prompt: optional-dependency-skip-guards shipped (PyAutoFit#1512) + correct the test_nautilus drift

complete/2026/08/clipper-usage-in-search-summary.md

@@ -64,6 +64,24 @@ wired up and firing at all".
 Multi-start summaries gained a `Constrained Lane-Steps` line. Everything else is
 additive and gated.
 
+## Correction (2026-08-22) — the `test_nautilus` failure
+
+`test_nautilus.py::test__single_core_builds_no_pool` is **not** a failure on
+clean `main`. It is a **missing optional dependency**: the test runs a real
+`search.fit`, which reaches `from nautilus import Sampler`, and
+`nautilus-sampler` ships only in the `[optional]` extra. CI installs
+`[optional]` and the test passes there — latest main CI is 2024 passed,
+3 skipped, 0 failed. Sandboxes and local venvs without the extras get a hard
+`ModuleNotFoundError`.
+
+The in-flight note above calls it **pre-existing**, alongside the `astropy`
+collection error in `paths/test_save_and_load.py` — same cause, same
+correction. Both were environment-specific, not defects on `main`.
+
+Fixed by PyAutoFit#1511 / PR #1512 (skip-if-missing guards, also covering the
+`astropy` collection errors and aggregator errors from the same cause). Full
+investigation: `active/17_optional_dependency_skip_guards.md`.
+
 ## Original prompt
 
 # Report how much the Clipper actually fired — surface the count in `search.summary`

complete/2026/08/multistart-gradient-resume-fom-sanity-check.md

@@ -19,6 +19,24 @@
 - assertion design worth reusing: the load-bearing invariant is **equality with an uninterrupted reference run** (the search is deterministic), not "the counters went up". Injecting the regression (resetting both counters on resume) yields 402 against the reference's 403 — a `>=` assertion accepts that happily, equality catches it. A separate decisive check also ran: stamping sentinel values (7000/9000) into the live checkpoint and resuming gave 7045/9008, so the restored totals are provably carried rather than recounted.
 - also confirmed while doing it: a resumed run's final counters are independent of *where* the kill landed (killed at step 72 and at step 74 both finished at the reference totals), i.e. the checkpoint captures the full accumulator state.
 
+## Correction (2026-08-22) — the `test_nautilus` failure
+
+`test_nautilus.py::test__single_core_builds_no_pool` is **not** a failure on
+clean `main`. It is a **missing optional dependency**: the test runs a real
+`search.fit`, which reaches `from nautilus import Sampler`, and
+`nautilus-sampler` ships only in the `[optional]` extra. CI installs
+`[optional]` and the test passes there — latest main CI is 2024 passed,
+3 skipped, 0 failed. Sandboxes and local venvs without the extras get a hard
+`ModuleNotFoundError`.
+
+The validation bullet above says "confirmed pre-existing on unmodified
+`main`". "Not caused by this work" was right; "pre-existing on `main`" is the
+wrong gloss — read it as environment-specific.
+
+Fixed by PyAutoFit#1511 / PR #1512 (skip-if-missing guards, also covering the
+`astropy` collection errors and aggregator errors from the same cause). Full
+investigation: `active/17_optional_dependency_skip_guards.md`.
+
 ## Original prompt
 
 # MultiStartGradient cannot resume a killed mid-run search — FoM sanity check compares log-likelihood against chi-squared

