mind: retire 13 prompts whose work already shipped

draft/docs/workspaces/extra_galaxies_feature_parity.md → complete/2026/07/extra-galaxies-feature-parity.md

@@ -1,3 +1,29 @@
+# extra_galaxies feature parity — the parent prompt, arc complete
+
+- completed: 2026-07-30 — ARC COMPLETE, recorded phase by phase.
+- phases (all merged):
+  | Phase | Issue | PR |
+  |---|---|---|
+  | 1 — point_source | autolens_workspace#374 closed | autolens_workspace#376 (`e005caca`) |
+  | 2a — autogalaxy multi_galaxy | autogalaxy_workspace#182 closed | autogalaxy_workspace#184 (`78568683`) |
+  | 2b — autolens multi_galaxy | autolens_workspace#387 closed | autolens_workspace#391 (`9b274ff2`) |
+- phase records: `complete/2026/07/extra-galaxies-point-source.md`,
+  `complete/2026/07/extra-galaxies-multi-galaxy.md`,
+  `complete/2026/07/extra-galaxies-multi-galaxy-lens.md`.
+- summary: the user's scoping held throughout — `imaging/features/extra_galaxies`
+  and its interferometer twin were the standard to mirror; `point_source/features`
+  got the same style without a `slam.py`; `multi_galaxy` got it in both workspaces;
+  `group/` and `cluster/` were deliberately left untouched.
+
+## Lifecycle note
+
+`complete/2026/07/extra-galaxies-multi-galaxy-lens.md` already names this file as
+its parent and declares the arc complete, but the parent prompt itself was never
+advanced, so it kept rendering as pickable backlog. Recorded here by the 2026-08-24
+completed-prompt reconciliation sweep.
+
+## Original prompt
+
 # extra_galaxies feature parity: point_source + multi_galaxy (both workspaces)
 
 Type: docs

draft/research/libraries/python_312_minimum.md → complete/2026/07/python-312-minimum-census.md

@@ -1,3 +1,29 @@
+# Adopt Python 3.12 as the PyAuto ecosystem minimum — the census
+
+- completed: 2026-07-30 — the census this prompt asked for was produced (it is the
+  body below), and the migration it planned then shipped in full.
+- implementation record: `complete/2026/07/python-312-floor.md` — every phase, from
+  the five core library floors through build/health machinery, a live
+  `2026.7.29.2` PyPI release, the independent downstream repos and the long-tail
+  selector census. The whole stack declares and enforces `Requires-Python >=3.12`.
+- parent issues: PyAutoNerves#142 (with companions PyAutoCTI#100, PyAutoReduce#59,
+  PyAutoHeart#113), all closed 2026-07-30.
+- sibling records: `python-312-workspace-runtime-pins.md`,
+  `python_312_floor_phase_4b_autoreduce.md`, `python-312-release-surfaces.md`,
+  `release-check-e-python-312.md`, `python-312-autofit-wiki-currency-ci.md`.
+- still open, deliberately: `draft/bug/pyautoreduce/published_autoreduce_09_missing_312_floor.md`
+  — the `autoreduce 0.9` wheel already on PyPI predates the floor and needs a
+  release, not a code change.
+
+## Lifecycle note
+
+A research/census prompt whose acceptance criteria were all met by the shipped
+migration; it was never advanced out of `draft/`, so it kept rendering as pickable
+backlog carrying the same verbatim user request the migration was executed from.
+Recorded here by the 2026-08-24 completed-prompt reconciliation sweep.
+
+## Original prompt
+
 # Adopt Python 3.12 as the PyAuto ecosystem minimum
 
 Filed: 2026-07-28 (backfilled from git)

draft/docs/workspaces/unify_ai_assistant_workspace_readmes.md → complete/2026/07/unify-ai-assistant-workspace-readmes.md

