test(repo): guard prose that has an expiry date

[donislawdev]

What and why

Follow-up to #43. That PR cleaned up four stale comments; this one stops them coming back.

The important bit about #43 is that it was not a set of false claims. All four sentences were
true when they were written and were supposed to die when a stage landed - "NOT read by targeting
yet - that is 2c" was accurate right up to the moment chunk 2c shipped. Convention 5 covers claims
that are wrong; check_notes.py deliberately does not check prose at all. So nothing anywhere
enforced the expiry, and the sentences outlived the code they described by several PRs, telling
every later session the opposite of what the code actually did. Discipline alone produced four
stale sites in a single transition, which is the argument for a mechanical step rather than
another reminder.

So: prose that is only true until a stage lands now carries a PENDING(<id>) marker naming
that stage, and the ids still open live in exactly one place. Closing a stage is one deletion from
that set, which turns the guard red on every marker still pointing at it - so the prose is
corrected in the same commit as the code that outdated it, instead of whenever somebody happens to
re-read it.

What a reviewer should know:

• test_no_stale_pending_markers checks both directions, and the second one is the
  non-obvious half: a marker naming a closed stage fails (the prose beside it is now a lie), and
  an id in OPEN_PENDING that nothing references also fails (a leftover entry). That is why
  OPEN_PENDING is documented as not a roadmap - it is the set of ids that prose points at.
• Mutation-checked, not asserted (convention 5): a probe file carrying a marker for an
  unlisted stage, and an unreferenced id added to the set, were each confirmed to turn the test
  red with the offending file:line in the message. The probe was removed; git status --untracked-files=all is clean.
• The guard skips its own file, which holds the ids rather than pointing at them - the same
  idiom this file already uses for scrollable.py and crashlog.py. The <id> placeholder form
  does not match the pattern, so docs and changelogs can name the token without registering it.
• First real marker goes on the SocketEvent comment in socketwatch.py, which was itself a
  leftover of 2c ("carried for the connection log later"). proto / remote_ip / remote_port /
  outbound are genuinely still unconsumed, so the comment now says so plainly and the marker
  forces that decision to be revisited rather than quietly kept. Whoever closes it deletes one
  line from OPEN_PENDING and the guard makes sure the comment does not survive alone.
• Rejected, measured not guessed: scanning for the word "yet". 25 hits across beantester/
  and tests/, roughly 21 of them permanently true ("width 1 == not laid out yet", "no honest
  answer yet"). A guard with that false-positive rate gets switched off within a week, so the
  token is opt-in by design.
• Known coverage limit, stated on purpose: this catches marked prose. Unmarked "not yet"
  free text still gets through, and the guard cannot see PROJECT_NOTES.md at all because that
  file is not in this repo. Convention 44 and a new definition-of-done step (grep for the id when
  closing a stage) cover that half manually, and say so explicitly.

Checklist

• python -m pytest tests passes locally. (661 tests, 0 failures - one more than before, which
  is the new guard; on an elevated shell, so the two known non-admin failures are absent
  rather than excused. smoke_gui.py: OK.)
• New behaviour has tests (see tests/ for the style). - the change is a test, and it was
  mutation-checked in both directions rather than trusted.
• UI text goes through i18n keys, with both lang/en.json and lang/pl.json updated. -
  n/a: no UI text.
• User-facing changes noted in CHANGELOG.md; technical ones and new tests in
  CHANGELOG-INTERNAL.md, under [Unreleased]. - internal only (convention 39): a new test
  and one comment, nothing a tester can see.
• Commits follow Conventional Commits (type(scope): summary).
• No version bump - the owner closes a version via VERSION.txt.

🤖 Generated with Claude Code