            continue
        m = RX_BYS.search(l)
        if m:
            struck = int(m.group(1))
            at = _p3(m.group(3))
            if "receipt cap reached" in l:
                d["bys_capped"] = 1
            # the receipt's own detonation follows within a few lines: DET
            # (carries the round's aim) then the mech delivery IMPACT
            aim = None
            paired = False
            for j in range(i + 1, min(i + 5, len(lines))):
                dm = RX_DETAIM.search(lines[j])
                if dm and aim is None:
                    aim = _p3(dm.group(2))
                if "[projectile] IMPACT damage=" in lines[j]:
                    paired = True
                    break
            d["bys_recs"].append((i + 1, struck, at, float(m.group(4)),
                                  m.group(5), aim, paired))
            d["bystander"] += 1
            continue
        m = RX_DET.search(l)
        if m:
            d["det"].append((float(m.group(1)), int(m.group(2))))
            continue
        m = RX_IMPACT.search(l)
        if m:
            d["impact"].append((float(m.group(1)), int(m.group(2))))
            continue
        m = RX_EMIT.search(l)
        if m:
            d["emitter"][m.group(1)] = d["emitter"].get(m.group(1), 0) + 1
            continue
        m = RX_AMMO.search(l)
        if m:
            d["ammo"][m.group(1)] = d["ammo"].get(m.group(1), 0) + 1
            continue
        m = RX_CASC.search(l)
        if m:
            z = int(m.group(1))
            life[z] = life.get(z, 0) + 1
            if life[z] > d["casc_worst"].get(z, 0):
                d["casc_worst"][z] = life[z]
            d["casc_total"] += 1
            continue
        if "Mech::Reset" in l:
            life = {}
            d["resets"] += 1
            m = RX_RESET.search(l)
            if m:
                mid = int(m.group(2))
                d["mechids"].add(mid)
                if d["own"] is None:
                    d["own"] = mid
            continue
        if RX_DEATH.search(l):
            d["deaths"] += 1
    return d
 def adjudicate(logs, verbose=True):
    """logs: {name: parsed-dict}.  Returns (fails, table_rows)."""
    fails = []
    def agg(key):
        out = []
        for nm in logs:
            out.extend(logs[nm][key])
        return out
    hits    = agg("dmghit")
    push    = agg("push")
    det     = agg("det")
    impact  = agg("impact")
    emitter = {}
    ammo    = {}
    for nm in logs:
        for k, v in logs[nm]["emitter"].items():
            emitter[k] = emitter.get(k, 0) + v
        for k, v in logs[nm]["ammo"].items():
            ammo[k] = ammo.get(k, 0) + v
    bystander = sum(logs[nm]["bystander"] for nm in logs)
    deaths    = sum(logs[nm]["deaths"] for nm in logs)
    resets    = sum(logs[nm]["resets"] for nm in logs)
    cascades  = sum(logs[nm]["casc_total"] for nm in logs)
    h_energy  = [h for h in hits if h[3] in (3, 4)]
    h_laser   = [h for h in hits if h[3] == 3]
    h_ppc     = [h for h in hits if h[3] == 4]
    h_ball    = [h for h in hits if h[3] == 1]
    h_missile = [h for h in hits if h[3] == 2]
    h_mis_multi = [h for h in h_missile if h[5] > 1]
    ball_multi  = [h for h in h_ball if h[5] != 1]
    push_ballistic     = [p for p in push if p[1] == 0]
    push_ballistic_dmg = [p for p in push_ballistic if p[0] != 0.0]
    push_guided_dmg    = [p for p in push if p[1] == 1 and p[0] > 0.0]
    det_dmg            = [x for x in det if x[0] != 0.0]
    imp_dmg            = [x for x in impact if x[0] > 0.0]
    imp_multi          = [x for x in imp_dmg if x[1] > 1]
    ammo_ac  = {k: v for k, v in ammo.items() if k.upper().startswith("AFC")}
    ammo_mis = {k: v for k, v in ammo.items()
                if k.upper().startswith(("LRM", "SRM", "STRK", "STREAK", "NARC", "NRK"))}
    # ---- A1 ENERGY ----
    a1 = []
    if not emitter:
        a1.append("no named [emitter] FIRED lines")
    if len(h_energy) < FLOOR_ENERGY_HITS:
        a1.append("laser/PPC dmghits %d < floor %d" % (len(h_energy), FLOOR_ENERGY_HITS))
    if not h_laser:
        a1.append("no type=3 laser dmghit landed")
    if sum(emitter.values()) < FLOOR_EMITTER_FIRED:
        a1.append("emitter FIRED total %d < floor %d"
                  % (sum(emitter.values()), FLOOR_EMITTER_FIRED))
    # ---- A2 AC HITSCAN (#171) ----
    a2 = []
    if len(h_ball) < FLOOR_AC_HITS:
        a2.append("type=1 dmghits %d < floor %d" % (len(h_ball), FLOOR_AC_HITS))
    if ball_multi:
        a2.append("%d type=1 dmghit(s) with burst!=1 (multi-panel ballistic -- "
                  "not the single-panel hitscan)" % len(ball_multi))
    if push_ballistic_dmg:
        a2.append("%d ballistic PUSH(es) with dmg!=0 -- a damage-carrying "
                  "ballistic round entered the pool (pre-#171 shape)"
                  % len(push_ballistic_dmg))
    if not push_ballistic:
        a2.append("no ballistic (guided=0) PUSH at all -- AC never cycled")
    if not ammo_ac:
        a2.append("no [ammo] AFC* FIRED receipt -- no autocannon cycled a round")
    # every damage-carrying DET must be attributable to a guided (missile) push,
    # per log (pool is node-local)
    for nm in logs:
        nd = len([x for x in logs[nm]["det"] if x[0] != 0.0])
        ng = len([p for p in logs[nm]["push"] if p[1] == 1 and p[0] > 0.0])
        if nd > ng:
            a2.append("%s: %d damage DETs > %d guided damage PUSHes -- an "
                      "unguided round detonated carrying damage" % (nm, nd, ng))
    # ---- A3 MISSILES ----
    a3 = []
    if not ammo_mis:
        a3.append("no [ammo] LRM/SRM FIRED receipt")
    if len(imp_dmg) < FLOOR_MISSILE_HITS:
        a3.append("missile IMPACT deliveries %d < floor %d"
                  % (len(imp_dmg), FLOOR_MISSILE_HITS))
    if len(imp_multi) < FLOOR_MISSILE_BURSTS:
        a3.append("salvo bursts (IMPACT dmg>0 burst>1) %d < floor %d"
                  % (len(imp_multi), FLOOR_MISSILE_BURSTS))
    if len(h_missile) < FLOOR_MISSILE_HITS:
        a3.append("type=2 dmghits %d < floor %d" % (len(h_missile), FLOOR_MISSILE_HITS))
    if not h_mis_multi:
        a3.append("no type=2 dmghit with burst>1 (no salvo cluster landed)")
    if deaths + resets + cascades < FLOOR_VITALITY:
        a3.append("rig vitality: deaths+resets+cascades = %d < %d"
                  % (deaths + resets + cascades, FLOOR_VITALITY))
    # ---- A4 NO REGRESSION ----
    a4 = []
    for nm in logs:
        dup = {z: c for z, c in logs[nm]["casc_worst"].items() if c > 1}
        if dup:
            a4.append("%s re-descended zones %s within one life (#174 ALIVE)"
                      % (nm, dup))
    # BYSTANDER receipts: every receipt must be a TRUE positive (see the
    # calibration note in the header -- rounds locked onto scenery legitimately
    # detonate on the other mech crossing the flight path; the binary's world
    # sweep fires on ANY solid [T1]).  False-positive conditions per receipt:
    all_mechids = set()
    for nm in logs:
        all_mechids |= logs[nm]["mechids"]
    bys_true = 0
    for nm in logs:
        own = logs[nm]["own"]
        if logs[nm]["bys_capped"]:
            a4.append("%s: BYSTANDER print cap (24) reached -- receipt count "
                      "unmeasurable, rerun a shorter window" % nm)
        for (ln, struck, at, bdmg, lockptr, aim, paired) in logs[nm]["bys_recs"]:
            bad = []
            if own is not None and struck == own:
                bad.append("SELF-STRIKE (sweep hit the shooter's own mech)")
            if all_mechids and struck not in all_mechids:
