prompt: close out aplt-output-drift-remaining-repos + file its three follow-ups

complete/2026/08/aplt-output-drift-remaining-repos.md

@@ -0,0 +1,184 @@
+Repaired the `aplt.Output` stale-API drift in the last two repos the
+2026-08-04 `plot-array-stale-kwargs` task deliberately stopped short of — and
+found that the drift's real shape was not what the prompt described.
+
+## Outcome
+
+| Repo | PR | Merged |
+|---|---|---|
+| PyAutoGalaxy | [#586](https://github.com/PyAutoLabs/PyAutoGalaxy/pull/586) | `d68a8f6` |
+| euclid_strong_lens_modeling_pipeline | [#39](https://github.com/PyAutoLabs/euclid_strong_lens_modeling_pipeline/pull/39) | `62b2fd4` |
+
+Classification changed during planning from workspace-only to **library +
+workspace**, which is why a PyAutoGalaxy PR exists at all.
+
+## The prompt was wrong in three ways, each found by verifying rather than trusting
+
+**1. `autocti_workspace_test` is out of scope — the hedge resolved, then the
+scope collapsed.** The prompt flagged 27 files as "UNVERIFIED — check first",
+suspecting PyAutoCTI might still export `Output`. It does not:
+`autocti/plot/__init__.py` is 49 lines of flat functions. So those files are
+genuinely broken — but an alias-aware scan found **31**, not 27, and:
+
+- **18 are under `legacy/`**, which `autocti_workspace_test/AGENTS.md:52`
+  forbids editing ("preserved Euclid VIS history"). Its README says outright
+  that they target the removed Plotter object stack and are not runnable.
+- **13 are top-level `imaging_ci/`** — undocumented in AGENTS.md's structure
+  block, absent from `smoke_tests.txt`, substantive history **2023-02**, the
+  same era as `legacy/` (created 2026-07-17 by "CTI resurrection Phase 5").
+  Its only 2026 commits are repo-wide mechanical sweeps that also hit `legacy/`.
+
+Following the prompt literally would have modernised dead code *and* violated
+the repo's own hard rule. Filed as
+`draft/maintenance/autocti_workspace_test/imaging_ci_heritage_sweep.md` for a
+human decision against the CTI resurrection epic (PyAutoCTI#82).
+
+**2. The drift was never really about `Output`.** The whole plotter-object stack
+is gone from every public namespace — `MatPlot2D`, `MatPlot1D`, `Cmap`,
+`Array2DPlotter`, `Visuals2D`, `MassProfileCentresScatter`, the `*Plotter`
+family. `Output` was 6 of 13 missing symbols in autocti, 2 of 6 in euclid.
+
+**3. Two library defects the prompt never mentioned.** `Cmap` is exported by
+**no** public plot namespace, yet:
+
+- `Scribbler.__init__` required a `Cmap`-shaped object for `cmap=`, so callers
+  had no public way to colour the GUI.
+- `Clicker.start()` built `aplt.Cmap(...)` on `autoarray.plot`
+  *unconditionally* — reproduced on the installed stack as
+  `AttributeError: module 'autoarray.plot' has no attribute 'Cmap'`. That GUI
+  was dead for every caller.
+
+Fixed library-side because the only workspace alternative was a private-path
+import — the "autoimmune reaction" the Bug Agent's fix-locus rule warns against.
+
+## The plan's worst assumption, caught before it shipped
+
+The approved plan said euclid's explicit `mask=` could simply drop out, since
+`plot_array` auto-derives the outline. It cannot. `data` there comes from
+`Array2D.from_fits` and is **unmasked**, so `auto_mask_edge(data)` returns
+`None` — verified directly. Worse, the mask being drawn is a *separately
+constructed* circular one whose radius is grown at line 160 to enclose the
+clicked galaxies, so it is the entire subject of the figure.
+
+A literal reading would have silently deleted the mask radius from the PNG that
+exists to show it, and nothing would have failed. Fixing it needed a second
+library change: autogalaxy's `plot_array` wrapper delegates to autoarray's but
+was dropping its `mask=` parameter.
+
+## Verification
+
+- PyAutoGalaxy CI green on **every leg**: `unittest (3.12)`, `unittest (3.13)`,
+  `unittest-nojax`, Docs. 1129 tests, 16 new across two files (`test_autogalaxy/gui/`
+  did not previously exist).
+- Alias-aware AST re-scan — aliases resolved from each file's own imports, the
+  correction carried forward from HowToGalaxy#56 — **0 residual stale symbols**
+  in both repos.
+- All 10 changed `aplt.*`/`al.*` calls bound against real signatures: 10 ok, 0 failures.
+- Mask passthrough verified by **rendering**: 2436 pixels differ vs. the same
+  call without `mask=`, confirmed by pixel diff rather than file size.
+- `compileall` clean; CRLF preserved uniformly (see below).
+
+## Not verified, stated plainly
+
+- **The two euclid GUIs were never run end-to-end.** They need TkAgg and FITS
+  data absent from the repo, and euclid has **no CI workflows at all**. The
+  merge was made with that stated and acknowledged.
+- **`Scribbler.__init__` has no direct test** — `matplotlib.use("TkAgg")` and
+  `wm_geometry` cannot run headless. The extracted `norm_from` helper carries
+  the logic and is fully tested; the constructor is covered only by binding.
+
+## Behaviour change accepted
+
