[braboj]

core/quality.md, which every chain loads, gains a "Static checks" section. Until now no template asked for a type checker, and the linter rule loaded only when adding a check.

Where	What changed
templates/core/quality.md	New "Static checks": strict type checker, annotated public signatures, linter and formatter, run before commit and in CI
docs/architecture/08-crosscutting-concepts.md	Generated context-cost table: +67 words a root, regenerated

Why, from the efficacy benchmark, round 5e:

Arms	Type errors per KLOC
Root names mypy --strict from the project brief	4 to 13
Templates alone	63 to 89
No file	99

• Stack-neutral: no language or tool named, so each rule passes the waist test (ADR-050)
• py tests/run.py all: 3 suites, 0 failed

Closes #2439

Also: this repository runs the three checks on itself. Until #2441 cleans the tooling (P1, before v3.0), the check reports its findings without failing.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DpT5xSQaG8RrSCyeihaQsm

[braboj]

Codex round 1: one finding, fixed in the two commits above.

Sev	Finding	Fix
P2	The new static-check rules had no runnable check	A fenced check with placeholders and a pass condition; its disposition runs it here

Fixing it showed this repository broke its own new rule (no type checker, no formatter). Owner's choice: keep the rules MUST and adopt them here.

• mypy strict and the ruff formatter, under a freeze tools/freeze_static.py writes into pyproject.toml from their own output
• pyflakes stays the linter
• CI installs the three (minor-range pins) and fails on a stale freeze
• Clearing the freeze: Fix the tooling's type errors and unformatted files before v3.0 #2441 (38 modules, 1,384 errors; 40 files)
• py tests/run.py all: 3 suites, 0 failed; every other CI gate passes locally

[braboj]

Codex round 2: three findings.

Sev	Finding	What became of it
P2	Regenerating the freeze could add a newly failing file	Moot: the freeze is dropped
P2	A failed checker run read as no findings	Moot: the freeze is dropped
P2	Local setup did not list mypy, ruff and pyflakes	Fixed: CONTRIBUTING.md Setup lists them with versions

Owner's choice: no freeze. The static checks read the whole tooling, so conformance's static check fails until #2441 lands; #2441 is P1 in the v3.0 milestone.

[braboj]

Codex round 3: one finding, P1, the static check fails on the existing tooling. By design: the owner chose no freeze and accepted conformance failing until #2441 (P1, v3.0 milestone) cleans the tooling. Not fixed here.

[braboj]

Owner's choice after round 3: the static check reports, and does not fail, until #2441 cleans the tooling. main stays green.

• Disposition: expect: [READ], so every finding line is a reading in the conformance report, never scored
• Each command ends || true, as the runner fails any nonzero exit
• Fix the tooling's type errors and unformatted files before v3.0 #2441 restores the gate: expect: [] and no || true
• py tests/run.py all: 3 suites, 0 failed; the report carries the findings

[braboj]

Codex round 4: one finding, fixed in a87085f.

Severity	Finding	What became of it
P2	The report-only static check read as PASS over real type errors, and added nothing to "awaiting a reading"	Fixed

The fix:

• The check is now manual, and its title is in reading-budget.txt
• A run says "1 reading still to read, so this run is not clean yet"
• The runner refuses any entry that ignores an exit with || true unless it is manual
• Once the tooling is clean the check prints nothing and fails, which prompts Fix the tooling's type errors and unformatted files before v3.0 #2441 to restore the gate

Before and after, py tests/run.py conformance "<type-check-command>":

before: 1 passed  ...  0 awaiting a reading
after:  0 passed  ...  1 awaiting a reading   (budgeted)

Without the disposition change, the runner's new rule fails the run.

py tests/run.py all: 3 suites, 0 failed. Codex round 5 is next.

[braboj]

Codex round 5: one finding, fixed in 29d0e52.

Severity	Finding	What became of it
P2	|| true also hid a checker that could not run (bad config, crash), so it read as a budgeted finding	Fixed

The fix:

• Each static command ends in FINDINGS_ONLY instead of || true
• Exit 1 (findings) is carried past, so the check stays a budgeted reading
• Any other exit stops the block with that code, so the check fails
• The runner now refuses a substitution that uses || true

Before and after, a block whose first checker crashes:

before: (exit 2) || true; (exit 1) || true      -> exit 0, budgeted REVIEW
after:  (exit 2) FINDINGS_ONLY; (exit 1) ...    -> exit 2, check fails

Without the change to the entry, the runner's new rule fails the run.

py tests/run.py all: 3 suites, 0 failed. Codex round 6 is next.

[braboj]

Codex round 6: one finding, fixed in 28a3d8d.

Severity	Finding	What became of it
P2	A checker missing from the interpreter exits 1 ("No module named ..."), the same as findings, so it read as a budgeted finding	Fixed

The fix:

• Each command now comes from findings_only(python, module, args)
• It imports the checker first, and exits 2 if that fails
• The runner refuses a findings-exit command with no import first

Before and after, with the checker not installed:

before: python -m nosuchchecker x  || {...}   -> exit 0, budgeted REVIEW
after:  python -c "import nosuchchecker" || exit 2; ...   -> exit 2, fails

Without the change to the entry, the runner's new rule fails the run.

py tests/run.py all: 3 suites, 0 failed. Codex round 7 is next.

[braboj]

Codex round 7: no findings. Smoke, conformance, doc sync and the quality ratchet pass; three readings wait as intended, the static checks among them. The review loop is done; the PR waits on the owner.