KB housekeeping sweep: end-of-arc staleness purge (2026-08-06)

Corrections (wrong claims fixed + swept):
- open-questions: the DuckState 'no CODE consumer, authentic [T1]' verdict
  RETRACTED (consumer = the unexported posture block; squat clips shipped in
  BTL4.RES all along); the searchlight fog swap re-marked DONE (was 'STILL
  DEFERRED'); crouch index pointer updated.
- combat-damage: old task-#60 kill-score section marked SUPERSEDED by the
  report tail (BTPostKillScore retired).
- pod-hardware: crouch census rows -> COMPLETE; gauges-hud: 07-20 note
  (generators 'unreconstructed') re-swept.

New knowledge:
- reconstruction-gotchas #20: EXPORT-GAP BLINDNESS (4 incidents, 2 falsely
  T1; the byte-scan/named-member/RES-TOC/manifest checklist) + the
  verification-viewpoint bullet in #13 (the crouch capture errors).
- decomp-reference: master posture/crouch block offsets (+0x3f8/+0x79c, the
  myomers cluster) + 8 new bench env gates.
- rendering: beam-material decode (btfx brighten) + look-pass acceptance.
- experience-levels: crouch novice lockout in the +0x25c consumer list.
- build-and-run: the stale-link build ritual; test-harness: bench-script
  gotchas (both previously oral tradition).
- glossary: export-gap-blindness; multiplayer: crouch/searchlight ride
  existing records; subsystems: searchlight visuals pointer.

Status refresh: CLAUDE.md front + project-overview current-state -> core
gameplay reconstruction COMPLETE (774->801), polish phase, #60 census next.
Ledger addenda: INPUT_PATH row 6 complete; RESPAWN_REARM VehicleDead-sender
note. .gitignore: content/ bench artifacts (pngs/logs/eggs/exe) out of
status. checkctx CLEAN.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

context/reconstruction-gotchas.md

@@ -278,6 +278,12 @@ Suspect ANY reconstructed per-frame code with narrow equality/window tests or `x
 state transitions: charge/seek loops, snap comparisons, timers compared with `==`.
 
 ## 13. Verification gotchas (don't fool yourself)
+- **Capture the viewpoint that can SEE the change.** The crouch "pose does not hold" and the
+  false "eye does not drop" residual (2026-08-05/06) were BOTH capture errors: cockpit-view
+  screenshots cannot show your own legs, and canopy-dominant diff crops mask eye-relative motion
+  (the canopy drops WITH the eye — only the through-window ground shifts). Anchor pixel-diff
+  crops on a region the effect MUST change, pair them with a state probe (joint values), and
+  READ one frame with your own eyes before declaring a visual regression.
 
 - **Lazy gauge build:** `GaugeRenderer::BuildConfigurationFile` runs LAZILY. A too-early process
   kill shows `[gskip]=0` / "not built" even though the widget is fine — **wait for the gauge
@@ -543,6 +549,30 @@ segment tables) — the same class of latent overflow.
    0xFEEEFEEE=freed).
 5. For exhaustive multi-function analysis: a read-only Workflow (understand), then implement hands-on.
 
+## 20. Export-gap blindness — absence in the EXPORT is not absence in the BINARY (4 incidents)
+The Ghidra export (`reference/decomp/`) has coverage gaps (#60), and BTL4.RES content never
+appears in it at all — so "no readers", "no caller", "no such clip/asset", and "unreconstructed"
+claims made by grepping the export or the port ALONE are structurally unsound. This class has
+produced four wrong conclusions, two of them [T1]-tagged at the time:
+1. **2026-07-25 searchlight "1995 latent bug"** — compared Searchlight's Performance against
+   THERMALSIGHT's handler (wrong class) and invented a missing bridge; retracted (#61).
+2. **2026-07-2x `DuckState` "has no CODE consumer, authentic [T1]" + "no SQUAT clip survives"** —
+   the consumer was the un-exported master-perf posture block (@0x4aa011), and the clips lived in
+   BTL4.RES (`squ/sqd/squi/sqdi` × 8 chassis); the search covered decomp strings + loose files,
+   never the RES TOC. Disproven by the CROUCH reconstruction (2026-08-06).
+3. **2026-07-31 morning: the myomer drive-scale DELETION** ("the binary has NO dynamic
+   myomer→speed coupling") — the consumer was un-exported; restored same day by raw capstone.
+4. **2026-08-05 "the myomer factor FEEDER (@004b8be3) is unreconstructed"** — it had been fully
+   reconstructed since 07-31 under NAMED members (`speedEffect`, `AvailableOutput`); the grep
+   searched raw offsets. Cost: a duplicate multiply + a dead crouch gate until 08-06.
+**The rule — before claiming "X does not exist / is not reconstructed":**
+- byte-scan `content/BTL4OPT.EXE` for the offset/immediate (disp32 patterns), never just the export;
+- grep the port for NAMED members (check the .hpp for the offset's name) — offsets rot after promotion;
+- for content claims, walk the **BTL4.RES TOC** (`tools/resscan.py` pattern), not the loose tree;
+- check `reference/BT410_SOURCE_MANIFEST.md` + `game/reconstructed/CLASSMAP.md`;
+- and tag the claim's tier by the WEAKEST source consulted — an export-only sweep caps at [T4].
+The #60 gap census (dark-region inventory + a fresh Ghidra re-export) is the systemic fix.
+
 ## Key Relationships
 - Applies to: every topic that reconstructs a class ([[subsystems]], [[combat-damage]], [[gauges-hud]], [[locomotion]]).
 - Uses: [[decomp-reference]] (offsets/ClassIDs), [[reconstruction-method]] (the loop).