                bad.append("phantom strike: id %d is no known player mech"
                           % struck)
            if aim is None or not paired:
                bad.append("no paired DET/IMPACT delivery (lost or duplicated "
                           "round)")
            elif at is not None and _dist(at, aim) < 30.0:
                bad.append("strike within %.0fu of the round's own aim -- the "
                           "sweep raced the 4.0u fuze at the lock itself "
                           "(target-exclusion suspect)" % _dist(at, aim))
            if bad:
                a4.append("%s:%d BYSTANDER FALSE POSITIVE: %s"
                          % (nm, ln, "; ".join(bad)))
            else:
                bys_true += 1
    # ---- A5 FLOORS ----
    a5 = []
    if len(hits) < FLOOR_DMGHITS_TOTAL:
        a5.append("total dmghits %d < floor %d (quiet rig proves nothing)"
                  % (len(hits), FLOOR_DMGHITS_TOTAL))
    rows = [
        ("ENERGY",  a1, "emitterFIRED=%d(names=%d) hits t3=%d t4=%d"
                        % (sum(emitter.values()), len(emitter),
                           len(h_laser), len(h_ppc))),
        ("AC",      a2, "acFIRED=%d hits t1=%d (burst!=1: %d) ballPUSH=%d "
                        "(dmg!=0: %d)"
                        % (sum(ammo_ac.values()), len(h_ball), len(ball_multi),
                           len(push_ballistic), len(push_ballistic_dmg))),
        ("MISSILE", a3, "misFIRED=%d IMPACTdmg=%d(burst>1:%d) hits t2=%d"
                        "(burst>1:%d) FIZZ n/a"
                        % (sum(ammo_mis.values()), len(imp_dmg), len(imp_multi),
                           len(h_missile), len(h_mis_multi))),
        ("NO-REGR", a4, "cascades=%d worstRepeat=%d bystander=%d "
                        "(true-positive crossfire=%d, false=%d)"
                        % (cascades,
                           max([c for nm in logs
                                for c in logs[nm]["casc_worst"].values()] or [0]),
                           bystander, bys_true, bystander - bys_true)),
        ("RIG",     a5, "dmghits=%d deaths=%d resets=%d DETdmg>0=%d "
                        "guidedDmgPUSH=%d"
                        % (len(hits), deaths, resets, len(det_dmg),
                           len(push_guided_dmg))),
    ]
    for fam, fl, _ in rows:
        fails.extend("%s: %s" % (fam, f) for f in fl)
    if verbose:
        print("=== WEAPONS SWEEP -- per-family verdict ===")
        for fam, fl, info in rows:
            print("%-8s %-4s %s" % (fam, "PASS" if not fl else "FAIL", info))
            for f in fl:
                print("         - %s" % f)
        print("weapon receipts (named, both logs):")
        for k in sorted(set(list(emitter) + list(ammo))):
            n = emitter.get(k, 0) + ammo.get(k, 0)
            src = "emitter" if k in emitter else "ammo"
            print("   %-16s %4d  (%s)" % (k, n, src))
        for nm in logs:
            print("%s: dmghits=%d pushes=%d dets=%d impacts=%d casc=%d "
                  "resets=%d deaths=%d bys=%d"
                  % (nm, len(logs[nm]["dmghit"]), len(logs[nm]["push"]),
                     len(logs[nm]["det"]), len(logs[nm]["impact"]),
                     logs[nm]["casc_total"], logs[nm]["resets"],
                     logs[nm]["deaths"], logs[nm]["bystander"]))
        print("RESULT:", "PASS" if not fails else "FAIL")
        if fails:
            for f in fails:
                print("  FAIL:", f)
    return fails
 def load(fn):
    return parse(io.open(fn, encoding="latin-1", errors="replace").read().splitlines())
 # ---------------------------------------------------------------------------
 # --selftest: the detector must FAIL on pre-fix log shapes (negative controls)
 # ---------------------------------------------------------------------------
 HEALTHY = [
    # energy: named emitters + laser/PPC hits
 ] + [
    "[emitter] FIRED 'ERMLaser_1' damage=8 heat=2" for _ in range(6)
 ] + [
    "[emitter] FIRED 'ERPPC' damage=15 heat=6" for _ in range(6)
 ] + [
    "[dmghit] mech=42 zone=%d vital=0 type=3 amt=8 burst=1 lvl 0->0.1" % (i % 5)
    for i in range(8)
 ] + [
    "[dmghit] mech=42 zone=2 vital=0 type=4 amt=15 burst=1 lvl 0->0.2",
 ] + [
    # AC: named ammo fire, cosmetic tracer pushes, hitscan type=1 burst=1 hits
    "[ammo] AFC50 FIRED, rounds left=93" for _ in range(4)
 ] + [
    "[projectile] PUSH target=0x0 len=400 speed=250 dmg=0 guided=0 ttl=5 "
    "mz=(1,12,1) relY=12 lv=(fallback)" for _ in range(4)
 ] + [
    "[projectile] DET at(9,2,3) dmg=0 lead=1 tgt=0x0 aim(9,2,3)" for _ in range(4)
 ] + [
    "[dmghit] mech=42 zone=%d vital=0 type=1 amt=12.5 burst=1 lvl 0->0.1" % (i % 4)
    for i in range(6)
 ] + [
    # missiles: guided pushes (1 lead + visuals), IMPACT bursts, type=2 hits
    "[ammo] LRM15 FIRED, rounds left=110" for _ in range(4)
 ] + [
    "[projectile] PUSH target=0x5a len=300 speed=180 dmg=52.5 guided=1 ttl=13 "
    "mz=(1,14,1) relY=14 lv=(auth)" for _ in range(6)
 ] + [
    "[projectile] DET at(4,5,6) dmg=52.5 lead=1 tgt=0x5a aim(4,5,6)" for _ in range(6)
 ] + [
    "[projectile] IMPACT damage=52.5 subsys=21 v=540 burnLeft=0 burst=9 "
    "(direct dispatch) (zone cyl-resolved)" for _ in range(6)
 ] + [
    "[dmghit] mech=42 zone=%d vital=0 type=2 amt=3.5 burst=%d lvl 0->0.3"
    % (i % 6, 4 + (i % 8)) for i in range(24)
 ] + [
    "[cascade] zone 17 DESTROYED -> descend=1 destroySibs=0 crits=3",
    "[respawn] Mech::Reset 3:42 healed+moved to (0,0,0) alive=1 zones=22 subsys=34",
    "[cascade] zone 17 DESTROYED -> descend=1 destroySibs=0 crits=3",
    "[death] VehicleDead(-1) dispatched to the owning player",
    # the peer's mech id (a second life witness so 43 is a known player mech)
    "[respawn] Mech::Reset 2:43 healed+moved to (9,0,9) alive=1 zones=22 subsys=32",
 ]
 # a TRUE-positive bystander receipt: struck the OTHER mech (43 != own 42),
 # far from the round's own aim, one paired delivery -- must NOT fail A4.
 BYS_TRUE = [
    "[projectile] BYSTANDER id=43 t=0.7 at(100,10,100) dmg=5 lockedTgt=0BADF00D",
    "[projectile] DET at(100,10,100) dmg=5 lead=1 tgt=0BADF00D aim(500,7,900)",
    "[projectile] IMPACT damage=5 subsys=29 v=300 burnLeft=2 burst=3 "
    "(direct dispatch) (zone cyl-resolved)",
 ]
 def selftest():
    ok = True
    def expect(name, lines_a, want_frag):
        nonlocal ok
        logs = {"fx_a.log": parse(lines_a), "fx_b.log": parse(HEALTHY)}
        fails = adjudicate(logs, verbose=False)
        hit = any(want_frag in f for f in fails)
        print("  selftest %-28s %s" % (name, "DETECTED" if hit else "** MISSED **"))
        if not hit:
            ok = False
            for f in fails:
                print("      got:", f)
    print("=== DETECTOR SELF-TEST (negative controls; each MUST fail) ===")
    # 1. pre-#171: a damage-carrying ballistic round + multi-panel ballistic hit
    fx = list(HEALTHY) + [
        "[projectile] PUSH target=0x5a len=400 speed=250 dmg=12 guided=0 ttl=5 "
        "mz=(1,12,1) relY=12 lv=(fallback)",
        "[projectile] DET at(9,2,3) dmg=12 lead=1 tgt=0x5a aim(9,2,3)",
        "[dmghit] mech=42 zone=3 vital=0 type=1 amt=12 burst=3 lvl 0->0.2",
    ]
    expect("pre-#171 ballistic round", fx, "damage-carrying ballistic")
    expect("pre-#171 multi-panel t1", fx, "burst!=1")
    # 2. pre-#174: same zone cascades twice inside one life
    fx = list(HEALTHY) + [
        "[cascade] zone 18 DESTROYED -> descend=1 destroySibs=0 crits=4",
        "[cascade] zone 18 DESTROYED -> descend=1 destroySibs=0 crits=4",
    ]
    expect("pre-#174 double cascade", fx, "#174 ALIVE")
