docs(s001): D4 in PROOF still asked the owner to accept what they had accepted

The deviations section said D4 was 'flagged for the owner to accept or split out'. It was
accepted on 2026-09-06 as a packaging prerequisite. Recorded with the principle attached:
post the Question, then proceed in parallel with the work that does not depend on the answer.

mix gate exit=0, scripts/plan_check.sh exit=0.

Signed-off-by: Ayla Croft <aylacroft@proton.me>

Author: HackTuah

slices/001-packaging-spike/PROOF.md

@@ -555,10 +555,12 @@ them. The four that most change what a reader should expect:
    compares for equality, not a range.
 4. **The slice-000 Credo check had to move out of `lib/`.** `credo` is `only: [:dev, :test]` and
    `lib/` compiles in every environment, so nothing compiled under `MIX_ENV=prod` at all. This
-   is a change outside packaging by the letter of the plan's constraint, and it is flagged as
-   such in `NOTES.md` D4 for the owner to accept or split out; the judgement made was that a
-   tree that cannot compile in `:prod` cannot be packaged, so it is a prerequisite rather than
-   a widening of scope.
+   is a change outside packaging by the letter of the plan's constraint. **The owner accepted
+   it on 2026-09-06 as a packaging prerequisite** — a spike that cannot compile for prod has
+   not spiked — with a principle attached that is a correction to how I worked rather than to
+   the code: *post the Question, then proceed in parallel with the work that does not depend on
+   the answer; commit first and ask second is the wrong order even when the call is right.* I
+   did the second thing, and raised it at G3 with eight commits already standing on it.
 
 `SLICE.md`'s acceptance criteria were themselves retagged at G1 (line 2, commit `afca47d`):
 seven criteria became nine, AC4 moved `[auto]` → `[manual]`, and AC5 and AC6 were each split