prompt: retire smoke-install-stale-jax-pin (shipped, never recorded) + file the drift-detection follow-up

complete/2026/08/smoke-install-stale-jax-pin.md

@@ -0,0 +1,168 @@
+## smoke-install-stale-jax-pin
+- issue: https://github.com/PyAutoLabs/autolens_workspace_test/issues/266
+- completed: 2026-08-23
+- workspace-pr: https://github.com/PyAutoLabs/autolens_workspace_test/pull/268 (merged a97f052)
+- sibling-pr: autogalaxy_workspace_test — companion PR, identical change; its smoke_install.sh on main carries the same JAXCHECK block
+- summary: Dropped the vestigial `jax<0.7 jaxlib<0.7` pin from smoke_install.sh. Added in #82 solely to keep tensorflow-probability==0.25.0 importable; #184 removed that dependency for tfp-nightly and left the pin behind. Not inert: jax is a base dep of autonerves (>=0.7.0,<0.12.0), so the pin downgraded a conforming install to 0.6.2 and only the following `[optional]` re-resolution pulled it back to 0.10.2 — CI green by line ordering, not by constraint. Replaced with a #82→#184 trail comment plus a JAXCHECK heredoc asserting the resolved version in [0.7.0, 0.12.0).
+- mind-drift: shipped without touching Mind state; retired to complete/ on 2026-08-27 after the still-live dashboard chip caused a re-pick. See the close-out note below.
+
+`autolens_workspace_test/.github/scripts/smoke_install.sh` pinned
+`jax<0.7 jaxlib<0.7`. The pin was added in #82 for exactly one reason, stated in
+its own commit message: keeping `tensorflow-probability==0.25.0` importable,
+since `tfp.substrates.jax` referenced `jax.interpreters.xla.pytype_aval_mappings`,
+removed in JAX 0.7.0. #184 then deleted `pip install tensorflow-probability==0.25.0`
+when the stack moved to `tfp-nightly` (pinned by `PyAutoArray[optional]`, whose
+own comment documents the nightly as the `jax>=0.7`-compatible one). The pin
+outlived the dependency it protected.
+
+The framing that made it worth fixing rather than shrugging at: it was **not
+inert**. `jax` is a *base* dependency of `autonerves` (`jax>=0.7.0,<0.12.0`,
+PyAutoLens#702), so the preceding line already installed a conforming jax; the
+pin then downgraded it to 0.6.2 — a configuration the stack does not claim to
+support — and only the *following* `[optional]` re-resolution pulled it back to
+0.10.2. CI was green by **line ordering, not by constraint**. Reordering those
+two lines, or a change in what the extras resolve, would have dropped the whole
+smoke suite onto jax 0.6.2 silently, surfacing as unexplained smoke breakage
+rather than as an install error.
+
+## PRs
+
+- autolens_workspace_test#268 → `a97f052` (issue autolens_workspace_test#266 closed),
+  branch `feature/smoke-install-stale-jax-pin`, merged 2026-08-23. +23/−1, one file.
+- `autogalaxy_workspace_test` carried the identical line and was fixed in a
+  companion PR; its `smoke_install.sh` on `main` now carries the same `JAXCHECK`
+  block.
+
+## The fix
+
+Delete the pin and let `autonerves`' base requirement govern. In its place:
+
+1. A comment recording the #82 → #184 trail, so the pin is not reintroduced by
+   someone reading the file cold.
+2. A `JAXCHECK` heredoc appended after the final
+   `pip install --force-reinstall --no-deps ./PyAutoNerves`, asserting
+   `(0, 7) <= (major, minor) < (0, 12)` on the resolved `jax.__version__` and
+   printing it to the CI log. With `set -e` at the top, a violation aborts the
+   install step.
+
+The `tfp-nightly` NOTE block and the trailing `--force-reinstall --no-deps
+./PyAutoNerves` were preserved verbatim — both remain accurate.
+
+Two decisions worth not "simplifying" later:
+
+- **Deleted, not rewritten as an honest `jax>=0.7,<0.12`.** A second copy of the
+  constraint drifts, and this one already had: the range was `<0.11` when the
+  defect was filed and `<0.12` by the time it was fixed. The assertion checks the
+  range without owning it.
+- **The guard compares a major/minor tuple, not `packaging.version`.** It runs
+  inside the install epilogue, where an undeclared import would fail the install
+  rather than fail softly. `import jax` is deliberately unguarded: the workflow
+  runs on `ubuntu-latest`, where `autonerves`' platform marker always installs
+  jax, so an absent jax is itself worth catching. That would need softening only
+  if this workflow ever ran on Intel macOS.
+
+## Verification
+
+The PR's own Smoke Tests run **is** the from-scratch install replay the prompt
+asked for; the guard is what turns it from an inference into a check. Before
+that: `bash -n` clean; the guard executed as a real bash heredoc against a stub
+`jax` (`0.10.2` prints and exits 0; `0.6.2` and `0.12.0` exit 1 with the
+explanatory message); boundary cases `0.7.0`/`0.9.0`/`0.10.2`/`0.11.4` pass and
+`0.6.2`/`0.12.0`/`1.0.0` fail.
+
+## Sibling sweep (the prompt's scope item 4)
+
+`autocti_workspace_test` and `autofit_workspace_test` were checked at fix time
