feat: cortex sync-runs — run status from a campaign digest (hpc-campaign p5b) - #83

Summary

hpc-campaign phase 5b (#82). Adds cortex sync-runs <project> --digest <path> [--dry-run], which updates a project's run lines from a PyAutoFit campaign digest (digest.json + tasks.jsonl). This is new decision 64; the epic ledger calls it "decision 62", but 62 and 63 were already taken.

• What it changes: a run line moves from open to running once its tasks have started, with a run log entry. It moves to done, with state counts and wall time, once every task index the run names has finished. It is marked failed only when none completed.
• What it never touches: result, lesson and ## Now. It never adds or removes run lines. It is idempotent, and --dry-run writes nothing.
• Pending array slices: the scheduler reports these as compound rows (e.g. 3002_[2-3]). They are expanded into their indices and keep the run open.
• Matching: by the SLURM array job id plus the indices the run names.
• Writes: all edits are computed and validated first, then written once, so a refusal leaves the file untouched.
• Command contract: exits 0 on success; exits 1 with empty stdout on a malformed digest; the first stdout line is sync-runs: <project>: <n> changed.
• Ledger fields: projects.yaml gains optional campaign_file and digest, validated by cortex check.
• Docs: REFERENCE.md gets the boundary paragraph: run status is a cluster fact; results and lessons are the human's.

Heart RED override (development only)

Heart was RED at the door on 2026-10-30:

• release validation FAILED (stage integrate), and the Release Integrate workflow failed;
• PyAutoNerves test_fitsable.py::test__output_to_fits red;
• unit-test timing: 1 regression (>3× baseline), 14 slow;
• autogalaxy red; autolens_profiling red;
• release validation NOT release_ready;
• worktree drift: 1 dirty.

The human authorized the Human override for Heart RED (development only) live for hpc-campaign phase 5 ("Override + approve"). This PR doesn't touch any RED item. Merging needs green CI and a separate /prm.

Test plan

• tests/ 91 passed (21 new); cortex check OK on the real ledgers.

• Independent review: FINDINGS, fixed in d590eac:

  • HIGH: a run was marked done while compound pending slices remained;
  • MEDIUM: a failure partway left earlier runs written;
  • LOW: newest-first log order was not checked for the new entries.

  Each has a regression test. The fixes have not been re-reviewed (time-boxed session).

• Dry run on a temporary copy against the real DR1 digest: 58 euclid_dr1 runs would close. Applying that to the live ledger needs the human's ask.

Refs #82

🤖 Generated with Claude Code