Ψ-W2-T2 remediation: atomic temp+rename collapse write, honest unknown-channel fallback, [verify — DAW] markers, corrected+filed bake-collapse deferral, observable collapse message, quiet-NaN test

src/core/capture/CLAUDE.md

@@ -89,6 +89,14 @@ Detail specific to these pure modules:
   which is what the bit-identical-repeats invariant actually asks for. Do not "fix"
   this by hashing pre-collapse — that would make two entries with different audio
   layouts share one identity.
+- **The collapse's minimal rebuild also drops `bext`/iXML/LIST — a source-position
+  consequence, not only a hashing one.** REAPER's renderer writes a `bext` time
+  reference, and REAPER's own import paths can position an item at that BWF timestamp,
+  so a collapsed capture loses it while a declined (non-collapsed) capture from the same
+  action keeps it — two captures from one action behave differently on re-import.
+  `shell/capture/insert.cpp` is unaffected (it drives `SetEditCurPos` + `InsertMedia`
+  rather than reading BWF), so this is not a defect in the shipped insert path.
+  Accepted, not verified against a DAW re-import: `[verify — DAW]`.
 - `tail_control`'s `kDefaultManualTailMs`/`kManualStepMs` and
   `render_settings`'s `kMaxTailMs`/`kAutoTrimThresholdDb` are separate constants
   in separate files by design (panel-facing default/step vs. runaway-guard cap)