[derek73]

Part of #762. C1's ## Standing block landed on 2026-10-19 (#743), and the five decisions files split out the same day (C13, C14, P8, P9, T5) had none. #762 asks whether to roll Standing out to the old logs. New files don't carry that cost: a file's first entry is the cheap moment to write its block. So this PR makes new files carry one and leaves the old ones to #762.

What changes

• decisions/README.md and root AGENTS.md: a new decisions file opens with ## Standing and ## Log. The rule dates from C1's block landing; C12 and vocabulary-model were created earlier that day and are exempt.
• A ratchet, not prose. tests/v2/decisions_without_standing.json lists the 88 older files allowed to lack a block. test_a_decisions_file_off_the_list_opens_with_standing fails on:
  • a file off the list with no block. The heading must sit on a line of its own; the words quoted in prose don't count;
  • a listed file that has gained a block (it should leave the list);
  • a listed key with no file;
  • a listed key outside STANDING_EXEMPT_ORIGINAL, the 88 keys the list held when the rule landed (its size is asserted). So a new file can't join the list, alone or swapped in for one that leaves;
  • a heading that misspells ## Standing: a trailing space, another level, or indented. Otherwise a misspelled heading would switch every check off for that file.
• Placement: standing_findings reports a ## Log before the Standing block as a finding. A block after the log used to crash it.
• Standing blocks for C13, C14, P8, P9 and T5, each line pointing at the entry that decided it. Writing them, and then reviewing them, turned up stale or overstated facts:
  • P9's note on Doe, Jane PhD vd Ma: Should a name word after a credential join the suffix? John Smith PhD Jones reads middle Smith PhD #602 reversed the silent reading.
  • P8's subset default: it also holds i, since Josep Carod i Rovira reads family Rovira — Catalan and Polish link surnames with i, which is not a conjunction #397.
  • P8's "joins at every length": it joins only where P3's word count lets it, and the every-length claim applies to the initial tag.
  • C14 omitted its own exception for a run that a credential opens.
• A decision of another rule's that a log is the record of stays a bullet naming its rule: C13's line for C2 and P9's for S5. One recorded under another key is an Elsewhere: line.

Review

Run in this order:

• Two design-docs-reviewer rounds before the PR opened.
• A three-reviewer PR review: code, test coverage and comments.
• Two further rounds on its fix commits.

Every finding is fixed. The last round's one leftover, an indented heading, is fixed in a3c2265, which no reviewer has re-checked. Each new test row was run against the commit before it and fails there.

Outside this PR: P9's dated log claims DOE, JANE PHD VD PROF. MA reads middle 'VD PROF.', but it now reads family 'VD DOE', suffix 'PHD, MA'. That's a measurement in a dated entry, not a Standing line, so it's left for whoever revisits P9.

Test plan

• uv run pytest, after merging master: 22,063 passed, 123 skipped, 3 xfailed
• uv run mypy, uv run ruff check
• Negative controls: the real-tree test fails on origin/master's five files, and every synthetic row fails on the commit before the check it pins

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.05%. Comparing base (8275e8e) to head (9a76943).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #776   +/-   ##
=======================================
  Coverage   99.05%   99.05%           
=======================================
  Files          47       47           
  Lines        4642     4642           
=======================================
  Hits         4598     4598           
  Misses         44       44           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.