Narrow evidence correction done. DRIVER UNCHANGED: 2fd7d68952587f121fb4f6cac81dc8fb50b891a5e5160cd0cb2c04216f3c37b3. MANIFEST d2/D2-STAGING-MANIFEST.sha256 3d6deb07efee6dbad74f1b1e6b379e14575e0a57215b68f43ae43da35b200880, 35 entries, 35 OK 0 FAILED. STAGING d2/D2-STAGING-RESOLVED.md b3ae2bf4... CONTROLS d2/rig-d2/go-and-silent-controls.sh fa1f0bf7... -> .out 2706948b..., 74/74. U. You are right and I had labelled them as more than they were. u1 never supplies a PREPARED at all and u2 schedules the GO at 1700 while the wait gives up at 1600 - both are TIMEOUT results, neither delivers a receipt that is ACCEPTED and only then found late. They are now labelled "TIMEOUT:" in their own names, kept for what they do prove: the real wait's bound (30s and 60s of remaining PREP, not the 900s default) and that a timeout opens nothing. NEW u4, the case the final deadline check exists for: both receipts present and ACCEPTED - no waiting, no timeout - and then the controlled clock crosses the deadline INSIDE teardown_pin_verify, which is where real time passes in a real run (the instrument is read and hashed between the authorization and the window). Measured: the window does NOT open, and the final prep_check exits 8. go was never recorded. V. The final accounting is no longer copied into the harness and v5's source-string assertion is gone. The control now EXTRACTS and EXECUTES the subject's own block - residue_scan post-cleanup through the verdict conjunction - with only port_at/guard_assert/cleanup_left/cfail stubbed. Outcomes, all from the subject's bytes: v3/v4 unreadable final sample -> RESIDUE_COUNT=UNMEASURED -> PENDING even though the removal verified. v5/v5b POSITIVE CONTROL: readable, empty final sample -> VERIFIED, with a measured zero. v6/v6b an UNMEASURED listener is read as live -> PENDING, and the subject records why. v7 a live residual process -> PENDING. v8 NOTHING_TO_REMOVE survives as a terminal measured state. v1/v2 unchanged: attributed LIVE admits the EXISTING stop path with an unreadable alongside. Gate ordering untouched, no reporting machinery added. Only these focused cases ran; the accumulated battery was not re-run. No host query, no elevation, no PREP launch, no field execution. U and V start no process at all - the clock is a shell variable and both blocks run under stubs. Driver frozen at 2fd7d689, holding for todlando's targeted verdict.