@@ -1,3 +1,28 @@
+# Phase 2: workspace READMEs made assistant-first
+
+- shipped: 2026-07-24, inside the phase-1 task rather than as a separate PR —
+  see `complete/2026/07/assistant-first-docs.md`.
+- issue: https://github.com/PyAutoLabs/PyAutoLens/issues/645
+- workspace-prs: https://github.com/PyAutoLabs/autolens_workspace/pull/329,
+  https://github.com/PyAutoLabs/autogalaxy_workspace/pull/155
+  (library legs PyAutoLens#646 + PyAutoGalaxy#521 were phase 1).
+- summary: this prompt's whole scope — one concise **AI Assistant** route first in
+  each workspace README's Getting Started, **Three Ways to Learn PyAutoLens**
+  removed, Colab/notebook/browsable-example routes preserved, remaining links
+  described as human-readable documentation — landed as the workspace leg of
+  `assistant-first-docs`, which unified library README, Read the Docs *and*
+  workspace onboarding in one pass. The PyAutoGalaxy assistant URL was
+  deliberately allowed to 404 until its repository exists.
+
+## Lifecycle note
+
+Phase 1 recorded the workspace PRs but the phase-2 prompt file was never advanced,
+so it kept rendering as pickable backlog long after the work merged. Recorded here
+by the 2026-08-24 completed-prompt reconciliation sweep — the split into phases 1/2
+was a planning artefact that the shipping session collapsed.
+
+## Original prompt
+
 # Phase 2: Make workspace READMEs assistant-first
 
 Filed: 2026-07-24 (backfilled from git)

draft/research/pyautobrain/explore_dashboardify_the_brain_s_operational_sur.md → complete/2026/08/dashboardify-brain-operational-surfaces.md

@@ -1,3 +1,36 @@
+# Explore: dashboardify the Brain's operational surfaces — resolved, and built
+
+- completed: 2026-08-23 — investigation resolved *and* the implementation it led to
+  shipped; the Brain board has been iterated on several times since.
+- write-up (the deliverable this prompt asked for):
+  `PyAutoMind/docs/pyautobrain/brain_board_assessment.md` — inventory of
+  `/wake_up`'s legs and where each landed, what was deliberately NOT dashboarded,
+  and the sibling-board architecture.
+- human direction at unpark (2026-08-23): build the PyAutoBrain dashboard, move the
+  `/wake_up` routine onto it, make it the morning starting point, and run the local
+  repo sync as a terminal command. So the "recommend a subset, or none" outcome was
+  overtaken — the answer was "build it".
+- what is live on PyAutoBrain `main`: `board/_board.py` (+ `_publish.py`, `_theme.py`,
+  `board.sh`, `gallery/`), `.github/workflows/brain_board.yml` publishing the board
+  each morning, `bin/morning.sh` as the local sync/clean leg, and `/wake_up` demoted
+  to the fallback door — all documented in PyAutoBrain's `AGENTS.md` § command
+  surface. Later PRs #258–#263 (morning timer, cloud hygiene, organ logos, the
+  test-performance surface, the board gallery) have built on it.
+- related records: `complete/2026/08/actionable-health-board.md` (the Heart board
+  this pattern came from), `complete/2026/08/one-tap-dashboard-rollout.md`,
+  `complete/2026/08/organism-board-final-readmes.md` (the arc's finale).
+- follow-ups still open: `draft/feature/pyautobrain/brain_board_follow_ups.md`
+  ("what real mornings surface", filed 2026-08-23).
+
+## Lifecycle note
+
+The prompt was updated in place on 2026-08-23 with "investigation resolved;
+implementation begun" but never advanced out of `draft/`, so it kept rendering as
+pickable backlog for a board that already exists. Recorded here by the 2026-08-24
+completed-prompt reconciliation sweep.
+
+## Original prompt
+
 # Explore: dashboardify the Brain's operational surfaces with pasteable conductor prompts
 
 Type: research

complete/2026/08/delaunay-nan-probe-fix.md

@@ -1,3 +1,25 @@
+<<<<<<<< HEAD:complete/2026/08/delaunay-nan-probe-ell-comps-validator.md
+# The delaunay.py NaN-poisoning probe died in ell_comps validation
+
+- shipped: 2026-08-24 — autolens_workspace_test#277 (merged `995f0c9`)
+- classification: bug (autolens_workspace_test + PyAutoGalaxy) — the JAX NaN-lane
+  probe in `scripts/interferometer/jax_likelihood/delaunay.py`
+- validation: retime run 32759921650 — NEITHER on both matrix legs, full PASS
+  sequence restored on the script.
+- summary: `instance_from_vector` handed the deliberate all-NaN vector's
+  `ell_comps` to autogalaxy's `validate.validate_ell_comps`, which raised
+  `ModelParameterException` at instance construction — so the probe exited 1
+  before it could test the thing it exists for (that an invalid Delaunay mesh
+  reaches the raw interferometer likelihood as NaN rather than crashing or
+  returning a plausible number). Fixed probe-side.
+
+## Lifecycle note
+
+The shipping session closed the prompt out in place (Mind commit `9f5e6972`,
+`Status: shipped 2026-08-24`) but never advanced it out of `draft/`, so it kept
+rendering as pickable backlog on the dashboard. Recorded here by the 2026-08-24
+completed-prompt reconciliation sweep; no further work.
+========
 # The delaunay NaN probe injected above the boundary the validator guards
 
 autolens_workspace_test#277 (`995f0c9`), shipped 2026-08-24 — the first of the
@@ -45,6 +67,7 @@ None of its own. The sibling bug found by the same sweep —
 `gradient_eager_jit_divergence_py313.md`, root-caused to an NNLS
 `while_loop` branch flip — remains in `draft/` with its fix direction
 recorded (pin `nnls_solver_tol`/`max_iter`).
+>>>>>>>> origin/main:complete/2026/08/delaunay-nan-probe-fix.md
 
