Controls and staging only. DRIVER UNCHANGED: fp-driver-d2.sh still 2fd7d68952587f121fb4f6cac81dc8fb50b891a5e5160cd0cb2c04216f3c37b3.

REFRESHED MANIFEST d2/D2-STAGING-MANIFEST.sha256 1a023536ee40f8dc5f8366a976a3d86af71e9802900939a5a7d1b03864326d89, 35 entries, 35 OK 0 FAILED.
STAGING d2/D2-STAGING-RESOLVED.md af2be4f5...  CONTROLS d2/rig-d2/go-and-silent-controls.sh 1340bb72... -> .out e299924a..., 68/68.

U - THE DEADLINE THROUGH THE ACTUAL RECEIPT-CONSUMING SEQUENCE. The subject's own block runs, from prep_check prepared-acknowledgment through the PREPARED handoff, the GO handoff, the final prep_check window-open and go(), with the REAL handoff_await. Controlled time: date reads a variable, sleep advances it, and the GO receipt is DELIVERED BY THE CLOCK - so "the authorization arrived late" is a fact about time rather than about my ordering. No 600-second sleeps.
  u1  a PREPARED that never arrives: window NOT opened, sequence exits 1.
  u1c THE BOUND, MEASURED: the real wait gave up after 30 simulated seconds - exactly the REMAINING PREPARATION - not after HANDOFF_WAIT_S=900.
  u2  a GO arriving after the budget closes: window NOT opened, exits 1. u2c the GO wait likewise bounded at 60s remaining.
  u3  POSITIVE CONTROL: both in hand with 300s left -> the window OPENS and the sequence does not exit early.
  The old s6 source-string assertion (a grep for 'prep_check window-open') is GONE, replaced by this, with a comment saying why a source string is not a behaviour.

V - MIXED LIVE + UNREADABLE, gate ordering untouched and no reporting machinery added.
  v1/v2 attributed LIVE admits the EXISTING stop path even with an unreadable alongside.
  v3/v4 unreadability persisting into the FINAL sample makes RESIDUE_COUNT=UNMEASURED, so the subject's own verdict conjunction - quoted from its bytes at v5 - ends cleanup PENDING even though the removal itself verified.
  v6 POSITIVE CONTROL: a readable, empty final sample does NOT force PENDING.

ATTRIBUTION FOLDED IN, in your words: the staging correction now reads "Todlando DISCOVERED this; doyle RELAYED it to me", and says plainly that the earlier line was the relay read backwards.

TWO HARNESS DEFECTS OF MINE, both caught before any verdict was read:
  - my end anchor for the sequence extraction was 'abort_check', whose FIRST occurrence is the function DEFINITION hundreds of lines above the sequence - between() resolved an end BEFORE the start and silently extracted NOTHING, so the first run of U measured an empty block. Anchored on a line that occurs once, and the reason is a comment at the anchor.
  - section V needed stubs that section T had unset.
  Also corrected: I wrote 69/69 into the staging table before re-reading the total. The measured number is 68/68 and the document now says 68.

PROCESS DISPOSITION: four pwsh (argument receiver x3 forms, payload fixture harness); U and V start no process at all - the clock is a shell variable and the gate runs under stubs. No host query, no elevation, no PREP launch, no field execution, nothing signalled. Scratch dirs removed on exit.

Driver still frozen at 2fd7d689 and holding for todlando's targeted verdict.