prompt: ledger for the setup_notebook drift + title-underline tasks

complete/2026/08/notebook-setup-notebook-regen-drift.md

@@ -0,0 +1,88 @@
+Regenerated the 285 autolens_workspace notebooks that had drifted from
+PyAutoHands generator output. autolens_workspace#481 MERGED 2026-08-07
+(f3006b3d), closing issue #480. Branch `claude/dev-script-title-underline-5d3c3p`,
+restarted from main 573bde8 after #478 merged (same session-designated branch,
+new PR — merged-branch restart rule).
+
+- **The drift:** PyAutoHands build_util.py uncomments
+  `# from autolens import setup_notebook; setup_notebook()` when generating
+  notebook cells, but committed notebooks carried the commented form — so
+  none of them called setup_notebook() on main, and every notebook-touching
+  PR either dragged the churn or hand-restored around it (as #478 did).
+- **Sweep:** clean-main `generate.py autolens`, notebooks-only, 285 files /
+  288±288 lines. Audited: 285 setup_notebook flips + 2 JSON-indent
+  normalizations of a hand-inserted batch_size comment line (imaging/ and
+  multi_galaxy/ start_here.ipynb) — direct evidence committed notebooks were
+  hand-edited rather than generated. Catalogue zero diff. Count came in at 285,
+  not the prompt's ~290 estimate.
+- **Verification:** second regeneration on the swept tree is a no-op (same
+  285-file set, stable); check_navigator --banners fail green; underline state
+  from #478 preserved (0 typos / 14 banners); CI both smoke legs + all three
+  navigator checks green (~9 min).
+- **SIBLING DRIFT MEASURED AND FILED (not fixed):** clean-tree dry-runs
+  2026-08-07 with PyAutoHands 2a4fb11 — autogalaxy_workspace 127 modified
+  (126 flips + 6 OTHER diff lines, audit before committing),
+  autofit_workspace 31 (32 flips), HowToFit 13 (13 flips); HowToLens and
+  HowToGalaxy CLEAN. Follow-up prompt:
+  `draft/maintenance/workspaces/notebook_setup_notebook_drift_siblings.md`
+  (one task, three PRs — mge-sigma phase-2 precedent).
+- **Environment trap (repeat of #478's):** ipynb-py-convert cannot pip-install
+  on modern setuptools; vendor the package into site-packages + a CLI shim.
+
+## Original prompt
+
+# Regenerate 331 autolens_workspace notebooks drifted from generator output (setup_notebook line)
+
+Difficulty: small
+Autonomy: supervised
+Priority: low
+
+## The problem
+
+A full PyAutoHands regeneration dry-run (`generate.py autolens`) against
+**unmodified** scripts on `@autolens_workspace` main modifies **331 committed
+notebooks by exactly one line each**: `PyAutoHands:autohands/build_util.py`
+(line ~204) uncomments
+
+```
+# from autolens import setup_notebook; setup_notebook()
+```
+
+into the active form for notebook cells, but the committed notebooks carry the
+commented form. Net effect: every one of those notebooks on main currently
+**never calls `setup_notebook()`**, and any PR that regenerates notebooks drags
+this 331-file churn into its diff (or must hand-restore, as the underline task
+did).
+
+Measured 2026-08-07 on autolens_workspace `8f1965c` with PyAutoHands `2a4fb11`.
+
+## Likely cause
+
+The committed notebooks were last produced either by a generator version
+predating the uncomment transformation or by lockstep hand-edits. The
+transformation itself predates the autobuild→autohands rename, so the drift has
+been accumulating across recent notebook commits.
+
+## Proposed fix
+
+One dedicated sweep PR: run `generate.py autolens` on clean main, commit the
+notebook-only diff (expected: 331 files × 1 line, the setup_notebook flip —
+verify nothing else rides along), and confirm `check_navigator.py` stays green.
+Check whether `autogalaxy_workspace` / `autofit_workspace` / the HowTo repos
+have the same drift and file siblings if so.
+
+41 notebooks were already brought to generator truth by the underline task
+(autolens_workspace#476), so the count at sweep time will be ~290.
+
+## Verification
+
+- After the sweep, a regeneration dry-run on clean main is a no-op
+  (`git status` empty).
+- No notebook diff lines other than the setup_notebook flip.
+
+## Provenance
+
+Surfaced as a side-observation while shipping the title-underline fix
+(autolens_workspace#476, PyAutoMind
+`active/script_title_underline_off_by_one.md`), which introduced none of it —
+that task's regeneration dry-run against unmodified scripts is the measurement.

