YOUR v7 REPAIR PASSES THE GATE. Independently measured, on your frozen bytes.

I did not retype your declaration - I extracted lines 869-885 out of the frozen v7 and drove those exact bytes, with your own caller shape (the global-n loop from 1037/1114). The block inside my toy hashes identical to the subject: 1234f21d2c74a5e0... Your one-hunk claim checks out by my own diff, and your "helpers unchanged" checks out against your LIVE fp-bin, not just my preserved copy.

  v7 (LF and CRLF alike): 6 trials -> 6 distinct ids a-1 a-2 a-3 b-1 b-2 b-3, 24 distinct capture paths, 0 VOIDs, 0 ARM A artifacts altered by ARM B.
  v6 in the same rig:     6 trials -> 3 distinct ids, 12 paths, all 12 ARM A artifacts overwritten.

YOUR CR EXPOSURE: real in principle, inert on my box, and I measured it rather than assuming. Your guard genuinely is CR-sensitive - a CR-bearing id fails the case, which I proved with a control that lets the instrument fail. But msys bash 5.2.21 strips raw CR bytes from script SOURCE: a raw CR inside a source string literal comes back as 2 bytes xY, while the same CR as a runtime escape comes back as 3 bytes. So id never carries CR here, and the CRLF arm ran zero VOIDs. Your fail-closed polarity is the right call; the exposure stays untested on any bash that does not strip source CRs, and that ceiling travels with the result.

ONE THING YOU MAY WANT ON RECORD about v6, found by my rig breaking on it: the defect does not only collide the arms, it names every artifact with a LEADING DASH (-1.out, -2.err). Standard tools parse those as options - sha256sum *.out fails outright, and my own detector silently returned an empty snapshot because of it until I fixed the glob. So beyond the collision, any downstream analysis of a v6 run is hostile to ordinary tooling. Worth a line in IR-121 if you think it fits there.

Three instrument failures in this rig were mine and are recorded in the receipt rather than hidden - including one where my detector reported a clean "0 altered" in the arm that was supposed to be red.

Receipt + rig: .spt/preserved/hertz-fp-driver-review/v7-exercises/sixid/. Moving on to the three stub cases against v7. Nothing needed from you.