FIELD PLAN SAVED AND PINNED, as instructed. Ceiling rev 3 is in it verbatim.

  path:   C:\Users\decid\Documents\projects\spt-core\docs\304-W2-FIELD-PLAN.md
  commit: 2e611cafe63ff8db4134dac7f3074edcb166d6b9  (branch main, UNPUSHED)
  blob:   05426bb20d62add2f8b0fad685dc92df7481e715
  sha256: a99a1a58c95e6d5cdb42f1e9ef788b12a66c67d141064eaa5f7193c89d124ba5
  parent: 5d6c6315, tree clean after the commit, one file added and nothing else touched

Read it directly rather than through me.

WHY docs/ AND NOT .spt/preserved/: .spt/ is gitignored at .gitignore:52 and carries zero tracked
files, so nothing under it can take a commit pin. docs/ is where the project's other RCA, verdict
and plan documents already live, and it gives you a real sha to name.

WHAT IS IN IT, beyond the ceiling:
  1. The enforcement string change and the new ENFORCEMENT_REPRESENTATION_FAULT surface, with the
     reason a wrapper-only classifier inverts rather than merely under-counts.
  2. The rollout-state finding and your ruling on it: keep the path, no reset, no deletion, no
     fresh attempt-specific path; fresh isolated home and explicit --home mandatory.
  3. The provenance gap, with YOUR close rather than my rejected one -- pre-build hash and mtime as
     EVIDENCE not a mandatory inequality, the known-old hash rejected by name, subject HEAD/blob/
     clean-tree reasserted immediately before the producer as boundary observations, and the chain
     described in your four terms.
  4. The four outcomes kept separate, with the attempt-2 statuses recorded as UNEXPLAINED and the
     ValueMap 0..25 / absent-Values-qualifier rule against ever asserting a name for a code.
  5. Ceiling revisions 1 and 2 retained as superseded history with what each got wrong, per your
     instruction not to delete them.

Section 6 records what the plan does NOT do: no reachability probe proposed, build estimate stated
as UNKNOWN and not to be borrowed from a debug figure or a CI wall, and driver internals / fixture
controls / staged helper manifest named as hertz's, not duplicated here.

traceable-reqs check exits 0 on the tree with this commit in it. I minted no REQ and activated no
stage: the document is a plan, not requirement evidence, and tagging it would have claimed coverage
it does not carry.

Ownership understood and not crossed: hertz owns the integrated driver successor and the fixture
controls. I have made no driver edits and written no duplicate controls.

HOLDING THIS LANE. No build, no field run, no probe. I note the window is released but NOT
reassigned, and that admission needs an explicit grant plus fresh preflight evidence -- I am
treating the released window as no grant at all.