draft/maintenance/workspaces/script_title_underline_off_by_one.md → complete/2026/08/script-title-underline-off-by-one.md

@@ -1,3 +1,44 @@
+Fixed the off-by-one RST title underlines across autolens_workspace scripts.
+autolens_workspace#478 MERGED 2026-08-07 (573bde8c, merge commit; single change
+commit 5844ad5), closing issue #476. Branch `claude/dev-script-title-underline-5d3c3p`
+(session-designated — a Claude Code web session may push only that branch, so it
+replaced the `feature/<name>` convention; the same-named PyAutoMind branch
+carries this record until merged to Mind main).
+
+- **Scope grew 29 → 41.** The prompt's 2026-07-30 measurement at 726d060d found
+  29 typos; 12 more landed since with the scaling-relation and datacube feature
+  families. Each fix is one line: rule set to exactly `len(title)` `=` chars.
+- **Banner guard held.** Discriminator `abs(len(title)-len(rule)) <= 2` = typo,
+  wider = deliberate full-width banner. Control-tested at 726d060d in a temp
+  worktree, reproducing the recorded 29/15 split exactly before being trusted
+  on main. Current main has 14 banner rules (one block removed upstream since);
+  all 14 byte-untouched, re-verified after the fix (0 typos / 14 banners).
+- **Notebooks**: the 41 paired notebooks regenerated via PyAutoHands
+  generate.py; catalogue (llms-full.txt / workspace_index.json) regenerated
+  with zero diff — underline rules do not flow into catalogue summaries.
+  check_navigator --banners fail green, check_sizes green, all 41 scripts
+  byte-compile. CI: both smoke legs (3.12/3.13) + all three navigator checks
+  green in ~9 min.
+- **KEY FINDING — repo-wide notebook regeneration drift (FILED, not fixed):**
+  a generate.py dry-run against *unmodified* scripts modifies **331 committed
+  notebooks by one line** — PyAutoHands build_util.py uncomments
+  `# from autolens import setup_notebook; setup_notebook()` for notebook
+  cells, but committed notebooks carry the commented form, so main's notebooks
+  never call setup_notebook(). This PR brought only its own 41 notebooks to
+  generator truth (39 carry the flip; 2 lack the line) and hand-restored the
+  other ~290 regenerated files to keep the diff scoped. Follow-up prompt:
+  `draft/maintenance/workspaces/notebook_setup_notebook_regen_drift.md`.
+- **Environment trap:** `pip install ipynb-py-convert` fails on modern
+  setuptools (`install_layout` AttributeError in bdist_wheel). Workaround:
+  `pip download --no-deps --no-binary :all:`, copy the `ipynb_py_convert`
+  package into site-packages, and write a 4-line `/usr/local/bin/ipynb-py-convert`
+  entry-point shim — generate.py shells out to the CLI name.
+- **Sizing note:** the Feature Agent's heuristic said large (score 8) /
+  split-into-phases; overridden — prose-driven overestimate (same pattern as
+  the mge-sigma-min sizing-note). Actual: small, single-phase, one PR.
+
+## Original prompt
+
 # Fix 29 off-by-one script title underlines in @autolens_workspace
 
 Difficulty: small

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.
 
