[Jammy2211]

Phase 3 (last) of the cortex-checkin epic — PyAutoCortex#12.

Stacked on #350 / PyAutoCortex#11 — retargets to main as the stack merges. Base is feature/cortex-checkin-p2-the-door; that branch is itself based on feature/cortex-checkin-p1-shed-review-slot. Merge order for the epic's six PRs: PyAutoBrain#348 → PyAutoCortex#10 → PyAutoBrain#350 → PyAutoCortex#11 → this PR → PyAutoCortex#13.

What this finishes

The original request: "tell PyAutoCortex to download all results, give me a summary of where every project is in terms of runs and what folders I should check out, and give me starting claude prompts to continue work on any individual project." Phase 2 built the door. This is the summary and the last two prompts.

Before: every dashboard section and census were keyed by phase state, the Projects table omitted local_path / mirror / ral_root, and each phase's ## Where to look bullets — the folder answer itself — were parsed only to locate artefacts and never rendered.

## By project

One tree, three renderings, one builder:

Where	What it adds
## By project on dashboard.md / dashboard.html	copy buttons
pyauto-brain cortex census --by-project	nothing — the tree from the checkout alone, no pull, no cluster
the closing summary of cortex checkin --apply	the pull result and the six-leg health (replaces phase 2's minimal digest)

Per project: its three folders, its phase counts by state, then every phase still open — awaiting a ruling, still out there, ready to submit, gated, planned — each with its state, its ## Where to look bullets verbatim, and a copy chip per prompt its state carries. accepted / rerun / dropped are history: counted, not listed. A project gets a block when its row is status: active or it still holds an open phase; every other project folds to one line.

Where the section sits, and why

First — immediately after the counts table, above ## Awaiting ruling. The prompt offered "above the state sections" or "directly below the header"; on this page those are the same place, and the reason to take it is the phone. The state sections are cross-project lists: every row has to be read to find out whose project it is. This is the only section that answers "where is everything" a screen per project, and it opens with the folder paths — the thing the human came for. The counts table stays above it, so the urgency sections are still one tap away by anchor.

section_counts() is deliberately unchanged — board/_board.py scrapes the counts table by regex for the Brain board's strip, so ## By project has no count row of its own.

The two missing prompts

Both render as chips beside the existing rule prompt (the row keeps the payload it always had; these follow it as ↳ chips), and both appear in the by-project tree and in checkin's output.

_next_phase_payload — "the results are good", for pulled / awaiting-ruling. Example (--phase 5 is taken by a phase that has already run, so it offers the lowest free number):

The results for the PyAutoCortex phase phases/proj/four.md are good. Read its `## Witness` and the evidence under its `## Where to look`, draft the accept body for my approval, then file it and open the next phase:
python3 scripts/cortex.py rule phases/proj/four.md accept --body <file>
# phase 5 is taken (phases/proj/five.md, accepted) — the next free number is 10:
python3 scripts/cortex.py new proj <slug> --phase 10 --epic an-epic --title "<the tail only — `new` writes '<Project> — phase 10: ' itself>"
python3 scripts/cortex.py move phases/proj/<slug>.md ready
cd /s/proj && hpc/sync submit <script>

It prefills from the ledger: when the project already holds a phase numbered N+1 in planned or gated, that phase is opened (move … ready + the launch) rather than duplicated — epics.md points at phases/<project>/ for exactly this reason. The title rule is stated inline because cortex.py new writes # <Project> — phase N: in front of --title itself; passing the full headline doubles the prefix.

_rerun_payload — "run it again", for pulled / awaiting-ruling / running:

The PyAutoCortex phase phases/proj/four.md needs running again. Draft the rerun body — what came back, and what changes — for my approval, then file it and relaunch:
python3 scripts/cortex.py rule phases/proj/four.md rerun --body <file>
python3 scripts/cortex.py move phases/proj/four.md ready
cd /s/proj && hpc/sync submit <script>
python3 scripts/cortex.py move phases/proj/four.md submitted --run <jobid>   # one run per call; --after <run> chains the next

A rerun is a ruling first: rerun is the only verb that puts a phase back on the board. The payloads are short on purpose — they are pasted into a fresh chat, not run as scripts, and the <file> / <script> / <slug> / <jobid> placeholders are where the human's own reading lands.

phase_chips(r, c) is the single table saying which prompts a state carries, so a prompt cannot appear on the board and not in the door.

API Changes

Additive only — no signature broke.

• new: census --by-project; emit_by_project(), phase_chips(), project_groups(), by_project_keys(), project_paths(), project_counts(), _next_phase_payload(), _rerun_payload(), _where_bullets(), _submit_line().
• changed: census() gains a planned bucket; _phase_row() gains "where" (the bullets verbatim); project_digest()'s pull_line is now optional (census omits it); launch_payload() delegates its middle line to _submit_line(); SECTIONS gains by_project.
• unchanged on purpose: section_counts(), and therefore board/_board.py's counts strip.

Testing

tests/test_cortex_conductor.py: 67 passed (13 new) against the real Cortex fixture — the section leads the board, a project block names its three folders and its counts, the where-bullets render verbatim, history is counted not listed, census --by-project prints the same tree, a dormant project with nothing open folds to one line, a phase whose ## Where to look names nowhere still renders in all three views, all three branches of the accept-and-open prompt (planned sibling / no sibling / taken number), the rerun prompt's ordering, a project with no submit verb, the state→chips table, and the chips riding beside the rule prompt in md and html.

Full Brain suite: 889 passed, 2 failed — both tests/test_branch_sweep.py (test_an_unexpired_integration_branch_is_protected, test_delete_removes_only_the_expired_integration_branch), pre-existing on this stack and control-tested on the phase-2 base where they reproduce unchanged. Not touched.

Live: cortex census --by-project and cortex checkin --dry-run --skip-pull both run clean against the real Cortex tree; dashboard --check is current after regenerating (PyAutoCortex#13 carries the regenerated pages).

Heart

RED at ship time, not acknowledged by a human. Exact reasons from pyauto-heart readiness --json: red "release validation FAILED (stage integrate)"; yellow "PyAutoArray: open PR 11d old". Both are release-chain facts about other repos; neither PyAutoBrain nor PyAutoCortex is in the release chain, and this branch is a conductor renderer plus tests. PR opened, nothing merged — the merge decision needs the human and an ack (or a green Heart) before it.

🤖 Generated with Claude Code

https://claude.ai/code/session_013HsZA1ufn3msgPiDFxEXa6