Settle writ reconcile's surface, and stop teaching writ status - #773

Closes #772

Two plans described the same command differently. One recorded flags it will not have, and both still
taught writ status — a command being retired, not modernized. Documents only.

The settled surface

writ reconcile produces a report. It has no command flags. It takes the globals — --output /
-o, --filter, --jq, --store, --dry-run — and must utilize all of them faithfully. The selector
between fix, diff and summary behaviors is deferred to the reconciliation epic and is not designed
here.

The flags go

762:51 recorded the shape as writ reconcile [--drift] [--fix] [--json] [<project>...]. None of the
three survives: --json is replaced by the shared -o, and the other two are the deferred selector
under earlier names.

762:96 and 5.1:12 both said 5.1's "there is no --fix" was retired as a stale claim. That
overstates it — reconcile will repair, but there is no --fix today and the surface is undecided. Both
are corrected.

A caution, recorded where the selector will be designed

Two of the three candidate views are already projections of the one document:

Those are stage 2 of the pipeline #740 landed, and --jq and --filter are already reserved. Adding
--diff and --history flags would create a second way to select a subset of the same document, and the
two will drift. Only the repair half is genuinely new, because it is a mutation rather than a projection
— and --dry-run already owns the mutation axis.

Terminology corrected wherever it appears

Eight sites in 740. Six are the command's name and change outright. Two are not terminology:

• :212 is a path. Recorded now as cmd/writ/writ/status/report.go (reconcile/ under The writ lifecycle surface: reconcile, scopes, and the vocabulary settlement #762), so it
  is findable before and after the rename.
• :27 was a false claim about ordering, rewritten rather than renamed. The work lands as
  writ reconcile, so this epic's writ phase and The writ lifecycle surface: reconcile, scopes, and the vocabulary settlement #762's phase 2 are one piece of work, not two adjacent
  ones.

The measurements keep their date, not their vocabulary: lines 231–242 record what was found on
2026-08-30, and one dated note says so, rather than seven caveats through the text.

Test plan

• ./scripts/Test-GuideFrontmatter.sh — green
• No plan records a flag writ reconcile will not have; the surviving --drift/--fix/--json
  mentions are historical quotation or the explanation that they do not survive
• The globals commitment is stated in 762, which owns the command
• The only writ status left in 740 is line 231's dated note saying that is what it was called when
  measured
• :212 names both the path today and where The writ lifecycle surface: reconcile, scopes, and the vocabulary settlement #762 moves it
• Not covered: whether the deferred selector is designed correctly, because it is not designed

Open question, carried in the plan

Does the repair half become a mode of reconcile, or stay as the named commands each Entry already
points at in its Repair field? writ-deploy-family.md:136 chose the latter deliberately — "no
--fix: the repair for each finding is named instead" — and that is what the tree implements today.
Bigger than flag syntax; belongs to the reconciliation epic.

https://claude.ai/code/session_01SqQVFZz7AdiLZkGoAcUqRW