 ## Original prompt
 

draft/maintenance/pyautomind/history_policy_generated_block.md → complete/2026/08/history-policy-generated-block.md

@@ -1,5 +1,36 @@
 # Single-source the "Never rewrite history" policy as a generated AGENTS.md block
 
+- shipped: 2026-08-23 (Mind `18a6a82b`).
+- classification: maintenance (PyAutoMind + every repo in `repos.yaml`).
+- what landed, exactly as the prompt specified:
+  - the canonical text lives in a file, not a code constant —
+    `PyAutoMind/policy/never_rewrite_history.md` (`HISTORY_POLICY_FILE`), the
+    recommendation the prompt made so a safety policy is editable without
+    touching generator code;
+  - `scripts/repos_sync.py` gained `load_history_policy()` +
+    `HISTORY_BEGIN`/`HISTORY_END` (`<!-- repos_sync:history:begin/end -->`) and
+    writes the block into every checked-out repo's `AGENTS.md` on `--write`,
+    opt-in by markers and tolerant of absent repos, mirroring the `system_map`
+    rollout;
+  - `check_history_blocks()` is a registered check in `main()`, so a stale block
+    fails the drift guard rather than rotting.
+- the guardrail was not weakened: the full `## Never rewrite history` section is
+  still inline and always-loaded in every `AGENTS.md` (verified in PyAutoMind and
+  PyAutoBrain), and `PyAutoMind/AGENTS.md` § Hard rules keeps its one-line summary.
+  The 2026-07-12 proposal to simply delete the section stays rejected.
+- epic context: `complete/archive/epics/agents_md_standardization.md`, whose fourth
+  generated block this was.
+
+## Lifecycle note
+
+Shipped without a completion record and never advanced out of `draft/`, so it kept
+rendering as pickable backlog against machinery that is already on `main` and
+CI-gated. Recorded here by the 2026-08-24 completed-prompt reconciliation sweep.
+
+## Original prompt
+
+# Single-source the "Never rewrite history" policy as a generated AGENTS.md block
+
 Type: maintenance
 Target: PyAutoMind
 Repos:

draft/bug/autolens/interferometer_delaunay_nonpd_fitexception.md → complete/2026/08/interferometer-delaunay-nonpd-fitexception.md

@@ -1,3 +1,36 @@
+# interferometer Delaunay — non-PD FitException in test-mode bypass
+
+- completed: 2026-08-07 — both halves merged; the script is un-parked and in smoke.
+- prs: **PyAutoFit#1408** (phase 1 — `PYAUTO_TEST_MODE=2` bypass tolerates a
+  single-eval `FitException` as a resample-to-reject sentinel) and
+  **autolens_workspace#311** (phase 1 — un-park
+  `interferometer/features/pixelization/delaunay` + add it to smoke); then
+  **PyAutoArray#436** (merged `efaf3041`, phase 2 — the numerical producer fix,
+  CI green first run on both matrix legs).
+- implementation record: `complete/2026/08/interferometer-delaunay-flaky-fitexception.md`,
+  which shipped under the sibling filing
+  `draft/bug/autoarray/interferometer_delaunay_intermittent_qhull_nan.md`
+  (correct target: the autoarray Delaunay interpolator, not autolens).
+- the actual producer, against this prompt's guess: not the `fnnls.py:134`
+  divide-by-zero (599 exercises, zero degenerate denominators — those warnings are
+  a downstream symptom) but `cholesky_funcs.py -> cholinsertlast`, whose unchecked
+  `math.sqrt(x[index] - S12.dot(S12))` goes imaginary when two near-coincident
+  source-mesh vertices make the Schur complement zero to within rounding
+  (measured ±1.776e-15, 8·eps). This prompt's own instinct — *fix the producer,
+  do not swallow the FitException* — is what shipped.
+- earlier false start on the record: `complete/2026/07/drop-interferometer-delaunay-marker.md`
+  (marker-removal PR autolens_workspace#310 closed without merge; three passing
+  local runs are not enough to call a ~50%-flaky script fixed).
+
+## Lifecycle note
+
+Two prompts were filed for one defect from different repos. The autoarray-targeted
+one advanced correctly; this autolens-targeted duplicate stayed in `draft/` and kept
+rendering as pickable backlog. Recorded here by the 2026-08-24 completed-prompt
+reconciliation sweep.
+
+## Original prompt
+
 # interferometer Delaunay pixelization — non-PD FitException in test-mode bypass
 
