mind: retire jax-likelihood-smoke-pins-stale — fixed before it was filed

[Jammy2211]

Overview

/start_dev draft/bug/autolens/jax_likelihood_smoke_pins_stale.md found no reproducible bug left in the prompt. Retired unstarted: no GitHub issue, no active.md entry, no development.

Verified state

autolens_workspace_test main 8be3d59, Smoke Tests run 32922456742 (2026-08-26 02:24 UTC), green on Python 3.12 and 3.13:

Running 22 listed scripts
  scripts/imaging/jax_likelihood/lp.py ...              PASS (9.8s)
  scripts/interferometer/jax_likelihood/rectangular.py  PASS (21.1s)
  scripts/interferometer/jax_likelihood/mge.py ...      PASS (12.6s)
  ... 22/22 PASS

The prompt's headline — "every workspace_test smoke sweep reads 20/24 until this lands" — is now 22/22.

Disposition of each claim

Prompt claim	Current state
imaging/jax_likelihood/lp.py pin -1.34797827e09 is double	Repinned to -6.74165366e08 in 197ce6e (autolens_workspace_test#257), the same day the prompt was filed
fb1aefe0b sits unmerged in an orphan worktree	Not orphaned — it is the head SHA of PyAutoLabs/PyAutoLens#700, merged 2026-08-17, two days before the prompt called it unmerged. 197ce6e is the paired re-pin the prompt asked for
interferometer/jax_likelihood/rectangular.py misses pin by 0.0124%	Pin -3164.286252 unchanged and green (re-verified in 99d63b3 incl. the TransformerNUFFT cross-check). The 08-19 miss was control-run environment drift, not a stale pin
interferometer/jax_likelihood/mge.py stale pin	Repinned by the same 197ce6e; a PositionsLH-doubling casualty, not independent drift
multi_dataset/jax_likelihood/mge.py stale pin	Never a pin problem. Out of the smoke gate since #262 (2026-08-22) for the XLA compile stall tracked in autolens_workspace_test#245; its pin is untouched since the multi → multi_dataset rename

Changes

• complete/2026/08/jax-likelihood-smoke-pins-stale.md — the completion record, with the original prompt folded in
• draft/bug/autolens/jax_likelihood_smoke_pins_stale.md — removed
• complete/index.md — regenerated (1143 records)
• dashboard.md / dashboard.html — regenerated; the prompt no longer renders as pickable backlog

lifecycle.py check and lifecycle.py index --check both report OK.

Findings kept in the record

• Adjudicate an "orphan" worktree from the remote, not from the worktree. A local worktree over a merged-and-deleted branch is indistinguishable from one holding real unlanded work. Resolving the SHA against the remote's merged PRs settles it in one call.
• Do not group symptoms into a "family" by folder. Three of the four scripts were one root cause (the 2x-last PositionsLH penalty). Acting on the "stale-pin family" framing would have widened tolerances on pins that were already correct.
• A pass/total target is not a stable goal. Two entries left smoke_tests.txt for timing reasons, so the sweep reads 22/22 without any pin having been touched for that.

Follow-ups

• Local debris only: ~/Code/PyAutoLabs-wt/positions-lh-penalty-accumulation is a stale worktree over a merged branch — a /repo_cleanup item, not code work.
• Latent, owned elsewhere: multi_dataset/jax_likelihood/mge.py's pin has never been verified against a post-PyAutoLens#700 library. Re-derive it as part of autolens_workspace_test#245's re-enable. Deliberately not filed as its own prompt — it has no standalone trigger.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Egf8PFc5UKJfVWY1jbBKGU

---

Generated by Claude Code