    # 3. bystander FALSE positives (each must be flagged) ...
    fx = list(HEALTHY) + [
        "[projectile] BYSTANDER id=42 t=0.5 at(50,10,50) dmg=5 lockedTgt=0BADF00D",
        "[projectile] DET at(50,10,50) dmg=5 lead=1 tgt=0BADF00D aim(500,7,900)",
        "[projectile] IMPACT damage=5 subsys=29 v=300 burnLeft=2 burst=3 "
        "(direct dispatch) (zone cyl-resolved)",
    ]
    expect("bystander SELF-strike", fx, "SELF-STRIKE")
    fx = list(HEALTHY) + [
        "[projectile] BYSTANDER id=7 t=0.31 at(4,5,6) dmg=8 lockedTgt=0x5a",
    ]
    expect("bystander phantom strike", fx, "phantom strike")
    fx = list(HEALTHY) + [
        "[projectile] BYSTANDER id=43 t=1 at(499,7,899) dmg=5 lockedTgt=0BADF00D",
        "[projectile] DET at(499,7,899) dmg=5 lead=1 tgt=0BADF00D aim(500,7,900)",
        "[projectile] IMPACT damage=5 subsys=29 v=300 burnLeft=2 burst=3 "
        "(direct dispatch) (zone cyl-resolved)",
    ]
    expect("bystander fuze-race at lock", fx, "raced the 4.0u fuze")
    fx = list(HEALTHY) + BYS_TRUE + [
        "[projectile] BYSTANDER id=43 t=0.9 at(200,10,200) dmg=5 "
        "lockedTgt=0BADF00D (receipt cap reached)",
        "[projectile] DET at(200,10,200) dmg=5 lead=1 tgt=0BADF00D aim(500,7,900)",
        "[projectile] IMPACT damage=5 subsys=29 v=300 burnLeft=2 burst=3 "
        "(direct dispatch) (zone cyl-resolved)",
    ]
    expect("bystander print-cap reached", fx, "cap (24) reached")
    # ... and the TRUE positive must NOT be flagged (no false alarm)
    logs = {"fx_a.log": parse(HEALTHY + BYS_TRUE), "fx_b.log": parse(HEALTHY)}
    fails = adjudicate(logs, verbose=False)
    bysf = [f for f in fails if "BYSTANDER" in f or "SELF" in f]
    print("  selftest %-28s %s" % ("bystander TRUE pos accepted",
                                   "CLEAN" if not bysf else "** FALSE ALARM **"))
    if bysf:
        ok = False
        for f in bysf:
            print("      got:", f)
    # 4. quiet rig: both logs empty must FAIL floors
    logs = {"fx_a.log": parse([]), "fx_b.log": parse([])}
    fails = adjudicate(logs, verbose=False)
    quiet = any("quiet rig" in f for f in fails) and any("floor" in f for f in fails)
    print("  selftest %-28s %s" % ("quiet rig floors",
                                   "DETECTED" if quiet else "** MISSED **"))
    ok = ok and quiet
    # 5. the healthy fixture itself must PASS (no false alarms in the detector)
    logs = {"fx_a.log": parse(HEALTHY), "fx_b.log": parse(HEALTHY)}
    fails = adjudicate(logs, verbose=False)
    print("  selftest %-28s %s" % ("healthy fixture passes",
                                   "CLEAN" if not fails else "** FALSE ALARM **"))
    if fails:
        ok = False
        for f in fails:
            print("      got:", f)
    print("DETECTOR:", "OK -- all negative controls detected" if ok else "BROKEN")
    return ok
 if __name__ == "__main__":
    if len(sys.argv) >= 2 and sys.argv[1] == "--selftest":
        sys.exit(0 if selftest() else 1)
    if len(sys.argv) < 3:
        print("usage: weapons_check.py <log_a> <log_b> | --selftest")
        sys.exit(2)
    logs = {fn: load(fn) for fn in sys.argv[1:]}
    fails = adjudicate(logs, verbose=True)
    sys.exit(0 if not fails else 1)
@@ -1,73 +0,0 @@
 #!/usr/bin/env bash
 # weapons_sweep.sh -- ONE comprehensive 2-node bench proving every weapons fix
 # on the final (2026-08-13) tree: #171 AC trigger-time hitscan, missile
 # contact through the 4.0u fuze at drag-governed speeds (b94636b), energy
 # emitters, the #174 cascade guard, and the lane-12 bystander sweep's
 # no-false-positive property in a 2-mech fight.
 #
 # RIG (modeled on cascade_bench.sh / kd_bench.sh -- the rigs that demonstrably
 # engage).  KNOWN TRAP dodged: avatar-vs-avatar forms ZERO locks (three runs,
 # 0 dmghits even from lasers -- see HITSCAN_STATE.md / ac_bench.sh) while the
 # madcat family fights fine, so BOTH nodes fly madcat-family chassis:
 #   node A (-net 1501 -> pilot 127.0.0.1:1502) : mad2 "Zanin Neko" (Madcat V2)
 #       -- AFC50 autocannon + ERPPC + LRM15 + SRM6 x2 + ERMLaser x2:
 #       every family from one shooter (AC carrier per L4GAUGE.CFG :3008).
 #   node B (-net 1601 -> pilot 127.0.0.1:1602) : madcat
 #       -- AFC100 + LRM15 x2 + ERLLaser + ERSLaser x2 (kd_bench's proven
 #       engager), sparser trigger.
 # Engagement = the kd_bench pattern: BT_GOTO=enemy drives them onto each other
 # and the production aim ray designates the enemy (mech+0x388 lock).  NOT
 # BT_FIRE_AT_ICON (that designates buildings -- fine for the cascade rig's
 # splash chaos, useless for typed mech-vs-mech receipts).  AF periods 4/7:
 # throttled (unthrottled autofire trips the FailureHeat brick) and staggered.
 #
 # Assertions + PASS table: scratchpad/night16/weapons_check.py.  The checker's
 # --selftest (run FIRST, gate on it) proves the detector flags the pre-#171
 # and pre-#174 log shapes, bystander false positives (self-strike / phantom /
 # fuze-race / print-cap), and a quiet rig -- the week's hard rule: prove the
 # bench CAN detect failure before trusting PASS.
 #
 # RESULT 2026-08-13 (evidence: scratchpad/night16/ws_a.log ws_b.log): PASS.
 # ENERGY 184 FIRED / 72 laser + 7 PPC hits; AC 42 FIRED / 17 type=1 hits all
 # burst=1 / 84 ballistic pushes all dmg=0; MISSILE 151 FIRED / 57 damage
 # IMPACTs (50 salvo bursts) / 593 type=2 hits; 11 deaths; cascades 8, none
 # re-descended.  CALIBRATION FINDING: BYSTANDER receipts DO fire in a 2-mech
 # fight (11x) and all were TRUE positives -- the production aim ray locks
 # scenery, and the other mech crossing the round's path detonates it (the
 # binary world-sweep behavior).  The checker asserts the real false-positive
 # conditions instead of a zero count; see weapons_check.py header.
 set -x
 python /c/git/bt411/scratchpad/night16/weapons_check.py --selftest || {
  echo "DETECTOR SELF-TEST FAILED -- fix the checker before benching"; exit 1; }
 . /c/git/bt411/scratchpad/night6/bench_common.sh
 bt_assert_player_env
 cd /c/git/bt411/content || exit 1
 taskkill //F //IM btl4.exe >/dev/null 2>&1; sleep 3
 rm -f ws_a.log ws_b.log ws_r.log
 bt_expert_egg MP.EGG WS.EGG
 # map/time sed is MANDATORY (test-harness: MP.EGG authors cavern/night --
 # un-sedded copies park the mechs against cavern rock).  Vehicle seds are
 # EXACT-LINE (the ^vehicle=.* form would blindly stamp both pilots the same).
 sed -i "s/^map=.*/map=grass/; s/^time=.*/time=day/; s/^vehicle=bhk1$/vehicle=mad2/; s/^vehicle=ava1$/vehicle=madcat/" WS.EGG
 grep -q "^vehicle=mad2$" WS.EGG && grep -q "^vehicle=madcat$" WS.EGG || {
  echo "FAIL: WS.EGG vehicle sed did not land (check MP.EGG pilot pages)"; exit 1; }
 grep -q "advancedDamage=1" WS.EGG || { echo "FAIL: egg lacks advancedDamage=1"; exit 1; }
 # node B FIRST (back window): madcat, sparser trigger, still all families.
 ( export BT_GOTO=enemy BT_GOTO_STOP=80 BT_AUTOFIRE=1 BT_AF_MISSILE=1 BT_AF_PERIOD=7
  export BT_DMG_LOG=1 BT_PROJ_LOG=1 BT_DEATH_LOG=1 BT_AMMO_LOG=1
  bt_launch ws_b.log WS.EGG 0x0C -net 1601 )
 sleep 2
 # node A: mad2 (Zanin Neko), the dense shooter -- AC + missiles + energy.
 ( export BT_GOTO=enemy BT_GOTO_STOP=100 BT_AUTOFIRE=1 BT_AF_MISSILE=1 BT_AF_PERIOD=4
  export BT_DMG_LOG=1 BT_PROJ_LOG=1 BT_DEATH_LOG=1 BT_AMMO_LOG=1
  bt_launch ws_a.log WS.EGG 0x03 -net 1501 )
 sleep 5
 python ../tools/btconsole.py WS.EGG 127.0.0.1:1501 127.0.0.1:1601 > ws_r.log 2>&1 &
 R=$!
 sleep 300
 kill $R 2>/dev/null
 bt_kill_ours; sleep 2; taskkill //F //IM btl4.exe >/dev/null 2>&1
 python /c/git/bt411/scratchpad/night16/weapons_check.py ws_a.log ws_b.log
@@ -1,47 +0,0 @@
 #!/usr/bin/env bash
 # =========================================================================
 # #173 wave 2 -- the field-feel gap (3-4s to governor vs bench 12-13s) and
 # the 28Hz cadence residual, decisive runs:
 #   E) FIELD-SHAPED spawn: no pre-settle, weapons charging + THROTTLED
 #      autofire at scenery from the first frame, sprint at seek-4 --
 #      time-to-degradation + Condenser5/GeneratorD preheat at motion start.
 #   F) 28Hz ring cadence (BT_MYO_RING_HZ=28), balanced valves -- does the
 #      |a| gait-surge term collapse and t1000 stretch?
 #   G) 28Hz ring cadence + PROVEN detent 50 on Condenser5 (BT_VALVE5=2) --
 #      THE FLIP TEST: does boosted seek-4 land UNDER degradation 1000?
 # Same rig as m173v_bench.sh (solo expert vulture, grass/day, seek-4 forced,
 # full autodrive, 0.12 turn), 165s each.
 # =========================================================================
 set -x
 . /c/git/bt411/scratchpad/night6/bench_common.sh
 cd /c/git/bt411/content || exit 1
 taskkill //F //IM btl4.exe > /dev/null 2>&1
 sleep 2
 run_one () {   # run_one <tag> <secs> [extra env pairs...]