 Type: bug

draft/feature/autolens_profiling/jax_compile_time_profiling.md → complete/2026/08/jax-compile-time-profiling-absorbed.md

@@ -1,3 +1,32 @@
+# Profile and speed up JAX likelihood-function compile times — absorbed, then shipped
+
+- completed: 2026-08-10 — absorbed into the compile-axis arc and shipped there; no
+  work remained under this slug.
+- absorbed-into: `draft/feature/profiling/compile_warm_baseline_dashboard.md`
+  (phase 2 of the compile-axis arc), shipped as
+  `complete/2026/08/compile-warm-baseline-dashboard.md` — autolens_profiling#104
+  (`355d555`) + PyAutoBrain#220 (`74d7b1b`). Merged there because the dashboard and
+  the Profiling Agent that reads it are two halves of one thing, and building them
+  from opposite ends risks incompatible schemas.
+- the earlier legs were already spent by then: the compile-time speed-up arc
+  (`#71 → #77`) closed the "can we make it faster" question — settings suffice
+  (persistent cache 117.0 s → 2.3 s; autotune-off 498 s → 29 s) — and the
+  search-transform measurement scope was executed by the MultiStartProdigy census
+  (autolens_profiling#93).
+- one claim of this prompt is overturned on the record: its "autotuning ruled out"
+  section was wrong — that A/B never actually flipped the flag. See
+  `autolens_profiling/scripts/misc/jax_compile/README.md`.
+- arc record: `profiling-agent-compile-axis-arc`.
+
+## Lifecycle note
+
+Banner-marked "ABSORBED 2026-08-10 — do not start from this file" but left in
+`draft/` at `Priority: high`, so it kept rendering in the dashboard's top pick list.
+Recorded here by the 2026-08-24 completed-prompt reconciliation sweep; the
+measurements and overturned-claim trail are preserved in the original prompt below.
+
+## Original prompt
+
 # Profile and speed up JAX likelihood-function compile times (all use cases)
 
 > **ABSORBED 2026-08-10 — do not start from this file.** The surviving scope (the

draft/maintenance/autolens_workspace/latex_docstrings_invalid_escape_warnings.md → complete/2026/08/latex-docstrings-invalid-escape-warnings.md

@@ -1,3 +1,35 @@
+# LaTeX in non-raw docstrings — superseded, and the wider fix shipped
+
+- outcome: SUPERSEDED 2026-08-20; the defect it reported was then fixed at wider
+  scope on the same day.
+- superseded-by: `draft/maintenance/workspaces/latex_raw_string_docstrings.md`,
+  shipped as `complete/2026/08/latex-raw-string-docstrings.md` —
+  autolens_workspace#491, six PRs merged 2026-08-20 (HowToFit#47, HowToGalaxy#70,
+  HowToLens#73, autofit_workspace#145, autogalaxy_workspace#218,
+  autolens_workspace#492): 41 files, 180 literals, 131 silent corruptions
+  repaired. Prerequisite `hands-raw-string-docstring-prefix` (PyAutoHands#251)
+  cleared first.
+- why it was superseded rather than run: same defect, same fix, wider scope. The
+  2026-08-20 survey swept all six workspace repos; the 17 `autolens_workspace`
+  files it listed are exactly the 17 measured here (top-six warning counts match
+  line for line), and it found a second, *silent* class this filing never saw —
+  `\t` in `\theta`, `\f` in `\frac`, escapes Python DOES recognise, so they corrupt
+  the string with no warning at all.
+- carried across, so nothing was lost: the interpreter trap (§ TRAP) became the
+  successor's sweep spec and verification gate, and "do not reword the LaTeX or the
+  prose" became its § Constraints.
+- origin trail kept below: autolens_workspace#457 / PR#459, and the 2026-08-09
+  measurement (80 warnings across 17 files on `9974f891`).
+
+## Lifecycle note
+
+Marked `Status: SUPERSEDED` in place but left in `draft/`, so it kept rendering as
+pickable backlog after the successor merged. Recorded here by the 2026-08-24
+completed-prompt reconciliation sweep — which is also what widened the dashboard's
+drift detector to read a prompt's own `Status:` header, the signal that surfaced it.
+
+## Original prompt
+
 # LaTeX in non-raw docstrings emits SyntaxWarning: invalid escape sequence
 
