mind: land the autonerves-floor task state, one new prompt and one re-park

[Jammy2211]

Mind-state only — no library code. Records the work from one session across three tasks.

What's here

1. bump-autonerves-floor — shipped and closed out

Prompt advanced draft/ → active/ → complete/, with the completion record at complete/2026/08/bump-autonerves-floor.md and the original prompt folded in. Tracked as PyAutoArray#482, shipped as PyAutoArray#483 (merged 0f75c3d).

The record keeps two findings worth not re-deriving:

• The prompt was filed 2026-08-22 blocked on "the PyAutoNerves release", and autonerves 2026.8.23.1 was uploaded 2026-08-23T00:41Z — hours before the run. A prompt parked on an external release is worth re-checking on sight rather than trusting its Status:.
• The prompt framed converting the duplicated SMALLDAT literal as an optional cleanup. It wasn't optional in either direction: the comment justifying the duplication names the floor, so bumping the floor made it factually false. Kept the duplication on a decoupling rationale; a floor constrains resolution only, and an editable checkout can still present a pre-stamp autonerves.

2. untrack-generated-fits-test-artifacts — new prompt

draft/maintenance/libraries/. autoarray's suite writes test output into 13 tracked paths under output_test/, so any autonerves FITS-writer change dirties the tree and forces an unrelated binary into the diff. Surfaced when PyAutoNerves#155's header-comment fix silently modified a committed fixture during PyAutoArray#483.

The convention already exists and was applied inconsistently — .gitignore names two individual generated files and misses the other thirteen. Body carries the full inventory, the writers, and both candidate fixes (tmp_path preferred).

Filed via /intake, which classified low-confidence to triage/; re-homed by hand.

3. point_source_json_datasets_record_no_regime — re-parked, not started

That prompt's step 1 gates the work on two expiring facts and says to re-park rather than build if they still hold. Both re-verified 2026-08-23 and both still hold:

• PyAutoLens#480 still open — created 2026-04-28, updated_at identical to created_at, no assignees, no closing PRs.
• weak/simple still regime-invariant — scripts/weak/simulator.py calls via_tracer_from over an explicit 1500-galaxy annulus and contains no os.environ read at all.

So nothing was built, no issue opened, no branch cut. Added a dated re-check log so the next attempt doesn't repeat the verification, plus two path corrections that cost a search: the no_run.yaml exclusion lives in autolens_workspace, not autolens_workspace_test (whose copy has no multiple_sources entry), and it skips two scripts — simulator and modeling — not one. The original :41-42 line reference had already drifted, so the log matches on entry text instead.

Checks

scripts/lifecycle.py check → OK. dashboard.md / dashboard.html regenerated via pyauto-brain intake --apply dashboard (150 prompts).

---

Generated by Claude Code