+Extra-galaxy centre markers were cyan via `MassProfileCentresScatter(c="cy")`
+and are now the `plot_array` default. `autoarray/plot/array.py:258-262`
+hardcodes the overlay colour cycle, and `line_colors` (265-273) applies to
+`lines`, not `positions` — an early misreading of mine, corrected before it
+reached the code. Restoring the colour needs a PyAutoArray change; markers
+remain visible and distinguishable from the mask outline by size and z-order.
+
+## Process note
+
+An intermediate "suite green" reading was wrong twice and caught both times:
+pytest had exited 4 on an unrecognised `--timeout` flag without running, and a
+later run was corrupted when a `git stash` reverted the tree mid-run. Reported
+results come from clean runs on the final tree. Separately, a stop-hook flagged
+the PyAutoGalaxy branch as unpushed: the commit *was* on GitHub, but the
+`--depth 1` clone's single-branch refspec meant no remote-tracking ref existed,
+so `@{u}` could not resolve — repaired by widening the refspec.
+
+## Follow-ups filed, not fixed
+
+- `draft/maintenance/autocti_workspace_test/imaging_ci_heritage_sweep.md` — the
+  13 heritage files the `legacy/` sweep missed.
+- `autoarray` duplicates this normalisation inline in `plot/array.py` and
+  `plot/inversion.py`; a shared helper there is the real fix. This task added
+  one copy in autogalaxy serving both GUIs rather than a third inline copy.
+- euclid's three `tools/` files are **100% CRLF on `main`**, contradicting its
+  own AGENTS.md ("CRLF will break shell scripts on the HPC"). Preserved rather
+  than converted, to avoid whole-file diffs that would not fix the repo-wide
+  problem.
+
+## Original prompt
+
+# `aplt.Output` stale-API drift in the remaining workspace repos
+
+Type: bug
+Target: workspaces
+Repos:
+- autocti_workspace_test
+- euclid_strong_lens_modeling_pipeline
+Difficulty: small
+Autonomy: supervised
+Priority: normal
+Status: formalised
+Filed: 2026-08-04 (backfilled from git)
+Issued: 2026-08-24
+
+Split out of `plot-array-stale-kwargs` (HowToGalaxy#56, 2026-08-04), which
+repaired this same drift in `autolens_workspace_developer` but deliberately
+stopped at that repo's boundary.
+
+## The drift
+
+`aplt.Output` no longer exists on the **autolens / autogalaxy** plot namespace —
+verified: `hasattr(autolens.plot, "Output") == False`. It survives only as
+`autoarray.plot.Output`. The removal was deliberate and is already documented in
+`autolens_assistant/AGENTS.md:218` ("the `aplt.MatPlot2D` / `aplt.Output` objects
+have been removed — do not use them").
+
+Callers must move to the flat convention. Note the accepted kwargs differ per
+callee — check each signature rather than blanket-renaming:
+
+```python
+# plot_array takes all three
+aplt.plot_array(array=..., output_path=P, output_filename=F, output_format="png")
+# subplot_* take only path + format (no output_filename)
+aplt.subplot_tracer(tracer=..., grid=..., output_path=P, output_format="png")
+```
+
+## Sites
+
+| Repo | Files | Status |
+|------|-------|--------|
+| `autocti_workspace_test` | 27 | **UNVERIFIED — check first** |
+| `euclid_strong_lens_modeling_pipeline/tools/` | 2 (`psf_size.py`, `extra_galaxies_centres_gui.py`) | confirmed broken |
+
+**Do not assume the autocti files are broken.** Those import
+`import autocti.plot as aplt` — a *different* library's plot namespace.
+`autocti` was not installed in the 2026-08-04 session so it could not be
+checked. PyAutoCTI may still export `Output`, in which case those 27 files are
+correct as written and must be left alone. Verify with
+`hasattr(autocti.plot, "Output")` before touching anything.
+
+Confirmed **not** bugs, do not "fix" them:
+- `PyAutoArray/test_autoarray/plot/test_output.py` — there `aplt` *is*
+  `autoarray.plot`, which does export `Output`.
+- `autolens_assistant` markdown — documents the removal.
+
+## Verification
+
+Re-run an alias-aware AST scan after the fix (the 2026-08-04 session's first
+sweep hardcoded the alias `aplt` and **missed** a call site written as `aaplt`;
+resolve aliases from each file's own imports). Then bind each changed call's
+kwargs against the real callee signature via `inspect.signature`, since these
+repos have little or no CI to catch a wrong kwarg name.

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.
 
-1125 records across 7 buckets.
+1126 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -20,6 +20,7 @@ _(curate hard-won records here — survives regeneration.)_
 
 - [actionable-health-board](2026/08/actionable-health-board.md) — closed on ship
 - [active-prompt-orphan-triage](2026/08/active-prompt-orphan-triage.md)
+- [aplt-output-drift-remaining-repos](2026/08/aplt-output-drift-remaining-repos.md)
 - [arcticpy-install-standardisation](2026/08/arcticpy-install-standardisation.md)
 - [arxiv-inbox-tier](2026/08/arxiv-inbox-tier.md) — the nightly strong-lensing digest no longer ends in a Slack paste-block the
 - [astropy-cap-bump](2026/08/astropy-cap-bump.md) — closed manually post-merge — "Part of" in the PR bodies does not auto-close