 Type: maintenance

draft/bug/autolens_workspace_test/multi_dataset_jax_likelihood_xla_stall.md → complete/2026/08/multi-dataset-jax-likelihood-xla-stall.md

@@ -1,3 +1,26 @@
+# multi_dataset/jax_likelihood XLA compile stall — superseded by the jax-compile-stall epic
+
+- outcome: SUPERSEDED 2026-08-23, never issued — no PR, no code change under this slug.
+- superseded-by: `draft/bug/ci/jax_vmap_jit_compile_stall.md` (the `jax-compile-stall`
+  epic ledger), whose phases carry the same defect with cross-repo evidence.
+- shipped elsewhere: phase 1 `complete/2026/08/jax-compile-stall-evidence.md`;
+  phase 2 `complete/2026/08/jax-compile-stall-slow-vs-stall-audit.md`;
+  phase 3 remains open as `draft/bug/ci/jax_compile_stall_3_root_cause.md`.
+- summary: filed 2026-08-22 after `multi_dataset/jax_likelihood/mge.py` was parked
+  out of the smoke gate. One root cause is one task, so the filing was folded into
+  the epic the following day. The two asks this file carried that the epic did not
+  — dump a stack trace on timeout (solved library-side in phase 1) and re-enable
+  `mge.py` / `shared_preloads.py` in `smoke_tests.txt` (phase 3) — were both folded
+  in, so nothing was lost.
+
+## Lifecycle note
+
+Marked `Status: superseded` in place on 2026-08-23 but left in `draft/`, so it kept
+rendering as pickable backlog. Recorded here by the 2026-08-24 completed-prompt
+reconciliation sweep. Its evidence is preserved in the original prompt below.
+
+## Original prompt
+
 # multi_dataset/jax_likelihood scripts hang to the timeout cap (XLA compile stall)
 
 Type: bug

draft/research/autofit/priors_and_messages_math_audit.md → complete/2026/08/priors-and-messages-math-audit.md

@@ -1,3 +1,35 @@
+# Census of priors and messages — the audit, fully graduated
+
+- completed: 2026-08-18 — retired as a historical record; nothing was left to
+  action from this file.
+- A1–A9, all nine confirmed bugs: FIXED on PyAutoFit `main`. Batch 1 (A1–A5, A8)
+  via #1344 / PR#1345 (merged `c0b6c94b8`, 2026-07-10); batch 2 (A6, A9) via
+  #1346 / PR#1348 (merged `cf0cc4bbb`, 2026-07-10); A7 vectorised in PR#1345. All
+  five open decisions resolved on hub #1331 (2026-07-14 comment): Beta clamp →
+  raise; σ<0 rejected / σ=0 kept; Gamma `from_mode` matches mean+variance;
+  normalisation → Option A (drop constants + a `Prior.log_normalisation()` hook);
+  width modifier → `abs(mean)` + opt-in floor + `PriorException`. Locked in by
+  `test_priors_messages_fixes_1331.py` and `test_prior_width_safety.py`.
+- records: `complete/2026/07/priors-messages-fixes.md`,
+  `complete/2026/08/prior-width-safety.md`,
+  `complete/2026/07/ep-priors-fable-reassess.md`,
+  `complete/2026/08/prior-property-tests.md`,
+  `complete/2026/08/transformed-message-semantics-doc.md`.
+- C-findings: C2 shipped with the A5 decision (PR#1345); C6's reversal-convention
+  half via #1333 / PR#1334, its `LinearShiftTransform` half 2026-08-18 via PR#1499;
+  C3 shipped 2026-08-18 as the property sweep (PR#1499, `21288bb`).
+- still tracked elsewhere, deliberately: the remaining C-findings live as
+  `draft/bug/priors/` prompts (09, 11–14) under the tracker
+  `draft/bug/priors/z_features.md`, whose phase 4 is open.
+
+## Lifecycle note
+
+Marked `Status: RETIRED — historical record` in place on 2026-08-18 but left in
+`draft/` at `Priority: high`, so a spent census kept rendering near the top of the
+dashboard. Recorded here by the 2026-08-24 completed-prompt reconciliation sweep.
+
+## Original prompt
+
 # Census of priors and messages — confirmed bugs + redesign ideas
 
 Type: research