CORRECTION LANDED. SUCCESSOR PIN:

  commit: f1c400f9a20e711829d799a4ce4b0aea96a4b7ed   (branch main, UNPUSHED)
  blob:   9af0823840ff7bc1771135c9de4afe354c7e3186
  sha256: ebd0bf97d573cc6f797dc4606283ff4989cc8e218cef5b3ed39e7524e8e84f17
  parent: 2e611cafe63ff8db4134dac7f3074edcb166d6b9
  path:   C:\Users\decid\Documents\projects\spt-core\docs\304-W2-FIELD-PLAN.md
  tree clean, one file touched, traceable-reqs check exit 0

You were right and the error was worse than a loose sentence. I had taken THE OLD DRIVER'S COMMENT
-- the very interpretation the successor exists to remove -- and written it into the plan as the
MEANING of the string, inside the paragraph describing the defect. The document would then have been
the authority the driver quotes, carrying the defect forward under my name.

Section 1 now separates three claims that I had run together:

  - Generic LAN_FIREWALL_UNVERIFIED establishes ONLY that verify() did not return Ok(true). Many
    distinct causes arrive under that one wrapper and it distinguishes none of them. The
    absent-or-misspelled gloss is now labelled explicitly as the old driver's reading and as the
    defect under correction, not as a meaning.
  - ONLY the specific enforcement-code refusal establishes that decide() passed pair_satisfied_by,
    because the check order puts pair_satisfied_by first and reaching the enforcement arm requires
    it to have matched. That is what makes it the representation-accepted discriminator.
  - A REPRESENTATION FAULT ESTABLISHES NO SUCH THING. The query failed, so decide()'s verdict is not
    in evidence at all. Neither representation-accepted nor pair-absent. My previous text had
    grouped it with the enforcement refusal as though both proved the pair matched -- the same class
    of error as the one being fixed, two paragraphs after describing it.

Section 4 follows so the document is self-consistent: the enforcement-codes outcome has THREE
states, not two, and "not in evidence" is reported as its own value rather than collapsed into
accepted or refused. That is the only other line I touched.

Unchanged: the ceiling, the provenance requirements, the rollout-state ruling, section 6.

TRACEABILITY, reported by me as mine to report: traceable-reqs check exits 0 on the tree with
f1c400f9 in it. I minted no REQ and activated no stage for this document.

Hertz: the same distinction is what the integrated classifier has to express -- three outcomes under
one wrapper, and the fault is not a decide() verdict at all.

Holding the lane. No execution grant issued, none assumed; the released window is not a grant.