prompt: tracer-fits-existence-guard shipped (autolens_workspace#503)

autonomy_log.md

@@ -242,3 +242,4 @@ Outcome ∈ `merged-unchanged` / `amended` / `rejected` / `parked` /
 | 2026-08-01 | nautilus 1-core serial-pool — corrective for Heart RED "release validation FAILED (stage integrate)" (PyAutoFit#1442 / PR#1443) | corrective-PR exception; authorization = human session instruction quoted verbatim on #1442, given at launch ("fine if any blockers need sorting") rather than after the reason string was surfaced — deviation noted | regression test red-on-main/green-patched; test_autofit/non_linear 411p; py-spy deadlock stacks; control-vs-patched release-env run | corrective (PR-open; merge under the same session authorization; validation = Stage 2/3 dispatch + nightly on fresh wheels) |
 | 2026-08-24 | refactor-witness-map-audit (PyAutoBrain#269 / PR#271) | safe (header safe, maintenance cap safe) | tests PASS 468 (no downstream — organ repo, nothing imports Brain) / smoke n/a (organism repo, no script surface) / review CLEAN, lifted claim basis-cited by blobless tree reads of all 10 repos / heart NOT EVALUATED — pyauto-heart unreachable in this web-github session, no verdict obtained | merged-unchanged (PR#271 merge cf969f3 after #268, human-authorized via /prm; CI Brain Tests green on 2846b65; retargeted base main before merge — GitHub only auto-retargets a stacked PR on base-branch DELETION, not on merge) |
 | 2026-08-24 | pyautoconf-rename-leftovers (PyAutoBrain#267 / PR#268) | supervised (header supervised, bug cap supervised) | tests PASS 467 (no downstream — organ repo) / smoke n/a (organism repo, no script surface) / review CLEAN, 3 claim dispositions incl. one demoted to idle (a test count cited from a Mind record, not re-run) / heart NOT EVALUATED — pyauto-heart unreachable in this web-github session | parked at ship sign-off per the supervised contract, then PR-open on human sign-off same session; merged-unchanged (PR#268 merge 2e7ea30, human-authorized via /prm; CI Brain Tests green on 2088bad) |
+| 2026-08-24 | tracer-fits-existence-guard (autolens_workspace#503) | supervised (header safe, bug cap supervised) | tests n/a (workspace repo, no test dir) / smoke reproduced-then-fixed: imaging/modeling.py + multi_galaxy/modeling.py exit 1 with the reported FileNotFoundError under profile_smoke + PYAUTO_TEST_MODE=1 before, exit 0 after; TEST_MODE=2 default green; check_sizes.sh OK / review not run — parked before ship, so no ReviewSurface / heart NOT EVALUATED — pyauto-heart unreachable in this web-github session | parked at ship sign-off per the supervised contract, then PR-open on human sign-off same session (/prm); merged-unchanged (PR#504 merge b19750c, human-authorized via /prm; CI green on 0c14d00 — 3 runs / 7 jobs) |