  local TAG=$1 SECS=$2; shift 2
  bt_expert_egg MP.EGG M173V.EGG
  sed -i "s/^map=.*/map=grass/; s/^time=.*/time=day/; 0,/^vehicle=.*/s//vehicle=vulture/" M173V.EGG
  local LOG=m173_${TAG}.log
  rm -f "$LOG"
  ( export BT_MYO_LOG=1 BT_HEAT_LOG=1 BT_SPEC_LOG=1 BT_MYOMERS_LOG=1 BT_VALVE_LOG=1
    export BT_FORCE_SEEK=3 BT_AUTODRIVE=1.0 BT_FORCE_TURN=0.12
    for kv in "$@"; do export "$kv"; done
    bt_launch "$LOG" M173V.EGG 0x03 )
  sleep "$SECS"
  bt_kill_ours
  sleep 3
  taskkill //F //IM btl4.exe > /dev/null 2>&1
  sleep 2
 }
 run_one vul_E_field 165 BT_AUTOFIRE=1 BT_FIRE_AT_ICON=1 BT_AF_PERIOD=4 BT_FIRE_LOG=1
 run_one vul_F_28bal 165 BT_MYO_RING_HZ=28
 run_one vul_G_28d50 165 BT_MYO_RING_HZ=28 BT_VALVE5=2
 echo "=================== m173v wave-2 RUNS DONE ==================="
 for f in m173_vul_E_field.log m173_vul_F_28bal.log m173_vul_G_28d50.log; do
  echo "--- $f"
  grep -ac "myotemp" "$f" 2>/dev/null
  grep -a "valve-tx\|valve -> detent" "$f" 2>/dev/null
 done
@@ -1,65 +0,0 @@
 #!/usr/bin/env bash
 # =========================================================================
 # #173 DECISIVE EXPERIMENT -- does boosting coolant loop 5's valve sustain
 # seek-4 on a Vulture, as the equilibrium solver claims, or does it still
 # cook as the field reported ("loop 5 maxed and couldn't shed")?
 #
 # Rig: the m173 measurement pattern (night16 myo173.sh) -- solo expert
 # vulture, grass/day, BT_FORCE_SEEK=3 (seek-4) + BT_AUTODRIVE=1.0 +
 # BT_FORCE_TURN=0.12 (stays on-map near top speed), myomer/condenser temp
 # receipts under the heat-log envs.
 #
 # Valve control: the NEW BT_VALVE5=N hook (mech4.cpp) presses Condenser5's
 # MoveValve N times, spaced 30 frames, starting frame 300.  Every press is
 # receipted by the handler's unconditional "[valve] Condenser5 valve ->
 # detent D" line, and BT_VALVE_LOG receipts the zero-sum share redistribute
 # ("[valve] condenser#5 valveState=V flow=F (total=T)") -- the landed
 # detent is PROVEN, not assumed.
 #
 # Three measurement runs, ~150s sustained seek-4 cruise each, identical
 # otherwise:
 #   A) balanced (no presses)          -- expected: degradation ~14s, eq ~1800
 #   B) BT_VALVE5=2 (detent 1->5->50)  -- THE QUESTION: stays under 1000?
 #   C) BT_VALVE5=3 (wrap to detent 0) -- expected WORSE than A (share 0)
 # Plus one visual run:
 #   D) idle mech, 3 presses spaced 600 frames, BT_SHOT_EVERY=90 -- captures
 #      the Heat MFD valve slider at each detent 1/5/50/0 (indicator audit).
 # =========================================================================
 set -x
 . /c/git/bt411/scratchpad/night6/bench_common.sh
 cd /c/git/bt411/content || exit 1
 taskkill //F //IM btl4.exe > /dev/null 2>&1
 sleep 2
 rm -f vlv_*.png
 run_one () {   # run_one <tag> <secs> [extra env pairs...]
  local TAG=$1 SECS=$2; shift 2
  bt_expert_egg MP.EGG M173V.EGG
  sed -i "s/^map=.*/map=grass/; s/^time=.*/time=day/; 0,/^vehicle=.*/s//vehicle=vulture/" M173V.EGG
  local LOG=m173_${TAG}.log
  rm -f "$LOG"
  ( export BT_MYO_LOG=1 BT_HEAT_LOG=1 BT_SPEC_LOG=1 BT_MYOMERS_LOG=1 BT_VALVE_LOG=1
    for kv in "$@"; do export "$kv"; done
    bt_launch "$LOG" M173V.EGG 0x03 )
  sleep "$SECS"
  bt_kill_ours
  sleep 3
  taskkill //F //IM btl4.exe > /dev/null 2>&1
  sleep 2
 }
 DRIVE="BT_FORCE_SEEK=3 BT_AUTODRIVE=1.0 BT_FORCE_TURN=0.12"
 run_one vul_A_bal 165 $DRIVE
 run_one vul_B_d50 165 $DRIVE BT_VALVE5=2
 run_one vul_C_d0  165 $DRIVE BT_VALVE5=3
 run_one vul_D_shot 100 BT_VALVE5=3 BT_VALVE5_AT=900 BT_VALVE5_SPACING=600 \
        BT_SHOT_EVERY=90 BT_SHOT_PREFIX=vlv
 echo "=================== m173v RUNS DONE ==================="
 for f in m173_vul_A_bal.log m173_vul_B_d50.log m173_vul_C_d0.log m173_vul_D_shot.log; do
  echo "--- $f"
  grep -ac "myotemp" "$f" 2>/dev/null
  grep -a "\[valve" "$f" 2>/dev/null | head -40
 done
 ls vlv_*.png 2>/dev/null | head
@@ -1,68 +0,0 @@
 # #173 valve experiment digest -- per run: valve receipts (PROOF of detent),
 # time-to-degradation-1000, equilibrium T, held speed, C5/bank finals.
 import re, sys, statistics as st
 def digest(path):
    myotemp = []      # (t, T)
    myoheat = []      # (v, a, dt)
    myo     = []      # speedEffect samples in file order
    c5 = []; bank = []
    presses = []      # ([valve-tx] BT_VALVE5 press lines)
    detents = []      # ([valve] CondenserN valve -> detent D lines)
    redist  = []      # ([valve] condenser#N valveState=V flow=F (total=T))
    for ln in open(path, encoding='latin-1', errors='replace'):
        m = re.search(r"\[myotemp\] t=([\d.eE+-]+) T=([\d.eE+-]+)", ln)
        if m: myotemp.append((float(m.group(1)), float(m.group(2)))); continue
        m = re.search(r"\[myoheat\] v=([\d.eE+-]+) a=([\d.eE+-]+) m=[\d.eE+-]+ g=[\d.eE+-]+ ratio=[\d.eE+-]+ dmgGain=[\d.eE+-]+ dt=([\d.eE+-]+)", ln)
        if m: myoheat.append(tuple(map(float, m.groups()))); continue
        m = re.search(r"\[myo\] Myomers.* speed=([\d.eE+-]+).* gear=(\d+)", ln)
        if m: myo.append((float(m.group(1)), int(m.group(2)))); continue
        m = re.search(r"\[heat-t\] (\S+)( \(bank\))? T=([\d.eE+-]+)", ln)
        if m:
            name, isbank, T = m.group(1), m.group(2), float(m.group(3))
            if name == 'Condenser5': c5.append(T)
            elif isbank or name == 'HeatSink': bank.append(T)
            continue
        if '[valve-tx] BT_VALVE5' in ln: presses.append(ln.strip()); continue
        m = re.search(r"\[valve\] (Condenser\d+) valve -> detent (\d+)", ln)
        if m: detents.append((m.group(1), int(m.group(2)))); continue
        m = re.search(r"\[valve\] condenser#(\d) valveState=(\d+) flow=([\d.eE+-]+) \(total=(\d+)\)", ln)
        if m: redist.append(tuple(int(x) if i != 2 else float(x) for i, x in enumerate(m.groups()))); continue
    name = path.split('m173_')[-1].replace('.log', '')
    print("=" * 78)
    print("RUN %s" % name)
    if presses:
        for p in presses: print("  " + p)
    if detents:
        print("  detent walk: " + " -> ".join("%s:%d" % d for d in detents))
    # final redistribute state (last 6 lines)
    if redist:
        last6 = redist[-6:]
        print("  final shares: " + "  ".join("#%d v=%d f=%.4f" % (n, v, f) for (n, v, f, t) in last6)
              + "  (total=%d)" % last6[-1][3])
    if not myotemp:
        print("  NO MYOTEMP DATA"); return
    t_end = myotemp[-1][0]
    t1000 = next((t for t, T in myotemp if T >= 1000), None)
    t2000 = next((t for t, T in myotemp if T >= 2000), None)
    Tpk   = max(T for _, T in myotemp)
    late  = [T for t, T in myotemp if t > t_end - 40.0]
    k = max(2, len(myotemp) // 5)
    dT = (myotemp[-1][1] - myotemp[-k][1]) / max(1e-6, (myotemp[-1][0] - myotemp[-k][0]))
    # pair myoheat with myotemp by index (printed in lockstep)
    n = min(len(myoheat), len(myotemp))
    vlate = [myoheat[i][0] for i in range(n) if myotemp[i][0] > t_end - 40.0]
    vpk   = max(x[0] for x in myoheat) if myoheat else 0.0
    selate = [s for s, g in myo[-20:]]
    gears  = set(g for s, g in myo)
    print("  t1000=%s  t2000=%s  Tpeak=%.0f  eqT(last40s)=%.0f  endSlope=%+.2f/s  (window %.1fs)"
          % ("%.1fs" % t1000 if t1000 else "NEVER", "%.1fs" % t2000 if t2000 else "never",
             Tpk, st.mean(late), dT, t_end))
    print("  vPeak=%.1f  vHeld(last40s)=%.1f u/s  speedEffect(last~20 samples)=%.2f  gear(s)=%s"
          % (vpk, st.mean(vlate) if vlate else 0.0, st.mean(selate) if selate else 0.0, sorted(gears)))
    print("  Condenser5 end T=%.0f   bank end T=%.0f" % (c5[-1] if c5 else 0, bank[-1] if bank else 0))
 if __name__ == '__main__':
    for f in sys.argv[1:]:
        digest(f)