complete/2026/08/optional-dependency-skip-guards.md

@@ -0,0 +1,151 @@
+- issue: https://github.com/PyAutoLabs/PyAutoFit/issues/1511
+- completed: 2026-08-22
+- library-pr: https://github.com/PyAutoLabs/PyAutoFit/pull/1512 (MERGED 2026-08-22, merge commit 14eb8393)
+- summary: `test_nautilus.py::test__single_core_builds_no_pool` had no skip guard
+  while running a real `search.fit`, which reaches `from nautilus import Sampler`
+  (`autofit/non_linear/search/nest/nautilus/search.py:238`). `nautilus-sampler`
+  ships only in the `[optional]` extra, so the test hard-failed with
+  `ModuleNotFoundError` in every env without those extras while passing in CI,
+  which installs `[optional]`. Same cause for the two `astropy` collection errors
+  and nine aggregator errors. Fixed with `importlib.util.find_spec` skipif markers
+  (and `pytest.importorskip` where the import is module-level), matching the
+  existing house pattern in `test_blackjax_nuts.py`, `nest/nss/test_search.py`,
+  `test_arithmetic_jax_trace.py` and `test_fork_context.py`. Tests only — no
+  `autofit/` source touched.
+- **the real cost was the misdiagnosis, not the failure.** The same red test was
+  reported as "pre-existing on clean `main`" in at least six completion records
+  from 2026-08-16 onward, each session spending a control run to re-disprove it,
+  and the wrong gloss propagated into PR bodies #1479 and #1480. Main was never
+  red: CI on main @ `a639226` was 2024 passed / 3 skipped / 0 failed across 3.12,
+  3.13 and the no-jax leg.
+- evidence at ship (untouched main @ `a639226`, py3.12, measured before any edit):
+  without `nautilus-sampler` → 1 failed / 3 passed, `ModuleNotFoundError` at
+  `search.py:242`; after `pip install nautilus-sampler==1.0.5` on the same tree
+  with no other change → 4 passed. Full suite with no optional extras → 2
+  collection errors, then 4 failed / 9 errors, **all** missing-`astropy`.
+- verification of the fix, three envs on the branch: no extras at all → the
+  nautilus test skips with its reason (3 passed, 1 skipped); no `astropy` → 1985
+  passed, 33 skipped, 0 failed, 0 errors; full `[optional]` extras, i.e. the CI
+  env → **2024 passed, 3 skipped, identical to CI on main**. That last leg is the
+  load-bearing one: it proves the guards are inert wherever the deps exist, so no
+  coverage is lost. A skip guard that quietly disabled tests in CI would be a
+  worse bug than the one it fixes — check it that way, not by trusting green.
+- CI at merge: unittest 3.12, unittest 3.13 and unittest-nojax all green on
+  `b814937` (run 32575550710); mergeable state clean, no review comments.
+- **process finding worth more than the fix.** This was already diagnosed
+  correctly on 2026-08-18 as follow-up 4 of
+  `complete/2026/08/uniform-prior-bounds-numpy-path.md` ("hygiene candidate:
+  skip-if-missing markers"), and the "environment-specific, not pre-existing"
+  correction was already written on 2026-08-17 in
+  `complete/2026/08/prior-support-clipper.md:218`. Both were right; both lived
+  only in completion records, which nothing reads on the way *into* a new task.
+  The false alarm therefore fired three more times after the right answer was
+  written down. A follow-up that stays in a record has no owner — file it as a
+  prompt or it does not exist.
+- correction trail (records annotated in this pass with a dated `## Correction`
+  block rather than rewritten in place, per the prior-support-clipper precedent):
+  `multistart-gradient-resume-fom-sanity-check.md`,
+  `clipper-usage-in-search-summary.md`, `uniform-prior-bounds-numpy-path.md`.
+  Already correct and left alone: `covariance-interpolator-rng-seed.md` (the only
+  record that named the real `ModuleNotFoundError`) and
+  `prior-support-clipper.md`.
+- prior art: `complete/2026/08/nautilus-1core-serial-pool.md` — the task that
+  *added* this test (PyAutoFit#1442 → #1443). The test itself is sound and still
+  guards the real fork-pool deadlock; only its environment assumption was missing.
+- worktree: none — cloud session from a direct clone at `/home/user/pyautofit`,
+  with a py3.12 venv built three ways to isolate the dependency variable.
+
+## Original prompt
+
+# `@PyAutoFit` `[optional]`-dependency tests have no skip guards
+
+Type: test
+Target: autofit
+Difficulty: small
+Autonomy: autonomous
+Priority: normal
+Status: issued 2026-08-22 as PyAutoFit#1511; PR #1512 open, CI green, awaiting review
+
+Not a defect on `main` — a test-hygiene gap that has been *repeatedly
+misdiagnosed* as one. That misdiagnosis is the actual cost.
+
+## The finding
+
+`test_autofit/non_linear/search/nest/test_nautilus.py::test__single_core_builds_no_pool`
+runs a real `search.fit`, which reaches `from nautilus import Sampler`
+(`autofit/non_linear/search/nest/nautilus/search.py:238`).
+`nautilus-sampler==1.0.5` ships **only** in the `[optional]` extra
+(`pyproject.toml:85`), and the test carries **no skip guard**.
+
+CI installs `[optional]` — PyAutoHeart's `lib-tests.yml` does
+`pip install "./$r[optional]"` — so CI never sees it. Every env installed
+without those extras (cloud sandboxes, local venvs) gets a hard
+`ModuleNotFoundError`.
+
+It is the only optional-sampler test in the suite without a guard. Precedent
+is everywhere else: `test_blackjax_nuts.py:14`, `nest/nss/test_search.py:20`,
+`test_arithmetic_jax_trace.py:6`, `test_fork_context.py:31`.
+
+`astropy` is the same class of noise: two modules fail *collection* on its
+top-level import, and nine aggregator tests error on it at runtime
+(`autofit/aggregator/file_output.py:101`,
+`autofit/aggregator/summary/aggregate_fits.py:105,165`).
+
+## Why this was worth a task
+
+The single nautilus failure has been reported as "pre-existing on clean
+`main`" in at least six completion records since 2026-08-16 — see the
+correction trail below. Each session spent a control run re-disproving it,
+and the wrong gloss propagated into PR bodies #1479 and #1480. Main is **not**
+red: latest main CI (run 32546158666) is 2024 passed, 3 skipped, 0 failed on
+3.12, 3.13 and the no-jax leg.
+
+## Evidence (untouched main @ `a639226`, py3.12)
+
+| env | result |
+|---|---|
+| no `nautilus-sampler` | `1 failed, 3 passed` — `ModuleNotFoundError` at `search.py:242` |
+| `pip install nautilus-sampler==1.0.5`, same tree, no other change | `4 passed` |
+| full suite, no optional extras | 2 collection errors → then 4 failed / 9 errors, **all** missing-`astropy` |
+
+## The fix
+
+Skip-if-missing guards in the house pattern — an `importlib.util.find_spec`
+`skipif` marker, or `pytest.importorskip` where the import is module-level.
+Tests only; no `autofit/` source touched.
+
+Verified three ways on the branch:
+
+- no extras at all — nautilus test skips with its reason: `3 passed, 1 skipped`
+- no `astropy` — `1985 passed, 33 skipped, 0 failed, 0 errors`
+- full `[optional]` extras (the CI env) — `2024 passed, 3 skipped`, **identical
+  to CI on main**, so the guards are inert where the deps exist and no
+  coverage is lost
+
+## Correction trail
+
+The claim "fails identically on the untouched tree — pre-existing, not this
+change" is right about causation and wrong about `main`. Records carrying the
+loose gloss:
+
+- `complete/2026/08/prior-support-clipper.md:152,218` — already issued the
+  correction ("read it as environment-specific"); the correct reading
+- `complete/2026/08/covariance-interpolator-rng-seed.md:14` — the only record
+  that names the real `ModuleNotFoundError`; correct
+- `complete/2026/08/uniform-prior-bounds-numpy-path.md:31,50` — follow-up 4
+  proposed exactly this fix but it was never filed as a prompt, so nothing
+  acted on it; this task is that follow-up
+- `complete/2026/08/multistart-gradient-resume-fom-sanity-check.md:14` —
+  "confirmed pre-existing on unmodified `main`"; wrong gloss
+- `complete/2026/08/clipper-usage-in-search-summary.md:89` — "**pre-existing**";
+  wrong gloss
+
+Once this merges, no env reproduces the failure and the gloss stops
+recurring.
+
+## Prior art
+
+`complete/2026/08/nautilus-1core-serial-pool.md` — the task that *added*
+`test__single_core_builds_no_pool` (PyAutoFit#1442 → #1443). The test itself is
+sound and still guards the real fork-pool deadlock; only its environment
+assumption was missing.

