[derek73]

Follow-up to the S2 (#682) and C1 (#728) splits: a critique of how rules/, decisions/ and mechanisms.md are serving us found that the splits fixed file granularity but not the clause-by-clause growth that caused them, and that decisions files make the current answer a replay of the whole log. This PR lands the first three remedies.

What changes

• docs/design/AGENTS.md: one rule per piece of evidence. A rule states one signal in the written name and what it decides; a decision weighing several signals is a hub with a one-line map, and new evidence is a new rule plus a map line, never a new clause. This principle was applied by both splits but written only in their commit messages.
• Review axis 12: name the signal each new clause reads. A clause reading a signal the rule did not already read is reported as a structural finding (not style), which the reviewer's style ban had been hiding.
• decisions/README.md: append-only for decisions, not for text. Four permitted edits replace "append-only in spirit": a changed decision is a new entry (SUPERSEDES: / SUPERSEDED by, with a partial form); a factual error is corrected in place with a note; superseded entries and shipped measurement paragraphs compact on a stated trigger; and master's per-word-fact-to-TOML exception (Describe each word by its readings: per-word TOML vocabulary compiled to today's sets (phase 1) #737), folded in as the fourth when master was merged. New entries open with their verdict.
• ## Standing block, trialled on decisions/C1. An index of the decisions in force, each line pointing at the entry that decided it, plus Not standing: and Open: lines; the entries follow under ## Log. Writing it found the Have C1, S2 or P6 grown by patching the way M2 had, and should they be rethought? #611 entry's "the rethinks are filed, not done" overtaken since Rethink the suffix-comma rule (C1): decide the comma's structure once, after the words are tagged #613 and Rethink the credential-run rule (S2): read a name's trailing run once, over name units, and bind it before any join #614 shipped in 2.4.0.
• test_every_standing_line_points_at_a_live_entry (tests/v2/design_docs.py's standing_findings): every pointer must match an entry by exact date and first issue, and not one only SUPERSEDED by a later entry. Synthetic rows break each check; the review rounds' new rows each fail on the earlier detector. It cannot see a decision in force that Standing omits — that half stays with the reviewer.
• decisions/C1's Rethink the suffix-comma rule (C1): decide the comma's structure once, after the words are tagged #613 entry compacted: WHAT MOVES, COST and the review round's clock figures cut to one line each (Rethink the suffix-comma rule (C1): decide the comma's structure once, after the words are tagged #613 shipped in 2.4.0). The reviewer grepped the tree for every removed figure; nothing quotes them.

Review

Two design-docs-reviewer rounds: nine findings, then five on the fixes, all addressed in 0934efa (the main one: the detector read an entry ANNOUNCING a supersession as superseded). Not done here, by choice: the Signal: field in the rule grammar, a recorded per-rule size budget, and rolling Standing out past C1.

Test plan

• uv run pytest — 21,704 passed, 123 skipped, 3 xfailed (after merging master)
• uv run mypy, uv run ruff check
• Negative controls: each new synthetic row fails on the previous detector

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.05%. Comparing base (ef03011) to head (2887570).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #743   +/-   ##
=======================================
  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.