@@ -1,33 +0,0 @@
 #!/usr/bin/env bash
 # #21 CORRECTION bench (2026-08-13): the faithful overcharge clamp.
 #
 # THE OLD SHAPE (21b2bfc, 2026-07-21): 2.5 min of BT_SEEKTEST abuse (dispatches
 # the SAME ReceiverDataMessageOf id-0xb press the clickable seek button sends,
 # every ~2s) bricked the laser pre-rescue and produced "38 overcharge rescues"
 # post-rescue.
 #
 # THE NEW PASS (post-correction -- _DAT_004ba830 is a DOUBLE 1.0, overcharge
 # clamps rechargeLevel to EXACTLY 1.0f):
 #   1. gear cycling happens, including WRAPS past top (prev > next skips
 #      ResetFiringState -- the overcharge composition);
 #   2. overcharge compositions observed ([seek] gear-switch lines where the
 #      in-flight level lands ABOVE the new gear's snap window, level > 1.01*seekV);
 #   3. the weapon KEEPS FIRING after the last overcharge (no dark state);
 #   4. gauges non-zero: [fire-probe] pct=1 (rechargeLevel, the recharge-dial
 #      source) seen after the last overcharge;
 #   5. ZERO "rescue" prints anywhere in the log (the rescue block is gone).
 set -x
 . /c/git/bt411/scratchpad/night6/bench_common.sh
 cd /c/git/bt411/content || exit 1
 taskkill //F //IM btl4.exe >/dev/null 2>&1; sleep 3
 rm -f sk21.log
 bt_expert_egg MP.EGG SK21.EGG
 sed -i "s/^map=.*/map=grass/; s/^time=.*/time=day/; s/^vehicle=.*/vehicle=vulture/" SK21.EGG
 ( export BT_SEEKTEST=1 BT_AUTOFIRE=1 BT_FIRE_AT_ICON=1
  export BT_SEEK_LOG=1 BT_FIRE_LOG=1 BT_DMG_LOG=1
  bt_launch sk21.log SK21.EGG 0x03 )
 sleep 160
 bt_kill_ours; sleep 2; taskkill //F //IM btl4.exe >/dev/null 2>&1
 python /c/git/bt411/scratchpad/night17/seek21_check.py /c/git/bt411/content/sk21.log
@@ -1,68 +0,0 @@
 # #21 correction bench checker -- see seek21_bench.sh for the PASS contract.
 import io, re, sys
 path = sys.argv[1] if len(sys.argv) > 1 else "/c/git/bt411/content/sk21.log"
 L = io.open(path, encoding="latin-1", errors="replace").read().splitlines()
 # Emitter gear-switch lines (Myomers [seek] lines lack "seekV["):
 #   [seek] NAME -> gear N seekV[N]=X table={...} max=M level=L genV=...
 gear_re = re.compile(
    r"\[seek\] (\S+) -> gear (\d+) seekV\[\d+\]=([-\d.eE+]+) .*level=([-\d.eE+]+)")
 gears = []          # (line_idx, name, gear, seekV, level)
 for i, l in enumerate(L):
    m = gear_re.search(l)
    if m:
        gears.append((i, m.group(1), int(m.group(2)),
                      float(m.group(3)), float(m.group(4))))
 names = sorted(set(g[1] for g in gears))
 print("=== #21 CORRECTION BENCH (faithful overcharge clamp) ===")
 print("weapon(s) under seek cycling:", names or "NONE")
 ok = True
 if not gears:
    print("FAIL: no emitter [seek] gear lines -- rig broken (expert gate? seektest?)")
    ok = False
 for name in names:
    G = [g for g in gears if g[1] == name]
    wraps = sum(1 for a, b in zip(G, G[1:]) if b[2] < a[2])
    over  = [g for g in G if g[3] > 0.0 and g[4] > 1.01 * g[3]]
    fired_all  = [i for i, l in enumerate(L) if ("FIRED '%s'" % name) in l]
    last_over  = over[-1][0] if over else None
    fired_after = ([i for i in fired_all if i > last_over]
                   if last_over is not None else [])
    # gauge source: [fire-probe] NAME ... pct=1 after the last overcharge
    pct1_after = 0
    if last_over is not None:
        pr = re.compile(r"\[fire-probe\] %s .*pct=([-\d.eE+]+)" % re.escape(name))
        for l in L[last_over:]:
            m = pr.search(l)
            if m and float(m.group(1)) >= 0.999:
                pct1_after += 1
    # tail liveness: still firing in the last 25% of the run
    tail = int(len(L) * 0.75)
    fired_tail = sum(1 for i in fired_all if i >= tail)
    print("-- %s: gearSwitches=%d wraps=%d overchargeEvents=%d "
          "FIRED total=%d afterLastOver=%d pct=1 probes after=%d tailFIRED=%d"
          % (name, len(G), wraps, len(over), len(fired_all),
             len(fired_after), pct1_after, fired_tail))
    if wraps < 1:
        ok = False; print("FAIL(%s): no wrap past top gear" % name)
    if not over:
        ok = False; print("FAIL(%s): no overcharge composition ever formed" % name)
    if last_over is not None and not fired_after:
        ok = False; print("FAIL(%s): NEVER FIRED after the last overcharge -- dark state" % name)
    if last_over is not None and pct1_after < 1:
        ok = False; print("FAIL(%s): recharge gauge (pct) never read full after overcharge" % name)
    if fired_tail < 1:
        ok = False; print("FAIL(%s): not firing in the run tail -- possible late brick" % name)
 rescues = sum(1 for l in L if "rescue" in l)
 print("rescue prints in log:", rescues)
 if rescues != 0:
    ok = False; print("FAIL: rescue print still present")
 print("RESULT:", "PASS" if ok else "FAIL")
@@ -1,327 +0,0 @@
 #!/usr/bin/env python3
 """mksoundboard.py -- build SOUNDBOARD_BT411.zip, the distributable tester
 soundboard (no Python needed on the tester's machine).
 Run:  python tools/mksoundboard.py          (from the repo root, or anywhere)
 Output: dist/SOUNDBOARD_BT411.zip containing
  index.html   one self-contained page (inline CSS/JS, works from file://)
  AUDIO/       every soundbank wav (copies; a few ultra-low-rate zones get
               their header rate rebased x2^m so browsers can decode them --
               the page's playbackRate compensates exactly)
  README.txt   tester instructions
 The page plays every sample at its GAME rate:
    game rate = WAV header rate x 2^((note - 60) / 12)
 via <audio>.playbackRate with preservesPitch=false (the engine RESAMPLES --
 pitch and speed shift together, so pitch-preserving time-stretch would be
 wrong).  No fetch()/XHR of the wavs -- plain <audio src> works from file://.
 The trigger-note census is shared with tools/soundboard.py (imported).
 """
 import io, json, os, struct, sys, zipfile
 HERE = os.path.dirname(os.path.abspath(__file__))
 sys.path.insert(0, HERE)
 from soundboard import (GAME_TRIGGERS, VOICE_PRESETS, SEQ_PRIMARY_PATCHES,
                        AUDIO, load_zones, wav_fmt, note_shift)
 OUT = os.path.normpath(os.path.join(HERE, "..", "dist", "SOUNDBOARD_BT411.zip"))
 MIN_BROWSER_RATE = 3000     # Chromium refuses to decode wav below ~3 kHz
 def build_entries():
    zones = load_zones()
    files = sorted(f for f in os.listdir(AUDIO) if f.lower().endswith(".wav"))
    by_patch = {}
    for b, p, notes, trig in GAME_TRIGGERS:
        by_patch.setdefault((b, p), []).extend((n, trig) for n in notes)
    entries = []
    for f in files:
        fmt = wav_fmt(os.path.join(AUDIO, f))
        hz = fmt[0] if fmt else 0
        z = zones.get(f)
        plays, seen = [], set()
        if z:
            for n, trig in by_patch.get((z["bank"], z["patch"]), []):
                if z["keyLo"] <= n <= z["keyHi"] and n not in seen:
                    seen.add(n)
                    plays.append(dict(n=n, s=round(note_shift(n), 6), t=trig))
            if z["keyLo"] <= 60 <= z["keyHi"] and 60 not in seen:
                plays.append(dict(n=60, s=1.0, t="default start (note 60)"))
            # primary: sequence-driven patches lead with their first authored
            # sequence note; everything else leads with the note-60 default
            if (z["bank"], z["patch"]) not in SEQ_PRIMARY_PATCHES:
                plays.sort(key=lambda e: (e["n"] != 60, e["n"]))
        if not plays:
            plays = [dict(n=60, s=1.0, t="default start (note 60)")]
        # zip header rate: rebase ultra-low rates so browsers can decode
        zh, m = hz, 0
        while 0 < zh < MIN_BROWSER_RATE:
            zh, m = zh * 2, m + 1
        e = dict(f=f, hz=hz, zh=zh,
                 b=z["bank"] if z else 0, p=z["patch"] if z else -1,
                 lo=z["keyLo"] if z else 0, hi=z["keyHi"] if z else 127,
                 v=1 if (z and (z["bank"], z["patch"]) in VOICE_PRESETS) else 0,
                 plays=plays)
        entries.append(e)
    return entries
 README = """BT411 SOUNDBOARD -- what every game sound is called
 =====================================================
 1. Unzip this whole folder anywhere (keep index.html next to the AUDIO
   folder).
 2. Double-click index.html -- it opens in your browser.  Nothing to
   install, nothing goes online.
 3. Type in the filter box to narrow the list, click a sound to hear it.
 4. Found the sound from your bug?  Hit COPY on its card and PASTE the
   reference string straight into your Discord report.  That string is
   exactly what we need -- no descriptions like "the low buzzy one"
   required (though those are fun too).