complete/2026/08/tracer-fits-existence-guard.md

@@ -0,0 +1,85 @@
+- issue: https://github.com/PyAutoLabs/autolens_workspace/issues/503
+- completed: 2026-08-24
+- workspace-pr: autolens_workspace#504 (merged b19750cad64d2b5a0d809c132086371dccde27bc -> main)
+- what shipped: the "Loading From Output Folder" teaching section in `scripts/imaging/modeling.py` guarded on `files/tracer.json` and then, inside that branch, read `image/tracer.fits`. Split into two independent guards, one per file, so each read is gated on the file it actually reads. Same fix applied to `scripts/multi_galaxy/modeling.py`; both notebook mirrors regenerated.
+- the actual bug: under TEST_MODE the reduced search writes `files/tracer.json` but NOT the visualisation output `image/tracer.fits`, so the guard passed and the read raised FileNotFoundError. Long-standing (present since cfd5334c, 2026-04-14), not a regression. Never blocked a release — the `release` job's `needs:` excludes `run_smoke_tests` — but left the smoke gate permanently red, so a genuine workspace regression would have looked identical to this known failure.
+- guard shape is now canonical: guard-each-read-on-its-own-file, matching what `scripts/guides/results/start_here.py` already did. `draft/maintenance/workspaces/read_through_issues.md` asks for this section to be ADDED to the group/cluster/interferometer examples — this is the shape to propagate, or the same bug ships to those scripts.
+- scope decision: `scripts/multi_galaxy/modeling.py` was NOT in the reported scope but carried the byte-identical block and is also on `smoke_tests.txt`. Fixing only `imaging/` would have left the smoke gate red anyway, so both went in one PR.
+- trap for reproducing: the smoke profile's DEFAULT `PYAUTO_TEST_MODE=2` bypasses the sampler and writes no `tracer.json` at all, so the guard is never entered and the bug does NOT surface. `PYAUTO_TEST_MODE=1` (reduced search) is the mode that writes `tracer.json` but not `tracer.fits`. Anyone re-checking this under the plain smoke default will wrongly conclude it is already fixed.
+- validation: reproduced-then-fixed under profile_smoke.yaml + PYAUTO_TEST_MODE=1 — both scripts exit 1 with the reported FileNotFoundError before, exit 0 after. CI green on the merged head 0c14d00 across 3 workflow runs / 7 jobs (Smoke Tests changes+3.12+3.13, Navigator Check x3, Script Size Guard); mergeable_state clean at merge.
+- heart: NOT EVALUATED — `pyauto-heart` is unreachable from a web-github session, so leg 4 of the ship gate has no verdict. Stated on the PR and issue rather than assumed green.
+- environment note: run entirely in a cloud web-github session — no worktree, no `activate.sh`. autolens needs Python >= 3.12 (the PyPI 2026.7.29.1.post1 sdist raises a RuntimeError below that as a deliberate floor guard) while the session default was 3.11; a uv venv on /usr/bin/python3.12 was needed. `.github/scripts/run_smoke.py` could not be used (needs a sibling PyAutoHands checkout on PYTHONPATH), so profile_smoke.yaml's env was applied by hand. Notebook regeneration DID work by cloning PyAutoHands read-only and running `autohands/generate.py autolens` with the venv's bin on PATH — it rewrote only the two changed notebooks, no catalogue or unrelated drift.
+- gotcha for future cloud sessions: an `add_repo` clone arrives with `remote.origin.fetch = +refs/heads/main:refs/remotes/origin/main`, so `git push -u` sets the branch remote but can never create the `origin/<branch>` tracking ref — the stop-hook git check then reports pushed commits as unpushed. Fix is `git config remote.origin.fetch '+refs/heads/*:refs/remotes/origin/*'` then fetch.
+- follow-up not done (deliberately out of scope): the hard-coded `pixel_scales=0.1` / `0.05` in these blocks are the same literal pattern autolens_workspace#502 replaced with `dataset.pixel_scales` elsewhere.
+- autonomy: `--auto` launch, effective level supervised (prompt header safe, bug work-type cap supervised). Parked at ship sign-off per the contract; human typed /prm, which authorized PR-open, merge and the full close-out.
+
+## Original prompt
+
+# Bug: fix the tracer.fits existence guard in autolens_workspace imaging modeling.py
+
+Type: bug
+Target: workspaces
+Repos:
+- autolens_workspace
+Difficulty: small
+Autonomy: safe
+Priority: normal
+Status: formalised
+Filed: 2026-08-22 (backfilled from git)
+Issued: 2026-08-24
+
+Bug: fix the tracer.fits existence guard in autolens_workspace imaging modeling.py. The script crashes with FileNotFoundError because the guard checks the wrong file. It tests whether files/tracer.json exists, then reads image/tracer.fits, which TEST_MODE never writes. This fails the release smoke gate.
+
+
+## Evidence (2026-08-23, gathered during the 2026.8.23.1 release)
+
+`autolens_workspace/scripts/imaging/modeling.py:641`. The guard and the read
+name **different files**:
+
+```python
+if (result_path / "files" / "tracer.json").exists():
+    tracer = from_json(file_path=result_path / "files" / "tracer.json")
+
+    tracer_fits = al.Array2D.from_fits(
+        file_path=result_path / "image" / "tracer.fits", hdu=0, pixel_scales=0.1
+    )
+```
+
+Under TEST_MODE the search is reduced, so `files/tracer.json` IS written but the
+visualisation output `image/tracer.fits` is not. The guard passes and the read
+raises:
+
+```
+FileNotFoundError: .../output/test_mode/imaging/simple/modeling/<id>/image/tracer.fits
+FAIL: imaging/modeling.py
+```
+
+- Present since `cfd5334c` (2026-04-14); the script is untouched since 2026-08-04,
+  so this is long-standing, not a regression from the regime-stamp release.
+- **TEST_MODE-specific.** The same script passes at release fidelity — Heart's
+  Stage 3 integrate ran `run_scripts (3.12, autolens, imaging)` green
+  (672p/0f), because a full-fidelity fit does write `tracer.fits`.
+- Reproduced in PyAutoHands `release.yml` run 32542888112, job
+  `run_smoke_tests (3.12, autolens_workspace)`.
+
+## Why it matters (and why it is NOT urgent)
+
+It does not block releases: the `release` job's `needs:` is
+`[resolve_mode, release_test_pypi, version_number]` and excludes
+`run_smoke_tests`, so the publish succeeds regardless — run 32542888112 is marked
+`failure` yet published 2026.8.22.1. The real cost is that it leaves the smoke
+gate permanently red, so a genuine workspace regression would look identical to
+this known failure and be ignored.
+
+## Traps
+
+- Do **not** "fix" this by deleting the result-loading block. It is a documented
+  teaching section, and `draft/maintenance/workspaces/read_through_issues.md`
+  separately asks for the equivalent section to be **added** to the other
+  `modeling.py` examples (group, cluster, interferometer). Coordinate with that
+  prompt — whatever guard shape is chosen here should be the one propagated
+  there, or the same bug ships to four more scripts.
+- Guarding on `image/tracer.fits` alone would silently skip the `tracer.json`
+  load in test mode. Prefer guarding each read on the file it actually reads.
+
+<!-- formalised by the Intake (Conception) Agent on 2026-08-22 from user-intake -->

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.
 
-1123 records across 7 buckets.
+1124 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -233,6 +233,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [test-mode-samples-info-hook-contract](2026/08/test-mode-samples-info-hook-contract.md) — `NonLinearSearch._test_mode_samples_info()`'s docstring told subclasses
 - [test-performance-board](2026/08/test-performance-board.md)
 - [testmode-assertion-note-removal](2026/08/testmode-assertion-note-removal.md) — Follow-up to test-mode-bypass-assertion-ties (PyAutoFit#1520, merged
+- [tracer-fits-existence-guard](2026/08/tracer-fits-existence-guard.md)
 - [transformed-message-semantics-doc](2026/08/transformed-message-semantics-doc.md)
 - [undo-community-file-declutter](2026/08/undo-community-file-declutter.md) — Undid the 2026-08-19 community-file declutter (#248 Mind, #32 Memory):
 - [uniform-prior-bounds-numpy-path](2026/08/uniform-prior-bounds-numpy-path.md) — auto-closed by the merge's `Closes` line