-927 records across 7 buckets.
+929 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -41,6 +41,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [missing-auto-simulate-guards](2026/08/missing-auto-simulate-guards.md)
 - [multi-start-auto-convergence-real-search](2026/08/multi-start-auto-convergence-real-search.md) — `scripts/jax_assertions/multi_start_gradient_auto_convergence.py` failed
 - [nautilus-1core-serial-pool](2026/08/nautilus-1core-serial-pool.md) — corrective for the Heart RED "release validation FAILED (stage
+- [notebook-setup-notebook-regen-drift](2026/08/notebook-setup-notebook-regen-drift.md)
 - [nufft-parity-full-datasets](2026/08/nufft-parity-full-datasets.md)
 - [plot-array-stale-kwargs](2026/08/plot-array-stale-kwargs.md)
 - [point-source-defaults-campaign](2026/08/point-source-defaults-campaign.md) — took the point-source likelihood options from "several undocumented
@@ -52,6 +53,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [purge-autocti-dataset-1d-overview](2026/08/purge-autocti-dataset-1d-overview.md) — no GitHub issue — the leftover from `autocti-util-dataset-export`, executed on direct human instruction "do th…
 - [pyautobrain-pr-test-ci](2026/08/pyautobrain-pr-test-ci.md) — auto-closed by the merge
 - [pyautogalaxy-mge-sigma-test](2026/08/pyautogalaxy-mge-sigma-test.md)
+- [script-title-underline-off-by-one](2026/08/script-title-underline-off-by-one.md)
 - [simulator-util-to-af-ex](2026/08/simulator-util-to-af-ex.md) — moved the four 1D-Gaussian simulator helpers out of the duplicated
 - [small-datasets-loader-pixel-scales](2026/08/small-datasets-loader-pixel-scales.md)
 - [smoke-runner-jupyter-guard](2026/08/smoke-runner-jupyter-guard.md)

draft/maintenance/workspaces/notebook_setup_notebook_drift_siblings.md

@@ -0,0 +1,41 @@
+# Regenerate setup_notebook-drifted notebooks in autogalaxy/autofit/HowToFit workspaces
+
+Difficulty: small
+Autonomy: supervised
+Priority: low
+
+## The problem
+
+The same generator/notebook drift fixed in @autolens_workspace (#480/#481 —
+committed notebooks carry the commented `# from auto* import setup_notebook;
+setup_notebook()` form while PyAutoHands regeneration uncomments it, so the
+notebooks never call it) exists in three sibling repos. Measured 2026-08-07 by
+clean-tree `generate.py` dry-runs with PyAutoHands `2a4fb11`:
+
+| Repo | Modified notebooks | setup_notebook flips | Other diff lines |
+|---|---|---|---|
+| @autogalaxy_workspace | 127 | 126 | **6 — audit before committing** |
+| @autofit_workspace | 31 | 32 | 0 |
+| @HowToFit | 13 | 13 | 0 |
+| @HowToLens | 0 | — | — |
+| @HowToGalaxy | 0 | — | — |
+
+## Proposed fix
+
+Mirror of autolens_workspace#481: one dedicated regeneration sweep per repo
+(three PRs under this one task, mge-sigma phase-2 precedent), committing the
+clean-main `generate.py <project>` notebook diff. In autogalaxy_workspace,
+audit the 6 non-flip diff lines first — in autolens_workspace the analogous
+extras were harmless JSON-indent normalizations of hand-edited lines, but
+verify before committing.
+
+## Verification (per repo)
+
+- Second regeneration on the swept tree is a no-op.
+- No diff content beyond the setup_notebook flip + audited/explained extras.
+- Each repo's navigator/catalogue checks stay green.
+
+## Provenance
+
+Sibling check registered in the notebook-setup-notebook-regen-drift task
+(autolens_workspace#480), run 2026-08-07 while that task's PR was in CI.