fix: consolidate guides/results fits onto _quick_fit.py (IndexError)

[Jammy2211]

Overview

The two scripts/guides/results/aggregator/*.py tutorials crash with IndexError: list index out of range under the release-validation profile (surfaced in the 2026-07-13 release tail, PyAutoHeart#72, cluster E). Reproduced on current main — it is not covered by PyAutoFit#1368 (the release profile unsets PYAUTO_TEST_MODE for guides/results/, so is_test_mode() is False and the output-routing fix does not apply).

Root cause: three results tutorials each run their own capped n_like_max=300 Nautilus fit, but only _quick_fit.py disables samples_weight_threshold. The others prune to 2 samples on disk, and the tutorials then deep-index (from_sample_index(-10), parameter_lists[9]) assuming ≥10 samples → IndexError. It is order-dependent: start_here.py runs first and plants the 2-sample fit that the aggregator readers later choke on. The scripts' docstrings already claim they "resume from the cached fit," but their model/search config diverges from _quick_fit, so they don't — they run a separate shallow fit.

Fix: consolidate onto a single fit source (_quick_fit.py). The self-fitting scripts keep their model/search/analysis written out inline (these are teaching docs — no shared "builder" imports), but the inline definition is made to match _quick_fit so search.fit() genuinely resumes its 300-sample fit. Verified: the fit identifier is [search, model, unique_tag] only (analysis-independent), so a matching inline model + plain AnalysisImaging reproduces _quick_fit's on-disk identifier 8bc8f101….

Plan

• Keep _quick_fit.py as the single fit producer (threshold None, Isothermal + shear, MGE source, AnalysisLatent); make its primary fit unique_tag suffix-free (simple__no_lens_light) plus one more fit for the aggregator's multi-result iteration. (Bonus: fixes queries.py, which queries unique_tag == "simple__no_lens_light" — currently matches nothing.)
• start_here.py + galaxies_fits.py: add shear=al.mp.ExternalShear to the lens (they already use the identical MGE source and unique_tag=dataset_name) and match _quick_fit's identifier-affecting search fields, so search.fit() resumes the 300-sample fit. Deep indexing becomes valid.
• samples.py: swap the inline Sersic source for the same MGE bulge + shear so it resumes the primary fit; update prose/prints describing the source.
• Correct the now-accurate "resumes from the cached fit" docstrings.
• Leave the read-only aggregator scripts (samples_via_aggregator, models, queries, data_fitting, interferometer, latent_variables) unchanged — they will only ever read 300-sample fits.
• Regenerate the affected notebooks; validate via the ordered guides/results release-profile run.

Detailed implementation plan

Affected Repositories

• autolens_workspace (primary)

Branch Survey

Repository	Current Branch	Dirty?
./autolens_workspace	main	clean

Suggested branch: feature/aggregator-quick-fit-consolidation

Implementation Steps

1. scripts/guides/results/_quick_fit.py — make the primary fit unique_tag=dataset_name (suffix-free) and keep a second distinct fit for aggregator iteration; keep conf.instance["output"]["samples_weight_threshold"] = None, the shear + MGE model, LatentShear/AnalysisLatent. Confirm identifier via the identifier harness.
2. scripts/guides/results/start_here.py — add shear=al.mp.ExternalShear to the lens af.Model; align the af.Nautilus config with _quick_fit's identifier-affecting fields; fix the docstring (~line 104) that claims resumption. Keep the _quick_fit auto-trigger guard.
3. scripts/guides/results/aggregator/galaxies_fits.py — same model/search alignment; it still needs dataset for the local al.FitImaging refit. from_sample_index(-10) (lines ~176/238) becomes valid.
4. scripts/guides/results/aggregator/samples.py — change source al.lp.Sersic → MGE bulge (matching _quick_fit) + add shear; align search; update prose/prints referencing .source.bulge. parameter_lists[9] / log_likelihood_list[9] become valid.
5. Regenerate notebooks: PYTHONPATH=../PyAutoBuild/autobuild python3 ../PyAutoBuild/autobuild/generate.py autolens.
6. Validate: run_python.py autolens scripts/guides/results --env-config config/build/env_vars_release.yaml (with the PyAuto stack on PYTHONPATH) → 0 failures; assert every on-disk fit has ~300 samples and no 2-sample simple__no_lens_light fit is created; spot-run smoke (env_vars.yaml).