 The board plays each sound the way the GAME plays it -- some sounds are
 sped up or slowed down in-game from how they are stored (the "x0.25"
 tags).  The RAW toggle lets you hear the stored file instead, in case
 you are chasing a pitch/speed bug specifically.  Sounds with several
 small note buttons play at several different pitches in-game (warning
 voice phrases, explosion layers, the weapon ready clunk/blip) -- click
 each note button to hear each pitch.
 Cards tagged VOICE are the cockpit warning voice.  Cards tagged UNUSED
 are zones no game event ever triggers (we still include them for
 completeness).
 Thanks for testing!
 """
 def esc(s):
    return s.replace("&", "&amp;").replace("<", "&lt;").replace(">", "&gt;")
 def build_html(entries):
    data = json.dumps(entries, separators=(",", ":"))
    n_files = len(entries)
    n_shift = sum(1 for e in entries for p in e["plays"] if p["s"] != 1.0)
    return """<!DOCTYPE html>
 <html lang="en">
 <head>
 <meta charset="utf-8">
 <title>BT411 Soundboard</title>
 <style>
  :root { color-scheme: dark; }
  body { background:#101418; color:#cfd8e3; font:14px/1.4 "Segoe UI",system-ui,sans-serif; margin:0; }
  header { padding:14px 18px 10px; border-bottom:1px solid #2a3542; background:#151b22; position:sticky; top:0; z-index:5; }
  h1 { font-size:18px; margin:0 0 4px; color:#e8eef5; }
  .sub { color:#8fa3b8; font-size:12.5px; max-width:1000px; }
  .sub b { color:#c8d6e5; }
  .controls { display:flex; gap:10px; align-items:center; flex-wrap:wrap; margin-top:10px; }
  input[type=text] { background:#0c1013; color:#dfe8f2; border:1px solid #33414f; border-radius:4px; padding:5px 8px; width:260px; }
  button { background:#22303e; color:#dfe8f2; border:1px solid #3a4c5e; border-radius:4px; padding:5px 10px; cursor:pointer; }
  button:hover { background:#2c3d4e; }
  label.tog { color:#a9bccd; user-select:none; cursor:pointer; }
  #count { color:#6f8398; font-size:12px; }
  #grid { display:grid; grid-template-columns:repeat(auto-fill,minmax(340px,1fr)); gap:6px; padding:12px 18px 60px; }
  .card { background:#161d25; border:1px solid #26313d; border-radius:6px; padding:7px 9px; }
  .card.playing { border-color:#4d8edb; box-shadow:0 0 0 1px #4d8edb inset; }
  .ref { font-family:Consolas,monospace; font-size:13px; color:#e4ecf4; cursor:pointer; background:none; border:none; padding:0; text-align:left; width:100%; }
  .ref:hover { color:#8fc1f7; background:none; }
  .meta { margin-top:4px; display:flex; gap:6px; flex-wrap:wrap; align-items:center; }
  .tag { font-size:10.5px; padding:1px 6px; border-radius:3px; background:#243240; color:#9db4c9; }
  .tag.shift { background:#3d2f19; color:#e8b25a; }
  .tag.voice { background:#3a2440; color:#d79be8; }
  .tag.unused { background:#3a2626; color:#d99; }
  .noteb { font-size:11px; padding:1px 7px; background:#1d2a37; border:1px solid #33414f; border-radius:3px; cursor:pointer; color:#bcd; }
  .noteb:hover { background:#2c3d4e; }
  .copy { font-size:11px; padding:1px 8px; margin-left:auto; }
  #status { position:fixed; bottom:0; left:0; right:0; background:#151b22; border-top:1px solid #2a3542;
            padding:6px 18px; font-family:Consolas,monospace; font-size:12.5px; color:#9db4c9; }
 </style>
 </head>
 <body>
 <header>
  <h1>BT411 SOUNDBOARD <span style="color:#6f8398;font-weight:normal">&mdash; __NFILES__ samples, played the way the game plays them</span></h1>
  <div class="sub">
    Click a sound to hear it <b>at its in-game speed</b> (an <b>x0.25</b> tag means the game
    plays that file 4x slower than it is stored &mdash; the board reproduces that). Small
    <b>n##</b> buttons appear when one sample fires at several pitches in-game &mdash; click each
    to hear each. <b>When you report a sound in Discord: hit COPY on its card and paste the
    reference string</b> (it looks like <span style="font-family:Consolas,monospace">[2:113] Warnings01_z0 n16</span>) &mdash;
    that string tells us exactly which sound you mean. VOICE = the cockpit warning voice.
  </div>
  <div class="controls">
    <input type="text" id="filter" placeholder="filter... (e.g. laser, warn, foot)">
    <button id="stopb">STOP</button>
    <button id="playall">Play All (filtered)</button>
    <label class="tog"><input type="checkbox" id="raw"> RAW file rate (ignore game shift)</label>
    <span id="count"></span>
  </div>
 </header>
 <div id="grid"></div>
 <div id="status">ready &mdash; __NSHIFT__ (sample, trigger) pairs play rate-shifted in-game</div>
 <script>
 "use strict";
 const DATA = __DATA__;
 const grid = document.getElementById("grid");
 const statusEl = document.getElementById("status");
 const rawEl = document.getElementById("raw");
 const filterEl = document.getElementById("filter");
 const audioCache = new Map();
 let currentAudio = null, currentCard = null, playAllTimer = null;
 function refString(e, note) {
  let s = (e.p >= 0 ? "[" + e.b + ":" + e.p + "] " : "[-:-] ") + e.f.replace(/\\.wav$/i, "");
  if (note !== undefined && note !== 60) s += " n" + note;
  return s;
 }
 function getAudio(e) {
  let a = audioCache.get(e.f);
  if (!a) {
    a = new Audio("AUDIO/" + e.f);          // plain element src: file://-safe, no fetch
    audioCache.set(e.f, a);
  }
  return a;
 }
 function applyNoPreserve(a) {
  a.preservesPitch = false;                  // the engine RESAMPLES: pitch+speed together
  a.mozPreservesPitch = false;
  a.webkitPreservesPitch = false;
 }
 function stopAll() {
  if (playAllTimer) { clearTimeout(playAllTimer); playAllTimer = null; }
  if (currentAudio) { currentAudio.pause(); currentAudio.currentTime = 0; }
  if (currentCard) currentCard.classList.remove("playing");
  currentAudio = null; currentCard = null;
 }
 function playEntry(e, play, card) {
  stopAll();
  const a = getAudio(e);
  applyNoPreserve(a);
  // zh = the header rate of the wav in this zip (rebased x2^m when the true
  // rate is too low for browsers to decode); target = what the game outputs
  const target = rawEl.checked ? e.hz : Math.round(e.hz * play.s);
  let rate = target / e.zh;
  const clamped = rate < 0.0625 || rate > 16;
  rate = Math.min(16, Math.max(0.0625, rate));
  a.playbackRate = rate;
  a.currentTime = 0;
  a.play().catch(err => { statusEl.textContent = "PLAY FAILED " + e.f + " -- " + err; });
  currentAudio = a; currentCard = card;
  if (card) card.classList.add("playing");
  statusEl.textContent = "PLAYING " + refString(e, play.n) + "  |  " +
    (rawEl.checked ? "raw " + e.hz + " Hz" : "game " + target + " Hz (x" + play.s + ")") +
    "  |  " + play.t + (clamped ? "  [browser rate clamp -- approximate]" : "");
 }
 function copyText(txt, btn) {
  const done = () => { const old = btn.textContent; btn.textContent = "copied"; setTimeout(() => btn.textContent = old, 900); };
  if (navigator.clipboard && navigator.clipboard.writeText) {
    navigator.clipboard.writeText(txt).then(done, () => fallbackCopy(txt, done));
  } else fallbackCopy(txt, done);
 }
 function fallbackCopy(txt, done) {
  const ta = document.createElement("textarea");
  ta.value = txt; ta.style.position = "fixed"; ta.style.opacity = "0";
  document.body.appendChild(ta); ta.select();
  try { document.execCommand("copy"); } catch (e) {}
  document.body.removeChild(ta); done();
 }
 function shiftTag(s) {
  return s >= 0.1 ? "x" + s.toFixed(2).replace(/0$/, "") : "x" + s.toFixed(3);
 }
 function build() {
  const f = filterEl.value.toLowerCase();
  grid.textContent = "";
  let shown = 0;
  for (const e of DATA) {
    if (f && e.f.toLowerCase().indexOf(f) < 0) continue;
    shown++;
    const card = document.createElement("div"); card.className = "card";
    const primary = e.plays[0];
    const ref = document.createElement("button"); ref.className = "ref";
    ref.textContent = refString(e, primary.n) + (primary.s !== 1 ? "  " + shiftTag(primary.s) : "");
    ref.title = primary.t + "  |  stored " + e.hz + " Hz, game " + Math.round(e.hz * primary.s) + " Hz";
    ref.addEventListener("click", () => playEntry(e, primary, card));
    card.appendChild(ref);
    const meta = document.createElement("div"); meta.className = "meta";
    if (e.v) { const t = document.createElement("span"); t.className = "tag voice"; t.textContent = "VOICE"; meta.appendChild(t); }
    const seqOnly = !(e.lo <= 60 && 60 <= e.hi);
    if (seqOnly && e.plays.every(p => p.n === 60)) {
      const t = document.createElement("span"); t.className = "tag unused"; t.textContent = "UNUSED IN GAME"; meta.appendChild(t);
    }
    if (e.plays.length > 1) {
      for (const p of e.plays) {
        const nb = document.createElement("button"); nb.className = "noteb";
        nb.textContent = "n" + p.n + " " + shiftTag(p.s);
        nb.title = p.t;
        nb.addEventListener("click", () => playEntry(e, p, card));
        meta.appendChild(nb);
      }
    } else if (primary.s !== 1) {
      const t = document.createElement("span"); t.className = "tag shift";
      t.textContent = "game " + shiftTag(primary.s); t.title = primary.t; meta.appendChild(t);
    }
    const cp = document.createElement("button"); cp.className = "copy"; cp.textContent = "COPY";
    cp.addEventListener("click", () => copyText(refString(e, primary.n), cp));
    meta.appendChild(cp);
    card.appendChild(meta);
    grid.appendChild(card);
  }
  document.getElementById("count").textContent = shown + " shown / " + DATA.length + " total";
 }
 function playAll() {
  stopAll();
  const f = filterEl.value.toLowerCase();
  const list = DATA.filter(e => !f || e.f.toLowerCase().indexOf(f) >= 0);
  let i = 0;
  const step = () => {
    if (i >= list.length) { statusEl.textContent = "play-all done"; playAllTimer = null; return; }
    const e = list[i++];
    playEntry(e, e.plays[0], null);
    playAllTimer = setTimeout(step, 1100);
  };
  step();
 }
 filterEl.addEventListener("input", build);
 document.getElementById("stopb").addEventListener("click", () => { stopAll(); statusEl.textContent = "stopped"; });
 document.getElementById("playall").addEventListener("click", playAll);
 document.addEventListener("keydown", ev => { if (ev.key === "Escape") { stopAll(); statusEl.textContent = "stopped"; } });
 build();
 </script>
 </body>
 </html>
 """.replace("__DATA__", data).replace("__NFILES__", str(n_files)).replace("__NSHIFT__", str(n_shift))
 def rebased_wav_bytes(path, new_rate):
    """the wav with only fmt.dwSamplesPerSec/dwAvgBytesPerSec rewritten."""
