prompt: retire loggaussian-prior-support, file and work its four follow-ups - #350

Filed from the follow-up list in
complete/2026/08/loggaussian-prior-declares-own-support.md:

- refactor/autofit/optimisation_state_limit_guard_truthiness.md
- refactor/autofit/redundant_prior_limits_overrides.md
- research/graphical_ep/transformed_message_declares_support.md
- test/autogalaxy/prior_passing_loggaussian_lower_bound.md

Also corrects that record's follow-up 1. PyAutoFit#1527 describes
`OptimisationState.valid`'s guard as "falsy at 0.0", implying the fix disabled a
limit check for the prior it gave a 0.0 lower limit. It did not: `lower_limit`
there is a `VariableData` dict, so the guard tests non-emptiness, and the EP path
reads limits off the message, which #1527 left at +/-inf. What remains is a
readability defect, not a live bug — the prompt is filed as refactor on that
basis. Follow-up 2 likewise gains the `float()` coercion caveat that makes it a
type change rather than a pure deletion.

Dashboard regenerated in the same commit; `lifecycle.py check` OK.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0173PAAaCQ7myTu9SUboYA4Y

- 1 and 2 implemented together on PyAutoFit
  claude/loggaussian-prior-support-ngh59x (4c0f79b); status appended to both
  prompts, which stay in draft/ until a PR is opened.
- Doing 1 exposed a live bug: VariableData.any reduced through var_all, so
  OptimisationState.valid reported a parameter vector valid when only SOME
  components were outside their limits. Fixed in the same commit.
- 3 partly answered and the measurements appended: EP's check_limits does not
  enforce LogGaussian's support, but the message's own density returns a clean
  -inf, so nothing is wrong today. Design question stays open, lower priority.
- 4 done, null result, retired to complete/2026/08/ — neither PyAutoGalaxy nor
  PyAutoLens constructs a LogGaussianPrior, so there is no downstream exposure.

Dashboard regenerated; lifecycle.py check OK.

Co-Authored-By: Claude <noreply@anthropic.com>

…undant overrides

PyAutoFit#1532 merged as 6e2d8c8; issue #1531 closed. Two prompts retired, since
both shipped in that one PR:

- complete/2026/08/optimisation-state-limit-guard-truthiness.md
- complete/2026/08/redundant-prior-limits-overrides.md

The find worth keeping: VariableData.any reduced through var_all, so
OptimisationState.valid accepted a parameter vector with only SOME components
outside their limits, and MeanField's validity count under-reported the same
way. Neither had any test coverage, so the suite stayed green through it — the
#1477 process lesson, third time in this lineage.

Also repoints the parent record's follow-up list, whose "No PR opened yet" the
merge falsified.

complete/index.md refreshed, dashboard regenerated, reconcile clean (0 suspects
of 136), lifecycle.py check OK.

Co-Authored-By: Claude <noreply@anthropic.com>

The branch was 10 commits behind. Both conflicts were in GENERATED files --
complete/index.md and dashboard.md -- so they were resolved by regenerating from
the merged sources rather than hand-picking hunks:

  lifecycle.py index --apply            -> 1159 records
  pyauto-brain intake --apply dashboard -> 134 prompts

Both --check legs report current and lifecycle.py check is OK. No source-of-truth
file conflicted: the four completion records, the retired prompts and the
research prompt all merged cleanly.

Co-Authored-By: Claude <noreply@anthropic.com>