Key Files

• scripts/guides/results/_quick_fit.py — the single fit producer (threshold None; canonical model).
• scripts/guides/results/start_here.py — results-guide entry; runs its own fit.
• scripts/guides/results/aggregator/galaxies_fits.py — from_sample_index(-10) crash site.
• scripts/guides/results/aggregator/samples.py — Sersic→MGE; [9] indexing.
• config/build/env_vars_release.yaml — release profile (guides/results/ unsets PYAUTO_TEST_MODE).

Testing approach

The ordered guides/results release-profile run is both the reproduction and the validation harness (without --report-dir it sys.exit(1)s at the first failure; with it, it records each). Identifier equality is checked with autofit.mapper.identifier.Identifier([search, model, unique_tag]).

Original Prompt

Click to expand starting prompt

Aggregator scripts crash with IndexError (empty results) on release wheels

Type: bug
Target: autolens_workspace
Difficulty: medium
Autonomy: supervised
Priority: normal

Surfaced burning down the 2026-07-13 release-validation tail (PyAutoHeart#72). On the release wheels under the release profile, both aggregator guide scripts fail fast with the same error:

scripts/guides/results/aggregator/galaxies_fits.py        ... FAIL (13.8s) IndexError: list index out of range
scripts/guides/results/aggregator/samples_via_aggregator.py ... FAIL (4.5s) IndexError: list index out of range

IndexError: list index out of range = the aggregator queried results and got an empty list, then indexed [0]. Before assuming a new bug, VERIFY it still fails on current main. Reproduce under the release profile from the autolens_workspace checkout on current main; the aggregator reads the output of a prior modeling run, so ensure that prerequisite output exists — the empty-list may be a missing/misrouted prerequisite output rather than an aggregator bug. If it reproduces: get the full traceback and fix the producer, not the [0] access. Release run: PyAutoHeart workspace-validation 29279095224, TestPyPI 2026.7.13.1.dev65601.

---

Investigation note (2026-07-14): reproduced on current main. Real crash site is galaxies_fits.py:176 samples.from_sample_index(sample_index=-10) → parameter_lists[-10] on a 2-element list; the shallow capped fit retained only 2 samples (no samples_weight_threshold = None), while _quick_fit's fits retain 300. See PyAutoHeart#72 cluster E.

[Jammy2211]

⏸️ Parked at ship sign-off (supervised --auto) — awaiting human decision on Heart RED

The fix is implemented and validated in the worktree (~/Code/PyAutoLabs-wt/aggregator-quick-fit-consolidation), but it is uncommitted and unpushed, because the autonomous-ship gate cannot proceed. This comment is the checkpoint.

What was done (validated)

Consolidated the guides/results tutorials onto a single fit source, _quick_fit.py, per the approved plan:

• _quick_fit.py — primary fit unique_tag made suffix-free (simple__no_lens_light) + one more (_1) for aggregator iteration; keeps samples_weight_threshold = None, shear + MGE model, AnalysisLatent.
• start_here.py, aggregator/galaxies_fits.py — added shear=al.mp.ExternalShear so their inline model matches _quick_fit; search.fit() now resumes its 300-sample fit.
• aggregator/samples.py — source Sersic → MGE bulge + shear; the two source.bulge.sersic_index filter examples retargeted to lens.shear.gamma_1.
• Notebooks regenerated for the 4 changed scripts (unrelated drift reverted).