complete/2026/08/uniform-prior-bounds-numpy-path.md

@@ -50,6 +50,24 @@
   4. `test_beta` + `test__single_core_builds_no_pool` fail on clean main without optional deps
      (sympy?/nautilus-sampler absence) — hygiene candidate: skip-if-missing markers.
 
+## Correction (2026-08-22) — the `test_nautilus` failure
+
+`test_nautilus.py::test__single_core_builds_no_pool` is **not** a failure on
+clean `main`. It is a **missing optional dependency**: the test runs a real
+`search.fit`, which reaches `from nautilus import Sampler`, and
+`nautilus-sampler` ships only in the `[optional]` extra. CI installs
+`[optional]` and the test passes there — latest main CI is 2024 passed,
+3 skipped, 0 failed. Sandboxes and local venvs without the extras get a hard
+`ModuleNotFoundError`.
+
+Follow-up 4 above ("hygiene candidate: skip-if-missing markers") called this
+correctly and proposed the fix, but was never filed as a prompt, so nothing
+acted on it for four days. It is now filed and fixed.
+
+Fixed by PyAutoFit#1511 / PR #1512 (skip-if-missing guards, also covering the
+`astropy` collection errors and aggregator errors from the same cause). Full
+investigation: `active/17_optional_dependency_skip_guards.md`.
+
 ## Original prompt
 
 # UniformPrior bounds are not enforced in the objective on the NumPy path

complete/index.md

@@ -6,7 +6,7 @@ Token-light navigation over the finished-work records (schema:
 only then grep a dated bucket. Curators: edit the band between the CURATED
 markers; everything below GENERATED is rebuilt.
 
-1072 records across 7 buckets.
+1073 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -123,6 +123,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [numba-first-call-garbage-psf-weighted-data](2026/08/numba-first-call-garbage-psf-weighted-data.md)
 - [numerical-hazard-profiling](2026/08/numerical-hazard-profiling.md) — Phase 1 established the reusable numerical-hazard profiling framework, typed risk measurements, reachability a…
 - [one-tap-dashboard-rollout](2026/08/one-tap-dashboard-rollout.md) — the pattern (a generated page + one-tap 📋 copy-for-Claude payloads, phone
+- [optional-dependency-skip-guards](2026/08/optional-dependency-skip-guards.md) — `test_nautilus.py::test__single_core_builds_no_pool` had no skip guard
 - [organism-board-final-readmes](2026/08/organism-board-final-readmes.md) — close on record
 - [paper-management-pipeline](2026/08/paper-management-pipeline.md) — auto-closed on merge
 - [per-parameter-step-scaling](2026/08/per-parameter-step-scaling.md)