liam -> doyle (cc todlando, hertz). STOP — THE HASH YOU JUST CLEARED IS THE BROKEN SCRIPT. A6c-2 RAN AND IS VOID. Nothing was measured and nothing was written. Re-filed below; NOT running until you clear the new hash.
WHAT HAPPENED, 03:30:11Z-03:30:23Z: I ran the filed file (19ccce6b...3ec6b) and BOTH invocations died at the parser before executing a single statement:
  RUN 1  209 ms  exit=1  "Missing closing '}' in statement block or type definition. ParserError ... MissingEndCurlyBrace"
  RUN 2  166 ms  exit=1  same
  Censuses: PRE [] count 0, after run 1 [] count 0, after run 2 [] count 0, FINAL [] count 0.
  So the store never changed, the arm measured nothing, and the 209/166 ms are parse-failure times. The "delta run2 - run1 = -43 ms" my own output printed is meaningless and I am striking it before anyone quotes it.
MY BUILD ERROR, named exactly: I assembled the script by taking the first 32 lines of the A6c file as the wrapper+preamble. The OWNERSHIP preamble's Remove-Owned function runs to line 34. Line 32 is `$rule | Remove-NetFirewallRule -ErrorAction Stop`, so I cut the function inside its inner foreach and dropped its two closing braces. Brace balance of the filed file: 14 open, 12 close. The A6c file that ran fine is 12/12.
  It is a truncation-by-line-count error on a file whose structure I had already read, and the tell was available before I ran it: a brace count takes one command and I did not take it.
CORRECTED FILE, re-filed: same path scratchpad/a6c2-reconcile.ps1, now built from the first 34 lines, 57 lines total, brace balance 14 OPEN / 14 CLOSE.
  NEW SHA256: 062a87be7af2a245a5bb2e4a716f6c4fa35facbdb7aae0bf9eb99e743d421317
  THE ONLY DELTA from the bytes you reviewed is the two missing closing braces of Remove-Owned. Body, order, deviations, rule names, group, run plan and pre-registration are unchanged.
I AM NOT RUNNING IT ON YOUR EXISTING CLEARANCE. You cleared 19ccce6b explicitly by hash, and my own filing said a mismatch at run time voids the run. I could argue the repair is within scope — it restores the artifact to what the source actually says and changes nothing you reviewed — but that is me deciding my own permission, and I have refused to do that all night. One word from you and it runs.
SEQUENCE IMPACT, so you can weigh the delay: you gated hertz's capture on my A6c-2 stamp and todlando's dial on hertz's post, so both are behind this. If you would rather not spend a round trip, "run 062a87be" is the whole reply.
PRE-REGISTRATION UNCHANGED: run 1 near 1.6-1.8 s against an empty owned set; run 2 higher by two Assert-Owned walks and two removes; run 2 at or below run 1 would be a finding against my own expectation and would extend todlando's per-call result to the owned-set walk.