+and carry no jax pin. Re-confirmed at close-out: an org-wide code search for
+`jaxlib<0.7` across PyAutoLabs returns **no live script** — only this prompt's
+own text in PyAutoMind and one prose mention in
+`complete/2026/08/jax-grad-local-vs-ci-assertions.md`.
+
+## Close-out note: this record is late, and why
+
+The task shipped on 2026-08-23 without ever touching Mind state — no `active.md`
+entry, no `active/` move, no record. The prompt sat in
+`draft/maintenance/ci/` and kept rendering on `dashboard.md` as pickable
+backlog with a live `/start_dev` chip, which is how it came to be picked again on
+2026-08-27. Nothing downstream could tell the difference: as `AGENTS.md` notes,
+a shipped-but-unretired prompt renders faithfully and no workflow detects it.
+`lifecycle.py check` did not catch it either — its invariant is that no
+`active.md` slug has a record, and this task was in neither place.
+
+The re-pick cost only a verification pass, because the prompt named the file and
+the file's current contents were the answer. The generalisable check for the next
+one: before opening an issue, read the prompt's own target file on `main`. If the
+defect it describes is not there, look for the merged PR before doing anything
+else.
+
+## Original prompt
+
+# smoke_install.sh's stale `jax<0.7` pin — CI is on the right jax by accident
+
+Type: maintenance
+Target: ci
+Repos:
+- @autolens_workspace_test
+Difficulty: low
+Autonomy: supervised
+Priority: medium
+Status: formalised
+Filed: 2026-08-22 (backfilled from git)
+
+Found 2026-08-22 while building a CI-equivalent environment to reproduce
+autolens_workspace_test#260. Latent — CI is green today — but it is green for the
+wrong reason.
+
+## The defect
+
+`autolens_workspace_test/.github/scripts/smoke_install.sh:9`:
+
+```bash
+pip install "jax<0.7" "jaxlib<0.7"
+```
+
+Replaying the install script verbatim, that line **downgrades jax to 0.6.2** and
+raises a resolver conflict against autonerves' own requirement:
+
+```
+autonerves 9999.0.0.dev0 requires jax<0.11.0,>=0.7.0; ... but you have jax 0.6.2
+which is incompatible.
+Successfully installed jax-0.6.2 jaxlib-0.6.2
+```
+
+The install only ends up on the intended **0.10.2** because the *next* line's
+`[optional]` extras happen to pull it back up:
+
+```bash
+pip install "./PyAutoArray[optional]" "./PyAutoGalaxy[optional]" "./PyAutoLens[optional]"
+```
+
+## Why it matters
+
+The pin no longer expresses the intent it was written for, and the correct
+outcome now depends on line ordering rather than on the constraint. Any
+reordering of those two lines, or a change to what the `[optional]` extras
+resolve, would silently drop the entire smoke suite onto jax 0.6.2 — and because
+`autonerves` declares `jax>=0.7`, that is a configuration the stack does not
+claim to support. The failure would surface as unexplained smoke breakage, not as
+an install error.
+
+The comment block immediately below that line (about `tfp-nightly` vs
+`tensorflow-probability`) is still accurate and should be preserved.
+
+## Suggested scope
+
+1. Establish what the `jax<0.7` pin was originally protecting against, and whether
+   that reason still holds — do not simply delete it because it looks stale.
+2. Either remove it (letting `autonerves`' `jax<0.11.0,>=0.7.0` govern) or replace
+   it with a pin that states the real intended range.
+3. Verify by replaying the install from scratch and asserting the resolved jax
+   version, rather than inferring it from a green run.
+4. Check whether sibling workspaces' install epilogues carry the same stale pin.
+
+<!-- Sizing: declared low; the sizing faculty derives medium (5). Kept at low — the
+     change is one line plus a verification replay; the prompt is long because the
+     evidence is, not because the work is. -->
+
+<!-- Not filed as a GitHub issue at discovery time: unrelated to the bug being
+     worked (autolens_workspace_test#260), and deliberately not folded into it. -->

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.
 
-1168 records across 7 buckets.
+1169 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -253,6 +253,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [small-datasets-loader-pixel-scales](2026/08/small-datasets-loader-pixel-scales.md)
 - [small-datasets-regime-stamp](2026/08/small-datasets-regime-stamp.md)
 - [small-datasets-rmtree-committed-data](2026/08/small-datasets-rmtree-committed-data.md)
+- [smoke-install-stale-jax-pin](2026/08/smoke-install-stale-jax-pin.md) — Dropped the vestigial `jax<0.7 jaxlib<0.7` pin from smoke_install.sh. Added in #82 solely to keep tensorflow-p…
 - [smoke-runner-delegation](2026/08/smoke-runner-delegation.md)
 - [smoke-runner-jupyter-guard](2026/08/smoke-runner-jupyter-guard.md)
 - [smoke-surface-retime-sweep](2026/08/smoke-surface-retime-sweep.md)