    fmt = wav_fmt(path)
    data = bytearray(open(path, "rb").read())
    _, off, block_align = fmt
    struct.pack_into("<I", data, off + 4, new_rate)
    struct.pack_into("<I", data, off + 8, new_rate * block_align)
    return bytes(data)
 def main():
    entries = build_entries()
    html = build_html(entries)
    os.makedirs(os.path.dirname(OUT), exist_ok=True)
    rebased = []
    with zipfile.ZipFile(OUT, "w", zipfile.ZIP_DEFLATED, compresslevel=6) as zf:
        zf.writestr("SOUNDBOARD_BT411/index.html", html)
        zf.writestr("SOUNDBOARD_BT411/README.txt", README)
        for e in entries:
            src = os.path.join(AUDIO, e["f"])
            arc = "SOUNDBOARD_BT411/AUDIO/" + e["f"]
            if e["zh"] != e["hz"]:
                zf.writestr(arc, rebased_wav_bytes(src, e["zh"]))
                rebased.append((e["f"], e["hz"], e["zh"]))
            else:
                zf.write(src, arc)
    size = os.path.getsize(OUT)
    n_shift = sum(1 for e in entries for p in e["plays"] if p["s"] != 1.0)
    print("wrote %s  (%.1f MB, %d wavs, %d shifted (sample,trigger) pairs)" % (
        OUT, size / 1e6, len(entries), n_shift))
    if rebased:
        print("header-rebased for browser decode (playbackRate compensates):")
        for f, hz, zh in rebased:
            print("  %-28s %6d -> %d Hz" % (f, hz, zh))
 if __name__ == "__main__":
    main()
@@ -1,31 +1,16 @@
 #!/usr/bin/env python3
-"""soundboard.py -- click-to-play board for the BT soundbank (content/AUDIO/*.wav),
+"""soundboard.py -- click-to-play board for the BT soundbank (content/AUDIO/*.wav).
 playing every sample at its GAME-ACCURATE rate.
 Run:  python tools/soundboard.py          (from the repo root, or anywhere)
-THE RATE MODEL (see docs/AUDIO_FIDELITY.md F2 + L4AUDIO.cpp:21-24):
+- Click a button to play that sample (stdlib winsound, async -- click again to
-    game playback rate = WAV header rate x 2^((note - 60) / 12)
+  restart, click STOP to silence).
-The port bakes each zone's authored SoundFont tuning into the extracted WAV's
+- The label shows  [bank:patch]  from game/reconstructed/audiopresets.cpp so an
-declared rate (sf2extract.py), then applies AL_PITCH = 2^((note-60)/12) for
+  identified sound maps straight back to the game's (bankID, patchID).
-the triggering MIDI note.  Sources default to note 60 (AUDSRC.cpp DEFAULT_NOTE)
+- Type in the filter box to narrow (substring, case-insensitive).
-so most samples play at their file rate -- but AudioControlSequences streamed
+- "Play All" steps through the filtered list one per second (ESC/STOP to halt).
 from BTL4.RES trigger notes far off 60 (key-splits: the note SELECTS the zone
 AND transposes it).  The authored (patch, note) census below was parsed out of
 BTL4.RES sequence records and cross-checked against live SetupPatch logs [T1].
 - Click a button: plays the sample at its PRIMARY game rate (header-rewritten
  temp copy; winsound resamples like the engine does -- pitch and speed shift
  together).
 - Small per-note buttons appear when one sample fires at several pitches
  (explosion splits, the Warnings01 voice zones, weapon ready clunk/blip).
 - "Raw file rate" toggle: play the untouched WAV for comparison.
 - The label shows [bank:patch], the shift (e.g. x0.25), and V for the recorded
  VOICE zones (Warnings01/AllWarning -- the project's Yip recordings).
 - Filter box narrows (substring, case-insensitive); "Play All" steps through
  the filtered list; ESC/STOP halts.
 """
-import os, re, struct, sys, tempfile, threading, time
+import os, re, sys, threading, time
 import tkinter as tk
 from tkinter import ttk
 import winsound
@@ -33,118 +18,28 @@ import winsound
 HERE  = os.path.dirname(os.path.abspath(__file__))
 AUDIO = os.path.normpath(os.path.join(HERE, "..", "content", "AUDIO"))
 PRESETS_CPP = os.path.normpath(os.path.join(HERE, "..", "game", "reconstructed", "audiopresets.cpp"))
 TMPDIR = os.path.join(tempfile.gettempdir(), "bt411_soundboard")
-# ---------------------------------------------------------------------------
+def load_mapping():
-# The authored trigger-note census [T1]: every AudioControlSequence in
+    """file -> (bank, patch) from audiopresets.cpp (allPresets[b][p] ... file="X.wav")."""
-# BTL4.RES, parsed from the stream records (scratchpad seqparse.py,
+    mapping = {}
 # 2026-08-13) and resolved through its mixer chain to the target patch.
 # Everything NOT listed here is only ever started at DEFAULT_NOTE 60
 # (AUDSRC.cpp:18), i.e. at its file rate.  (bank, patch) are 1-based bank
 # as logged by SetupPatch.  Dynamic pitch-cent modulations (doppler, the
 # torso-twist pitch scale) are runtime effects and are NOT part of the
 # base rate.
 # ---------------------------------------------------------------------------
 GAME_TRIGGERS = [
    # (bank, patch, [notes in authored order], "trigger")
    (2, 113, [16, 19, 23], "coolant-leak voice loop (ReportLeak)"),
    (2, 113, [16, 33, 36], "ammo cook-off countdown voice (FireCountdownStarted)"),
    (2, 113, [29, 16, 26], "generator-out voice phrase (GeneratorState)"),
    (2, 113, [29, 16, 40], "SimulationState warning voice phrase"),
    (2, 115, [30, 18],     "death phrase A (random pick at death)"),
    (2, 115, [49, 43],     "death phrase B (random pick at death)"),
    (2, 115, [95],         "death phrase C (random pick at death)"),
    (2, 83,  [30, 48, 46, 96, 44, 63, 66], "mech-explosion sequence (brnext.scp)"),
    (2, 97,  [45, 38],     "mech fire-loop sequence (brnext.scp)"),
    (2, 64,  [36, 70],     "subsystem-destroyed interior (SimulationState)"),
    (2, 119, [57, 69, 46, 24], "big explosion sequence (bigexp.scp)"),
    (2, 120, [57, 69, 46, 24], "big explosion sequence LOD2 (bigexp.scp)"),
    (2, 119, [69, 51],     "medium explosion sequence (medexp.scp)"),
    (2, 120, [69, 51],     "medium explosion sequence LOD2 (medexp.scp)"),
    (1, 74,  [58],         "incoming-missile lock beeper (IncomingLock)"),
    (1, 83,  [51],         "weapon-configure ticker (ConfigureActivePress)"),
    (1, 36,  [36, 84],     "missile misfire (WeaponState)"),
    (1, 40,  [36, 84],     "missile loaded/ready (WeaponState)"),
    (1, 41,  [36, 84],     "autocannon misfire (WeaponState)"),
    (1, 56,  [36, 84],     "laser loaded/ready (WeaponState)"),
    (1, 70,  [36, 84],     "autocannon jam (WeaponState)"),
    (1, 71,  [36, 84],     "missile jam (WeaponState)"),
    (1, 75,  [36, 84],     "projectile loaded/ready (WeaponState)"),
 ]
 VOICE_PRESETS = {(2, 113), (2, 123)}    # Warnings01, AllWarning -- Yip's recorded voice
 # Patches the game drives ONLY through their sequences (live SetupPatch census:
 # no note-60 non-ZONESKIP starts) -- their PRIMARY board rate is the first
 # authored sequence note, not the note-60 default.  ProgramButton01 (1,83) is
 # deliberately absent: button presses start it at 60 (live-trace evidence);
 # the configure-ticker 51 stays a secondary note button.
 SEQ_PRIMARY_PATCHES = {
    (1, 36), (1, 40), (1, 41), (1, 56), (1, 70), (1, 71), (1, 75),   # loaded/jam/misfire
    (1, 74),                                                          # lock beeper (F7: sequence is the only path)
    (2, 64), (2, 83), (2, 97), (2, 113), (2, 115), (2, 119), (2, 120),
 }
 def note_shift(note):
    return 2.0 ** ((note - 60) / 12.0)
 def load_zones():
    """file -> dict(bank, patch, keyLo, keyHi) from audiopresets.cpp Z() lines."""
