prompt: close out the Clipper handover — retire shipped phase 1, file what was owed

[Jammy2211]

Closes out the Clipper handover. Nothing from the autolens_profiling#128 investigation is left living only inside another prompt file, and the Mind's view of the Clipper work now matches reality.

Why this needs to be on main: the same reason #188 gave. A cold session gets whatever is in its checkout, which comes from main. Until this merges, the phase-1 prompt still reads as unstarted high-priority backlog on the dashboard people pick work from, and the four filed follow-ups do not exist.

Phase 1 had already shipped and the Mind had not noticed

PyAutoFit#1477 merged as 1f4b66a — the current tip of PyAutoFit main — shipping AbstractClipper / ClipperNone / ClipperPriorBox, wired into both AbstractMultiStartGradient and AbstractBFGS, opt-in, default ClipperNone, bit-identical under it, 22 tests.

prior_support_clipper.md was still sitting in draft/ as a high-priority backlog item. Retired to complete/2026/08/prior-support-clipper.md via lifecycle.py record, prompt folded in, with the findings worth keeping written up: the inset keyed on bound kind rather than unguarded upper - lower; LogGaussianPrior reporting (-inf, inf) for a support of (0, inf); scipy reading a bounds tuple as (min, max) pairs and silently mis-fitting two-parameter models.

Four prompts filed, three of them previously homeless

Two incidental PyAutoFit bugs that prior_support_clipper.md flagged under "do not lose these — file separately if confirmed". Both confirmed by reading PyAutoFit main, and both still present at 1f4b66a — phase 1 did not fix either. Both are now implemented and in flight:

• save_json_numpy_scalar_typeerror.md → PyAutoFit#1479. paths/directory.py:80 is a bare json.dump with no default= hook, and there is no JSON encoder anywhere in autofit/. float32 raises at the end of a successful run. It did not fire while 14/16 lanes were dead, so it starts firing exactly when lane survival improves. Implementing it turned up a second unguarded site the prompt never named — Samples.info_to_json — which is the more dangerous of the two, since samples_info is a search's own diagnostic channel.

• crashed_run_poisons_resume.md → PyAutoFit#1480. directory.py:79 opens "w+" (truncates, no atomic rename) → is_complete tests only the .completed marker → abstract_search.py:582 dispatches to resume → the resume reads the truncated JSON → and JSONDecodeError subclasses ValueError, so the guard at multi_start_gradient/search.py:720 catching (FileNotFoundError, TypeError, KeyError) never falls through to the fresh-start branch.

  Correction, from reproducing it. The prompt as filed described the result as "a zero-step no-op reported as a clean run". That did not reproduce. What reproduces against main is a hard JSONDecodeError on every rerun of that search name, naming no file and offering no remedy. The no-op variant presumably needs a surviving search_internal whose restored total_steps short-circuits the loop — the crash path deletes it first. Same root cause, and #1480 covers both paths, but only the crash is evidenced, and the prompt now says so.

One research task, follow-up (2) owed by active/mge_lane_death.md:

• ell_comps_trapping_unmasked.md — the plateau was masked, not cleared. The baseline n_constrained_lane_steps = 0 meant "nothing got that far"; lanes died of prior-exit first. Records that the 27.79% came from the prior-neutered diagnostic arm and is not a citable production number, so the task sequences after phase 1 and measures the clipped run instead.

And one new feature task:

• clipper_usage_in_search_summary.md → PyAutoFit#1478. Surface the clip count in search.summary.

Phase 2 unblocked

clipper_validation_campaign.md marked ready, with what phase 1 changed for it: record n_clipped_lane_steps as a fourth counter from search_internal; a ClipperPriorBox arm reporting zero clips has not exercised the clipper and its "no change" result is meaningless; the momentum-reset arm needs writing since nothing yet consumes the mask; and the save_json trap stands until #1479 lands.

Bookkeeping

All seven references to the moved prompt repointed. lifecycle.py check and index --check both OK; dashboard.md regenerated.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FzF2XmKQaqZRWZfZMxvTNR