[Jammy2211]

Phase 1 of the cortex-checkin epic (ledger PyAutoMind/draft/maintenance/pyautocortex/cortex_checkin_epic.md), closing PyAutoCortex#9. Deletion only, plus the one decoupling the phase-2 door needs. Every historical record stays readable.

The 2026-09-03 assessment found the core right-sized and the review-slot apparatus unused: 0 slots opened by the conductor, 0 rulings from a packet, 0 partial reviews, review-minutes-actual never filled — all 22 rulings were reached in a live session. Gate grading saw 2 gated refs and flipped 0. rule --also was never used. Lane: read local-dev on 32 of 32 phases. The restricted-YAML parser's own PyYAML-parity test proved it redundant.

What went

• Gate grading. .github/workflows/gates_grade.yml (the daily cron), GATE_JQ / _http_gate / fetch_gate_states / grade_gate and the --grade/--write half of gates_report, the Gates-cleared: / Gate-override: header keys and the three check invariants over them, and move --override. Kept: the plain Gates: header, the gated state, GATE_REF_RE / gate_url, and the read-only gates listing — which now also prints the URL each ref resolves to, because a human opening them is the whole mechanism. gated → ready is now cortex.py move <phase> ready, no flag.
• rule --also — flag, fan-out, tests, docs.
• Lane: — the schema, the check invariant, the new template, and one mechanical sweep over the 32 phase files and the fixtures. The fact it recorded (every phase runs on the laptop) now lives once, in AGENTS.md.
• The restricted-YAML parser. parse_projects is yaml.safe_load plus the same field validation (required/optional field sets, status/partition enums, path shape, sync_verbs bare words). PyYAML is declared in cortex_check.yml — moved before the check step, not only before the tests — and in a new README requirements line.
• The batch record + packet schema. batches/packets/ (TEMPLATE, AGENTS, three HTML pages), batch_problems / _review_section_problems / _record_members / _record_values and their grammars, the REFERENCE.md schema section, and the slot/shift/review-minute/partial-review vocabulary in batches/AGENTS.md and batches/reviews/AGENTS.md.

What stayed, deliberately

batches/2026-*.md and batches/reviews/*.md are kept as read-only history — 13 rulings cite the human's verbatim words in them. Both AGENTS.md files now say exactly that, and scripts/ledger_merge.py puts batches/ in APPEND_ONLY_DIRS beside rulings/: history, never modified, only added. (That is the answer to the prompt's "state which".) A ruling's Batch: is now optional-historical — where present it must still resolve to a batches/<slot>.md; nothing writes a new one.

projects.yaml and rulings/ are untouched — no projects.yaml change was needed for the PyYAML swap.

API Changes

scripts/cortex.py is a repo-local script, not a published API, but its CLI surface changed:

• gates loses --grade and --write; it is read-only and offline.
• rule loses --also.
• move loses --override; gated → ready needs no flag, and ready → gated is now a hand edit rather than a refusal pointing at the retired grader.
• parse_projects(text) keeps its (rows, problems) signature; _strip_comment/_scalar are gone. cortex.py now needs PyYAML — its one dependency.
• Deleted module-level names other tools might import: GATE_JQ, _http_gate, fetch_gate_states, grade_gate, batch_problems, _record_members, _record_values, MEMBER_RE, SLOT_RE, REVIEW_PART_RE, LANE, HEALTHS, GATES_SETTLED_STATES, GRADED_STATES. The PyAutoBrain side of this epic (PyAutoBrain PR, same branch) is the only consumer and is updated with it.

Verification

• python3 scripts/cortex.py check → OK.
• pytest -q → 115 passed (134 before; the 19 fewer are the tests of the deleted things).
• The dashboard is regenerated from inside this checkout against this branch's Brain worktree; dashboard --check is current. It loses the batch status box and the two blurbs that named the retired grader.
• Schema decision 58 appended to docs/schema_decisions.md with the measurements above.

Left alone

rulings/AGENTS.md still documents rule --also in one sentence. rulings/ is append-only and was declared out of bounds for this phase; it is a one-line doc fix for whoever next touches that file.

Merge order: PyAutoBrain#348 first, then this one. They are one change on one branch name, and the dependency runs that way: dashboard_refresh.yml here renders through PyAutoBrain main, so until #348 lands, this PR's refresh check runs the old conductor against the new script and fails on the MEMBER_RE this PR deletes. The other direction is safe — the new conductor works against either script. check (cortex.py check + pytest) is green on the branch's own tree.

🤖 Generated with Claude Code

https://claude.ai/code/session_013HsZA1ufn3msgPiDFxEXa6

[Jammy2211]

Expected red until the Brain PR merges. The refresh check runs dashboard --check through PyAutoBrain's cortex conductor, checked out at main — that copy still calls mod.MEMBER_RE, which this PR deletes. The workflow's own comment names this case ("a Brain/Cortex version skew"). PyAutoBrain#348 is the other half of this change on the same branch name; merge it first and re-run refresh here.

The check job (cortex.py check + pytest) is green on the branch's own tree.

Follow-up commit beec64c declares pyyaml in dashboard_refresh.yml and ledger_merge.yml too — both run cortex.py, which now reads projects.yaml with PyYAML.