    zones = {}
    try:
        text = open(PRESETS_CPP, encoding="utf-8", errors="replace").read()
-        for m in re.finditer(r'Z\((\d+),(\d+),"([^"]+)",(-?\d+),(-?\d+),', text):
+        # blocks look like: allPresets[0][12].samples[0]; ... s.file = "LaserAFire01.wav";
-            zones[m.group(3)] = dict(bank=int(m.group(1)) + 1, patch=int(m.group(2)),
+        for m in re.finditer(r'allPresets\[(\d+)\]\[(\d+)\]\.samples\[0\];.*?s\.file = "([^"]+)";',
-                                     keyLo=int(m.group(4)), keyHi=int(m.group(5)))
+                             text, re.S):
            bank, patch, fname = int(m.group(1)) + 1, int(m.group(2)), m.group(3)
            mapping[fname] = (bank, patch)
    except OSError:
        pass
-    return zones
+    return mapping
 def wav_fmt(path):
    """(rate, fmt_chunk_file_offset, block_align) from the RIFF header."""
    with open(path, "rb") as f:
        head = f.read(12)
        if head[:4] != b"RIFF" or head[8:12] != b"WAVE":
            return None
        while True:
            ch = f.read(8)
            if len(ch) < 8:
                return None
            cid, sz = ch[:4], struct.unpack("<I", ch[4:])[0]
            if cid == b"fmt ":
                off = f.tell()
                data = f.read(sz)
                rate, = struct.unpack_from("<I", data, 4)
                block_align, = struct.unpack_from("<H", data, 12)
                return rate, off, block_align
            f.seek(sz + (sz & 1), 1)
 def make_rate_copy(src, dst, new_rate):
    """byte-exact copy of src with only the fmt chunk's dwSamplesPerSec and
    dwAvgBytesPerSec rewritten to new_rate."""
    fmt = wav_fmt(src)
    if fmt is None:
        return False
    _, off, block_align = fmt
    data = bytearray(open(src, "rb").read())
    struct.pack_into("<I", data, off + 4, new_rate)                  # dwSamplesPerSec
    struct.pack_into("<I", data, off + 8, new_rate * block_align)    # dwAvgBytesPerSec
    open(dst, "wb").write(data)
    return True
 class SoundBoard(tk.Tk):
    def __init__(self):
        super().__init__()
-        self.title("BT411 Soundboard (game rates) -- " + AUDIO)
+        self.title("BT411 Soundboard -- " + AUDIO)
-        self.geometry("1400x800")
+        self.geometry("1150x760")
-        os.makedirs(TMPDIR, exist_ok=True)
+        self.mapping = load_mapping()
        self.zones = load_zones()
        self.files = sorted(f for f in os.listdir(AUDIO) if f.lower().endswith(".wav"))
        self.rates = {}
        for f in self.files:
            fmt = wav_fmt(os.path.join(AUDIO, f))
            self.rates[f] = fmt[0] if fmt else 0
        self.plays = self.build_play_table()
        self.playing_all = False
        top = ttk.Frame(self); top.pack(fill="x", padx=6, pady=4)
@@ -155,9 +50,6 @@ class SoundBoard(tk.Tk):
        self.filter_var.trace_add("write", lambda *a: self.rebuild())
        ttk.Button(top, text="STOP", command=self.stop).pack(side="left", padx=8)
        ttk.Button(top, text="Play All (filtered)", command=self.play_all).pack(side="left")
        self.raw_var = tk.BooleanVar(value=False)
        ttk.Checkbutton(top, text="Raw file rate (ignore game shift)",
                        variable=self.raw_var).pack(side="left", padx=10)
        self.status = tk.StringVar(value=f"{len(self.files)} samples loaded")
        ttk.Label(top, textvariable=self.status, foreground="#06c").pack(side="left", padx=12)
@@ -177,94 +69,30 @@ class SoundBoard(tk.Tk):
        self.rebuild()
    def build_play_table(self):
        """file -> list of (note, shift, trigger); primary first.
        A zone whose key range contains 60 plays at note 60 (= file rate)
        from every plain watcher Start; sequence notes are added on top.
        A zone whose range EXCLUDES 60 is reachable ONLY via its sequence
        notes -- its file-rate rendering never occurs in game."""
        by_patch = {}
        for b, p, notes, trig in GAME_TRIGGERS:
            by_patch.setdefault((b, p), []).extend((n, trig) for n in notes)
        plays = {}
        for f in self.files:
            z = self.zones.get(f)
            entries = []
            if z:
                seen = set()
                for n, trig in by_patch.get((z["bank"], z["patch"]), []):
                    if z["keyLo"] <= n <= z["keyHi"] and n not in seen:
                        seen.add(n)
                        entries.append((n, note_shift(n), trig))
                if z["keyLo"] <= 60 <= z["keyHi"] and 60 not in seen:
                    # default-60 start reaches this zone at file rate
                    entries.append((60, 1.0, "default start (note 60)"))
            if not entries:
                entries = [(60, 1.0, "default start (note 60)")]
            # primary: sequence-driven patches lead with their first authored
            # sequence note; everything else leads with the note-60 default
            if not (z and (z["bank"], z["patch"]) in SEQ_PRIMARY_PATCHES):
                entries.sort(key=lambda e: (e[0] != 60, e[0]))
            plays[f] = entries
        return plays
    def label_for(self, f):
-        z = self.zones.get(f)
+        bp = self.mapping.get(f)
-        tag = f"[{z['bank']}:{z['patch']}] " if z else "[-:-] "
+        tag = f"[{bp[0]}:{bp[1]}] " if bp else "[-:-] "
-        if z and (z["bank"], z["patch"]) in VOICE_PRESETS:
+        return tag + f[:-4]
            tag += "V "
        primary = self.plays[f][0]
        shift_txt = "" if primary[1] == 1.0 else f"  x{primary[1]:.2f}" if primary[1] >= 0.1 else f"  x{primary[1]:.3f}"
        return tag + f[:-4] + shift_txt
    def rebuild(self):
        for w in self.grid_frame.winfo_children():
            w.destroy()
        flt = self.filter_var.get().lower()
        shown = [f for f in self.files if flt in f.lower()]
-        cols = 3
+        cols = 4
        for i, f in enumerate(shown):
-            cell = ttk.Frame(self.grid_frame)
+            b = ttk.Button(self.grid_frame, text=self.label_for(f), width=38,
            cell.grid(row=i // cols, column=i % cols, padx=2, pady=1, sticky="w")
            b = ttk.Button(cell, text=self.label_for(f), width=44,
                           command=lambda ff=f: self.play(ff))
-            b.pack(side="left")
+            b.grid(row=i // cols, column=i % cols, padx=2, pady=1, sticky="w")
            notes = self.plays[f]
            if len(notes) > 1:
                for n, sh, trig in notes:
                    txt = f"n{n}"
                    nb = ttk.Button(cell, text=txt, width=4,
                                    command=lambda ff=f, nn=n: self.play(ff, nn))
                    nb.pack(side="left")
        self.status.set(f"{len(shown)} shown / {len(self.files)} total")
        self.canvas.yview_moveto(0)
-    def path_for(self, f, note):
+    def play(self, f):
        """the file to hand winsound: raw, or a header-rewritten game-rate copy."""
        if self.raw_var.get() or note == 60:
            return os.path.join(AUDIO, f)
        game_rate = int(round(self.rates[f] * note_shift(note)))
        dst = os.path.join(TMPDIR, f"{f[:-4]}@n{note}.wav")
        if not os.path.exists(dst):
            if not make_rate_copy(os.path.join(AUDIO, f), dst, game_rate):
                return os.path.join(AUDIO, f)
        return dst
    def play(self, f, note=None):
        self.playing_all = False
-        entries = self.plays[f]
+        path = os.path.join(AUDIO, f)
        if note is None:
            note, shift, trig = entries[0]
        else:
            shift, trig = next((s, t) for n, s, t in entries if n == note)
        path = self.path_for(f, note)
        winsound.PlaySound(path, winsound.SND_FILENAME | winsound.SND_ASYNC)
-        z = self.zones.get(f)
+        bp = self.mapping.get(f)
-        bp = f"bank {z['bank']}, patch {z['patch']}" if z else "unmapped"
+        self.status.set(f"PLAYING  {f}" + (f"   (bank {bp[0]}, patch {bp[1]})" if bp else ""))
        mode = "RAW" if self.raw_var.get() else f"note {note} x{shift:.4g}"
        game_rate = int(round(self.rates[f] * (1.0 if self.raw_var.get() else note_shift(note))))
        self.status.set(f"PLAYING  {f}  ({bp}; {mode}; {game_rate} Hz; {trig})")
    def stop(self):
        self.playing_all = False
@@ -279,10 +107,9 @@ class SoundBoard(tk.Tk):
            for f in shown:
                if not self.playing_all:
                    return
                note = self.plays[f][0][0]
                path = self.path_for(f, note)
                self.after(0, lambda ff=f: self.status.set("PLAYING  " + ff))
-                winsound.PlaySound(path, winsound.SND_FILENAME | winsound.SND_ASYNC)
+                winsound.PlaySound(os.path.join(AUDIO, f),
                                   winsound.SND_FILENAME | winsound.SND_ASYNC)
                time.sleep(1.0)
            self.playing_all = False
        threading.Thread(target=run, daemon=True).start()