Verification (fresh, current main, in the worktree):

• Identifier check: inline tutorial config computes aceb0b5a8cddadad1a489ab941e694cb, identical to _quick_fit's primary fit → resume confirmed (identifier is [search, model, unique_tag], analysis-independent).
• Ordered guides/results run under config/build/env_vars_release.yaml: 14/14 pass (previously galaxies_fits.py and samples_via_aggregator.py failed with IndexError).
• On disk: both fits retain 300 samples; no 2-sample fit is created. queries.py's unique_tag == "simple__no_lens_light" query now matches (was matching nothing — a latent bug this also fixes).

Diff: 8 files, +90/−22 (4 scripts + their 4 notebooks). Scope is exactly the consolidation — no unrelated changes.

Why it is parked — Heart is RED

Per AUTONOMY.md, Heart RED forbids commit, push and PR-open at every autonomy level, and the corrective-PR-for-RED exception "never fires under --auto" — it is a live human act. Heart's current verdict (verbatim):

verdict: RED
  - PyAutoLens: 1 commit(s) behind origin
  - release validation FAILED (stage integrate)
  - workspace validation not passing (3 failed, 2026-07-09T09-48-30Z)
  - 58 stale parked script(s)
  - PyAutoMind: open PR 14d old
  - install verification not run

This fix is causally corrective for the reason release validation FAILED (stage integrate) — the aggregator IndexErrors are cluster E of the release-validation tail (PyAutoHeart#72). It does not address the other five reasons, which remain.

Decision needed

To ship this, the corrective-PR-for-RED exception must be invoked by you, live. If you want to authorize it, reply on this issue with something like:

Authorize the corrective-PR exception for Heart RED reason "release validation FAILED (stage integrate)" to commit, push, and open one pending-release PR for this aggregator-consolidation fix (issue #274).

On that authorization I will: commit → push feature/aggregator-quick-fit-consolidation → open one pending-release PR (stating it clears only the aggregator/cluster-E portion of the stage-integrate reason; the other RED reasons remain, so Heart stays RED and release stays blocked) → record the four sinks. Merge stays a separate human act.

Otherwise this stays parked (awaiting-input); the validated changes are safe in the worktree.

[Jammy2211]

Corrective-PR exception for Heart RED — AUTHORIZED (human, 2026-07-14).

Authorize the corrective-PR exception for Heart RED reason «release validation FAILED (stage integrate)» to commit, push, and open one pending-release PR for #274.

Heart re-checked at ship time: still RED; the named reason release validation FAILED (stage integrate) is present and unchanged. Proceeding to commit → push → open one pending-release PR that clears only the aggregator/cluster-E portion of that reason. The other RED reasons remain, so Heart stays RED and release stays blocked. Merge remains a separate human act.

[Jammy2211]

Corrective PR opened: #275 (pending-release).

Under the authorized corrective-PR exception, this clears only the aggregator/cluster-E portion of the Heart RED reason release validation FAILED (stage integrate). The other RED reasons remain, so Heart stays RED and release stays blocked. Merge is a separate human act; after merge, build fresh wheels + re-run release-integration on main for a new Heart verdict. Issue kept open until then.

[Jammy2211]

Corrective PR #275 merged (squash c17ddfa5715d81b60cb3a607af08815f9257588e) by explicit human direction. All CI green (smoke 3.12/3.13, navigator, catalogue staleness).

Issue kept open: per the corrective-PR exception, release work resumes only after fresh post-merge wheels are built and release-integration validation is re-run on main for a new Heart verdict. This cleared the aggregator/cluster-E portion of release validation FAILED (stage integrate); the other RED reasons remain, so Heart stays RED and release stays blocked.

[Jammy2211]

Corrective validation GREEN — closing

The fresh mode=release validation run on the current main HEADs passed end-to-end:

• Stage 2 TestPyPI rehearsal v2026.7.14.1.dev66001 — all 5 wheels built + uploaded.
• Stage 3 wheel-based real-search integration — 542 passed, 0 failed, 0 timeout, 87 skipped.
• Ingested → release_ready: true; Heart is now YELLOW (no RED reasons remain).

This fix is on the validated main HEADs (commit_shas matched), so its corrective validation is satisfied. Closing.