Release-fidelity validation: 14 workspace script legs fail under mode=release (run 28784914443)

[Jammy2211]

Summary

The first real end-to-end release-validation run surfaced 14 failing run_scripts legs (of ~28) in workspace-validation.yml mode=release. This is a tracking issue for the workspace/library maintainers.

• Run: https://github.com/PyAutoLabs/PyAutoHeart/actions/runs/28784914443 (mode=release, 2026-07-06)
• Rehearsed wheels: autoconf / autoarray / autofit / autogalaxy / autolens == 2026.7.6.1.dev65001 (TestPyPI), built from the current library main HEADs.
• Not a packaging problem: all 5 wheels installed cleanly from TestPyPI, and verify_install A–E passed. The failures are genuine per-script failures under release fidelity (real sampler / full evaluation, no source on PYTHONPATH).
• Gate impact: pyauto-heart validate --ingest recorded release_ready=false → readiness is RED (release validation FAILED (stage integrate)). This is the expected, correct behaviour — the run caught real regressions.

Failing legs (project → directory)

Workspace	Failing script directories
autofit_workspace	cookbooks
autofit_workspace_test	database
autogalaxy_workspace	multi, guides, imaging, interferometer, ellipse
autogalaxy_workspace_test	ellipse, interferometer
autolens_workspace	multi, cluster, imaging, interferometer, guides

All other legs (autofit overview/model/plot/searches/simulators/jax_assertions/model_composition, autolens weak/point_source, autogalaxy_test aggregator, …) passed, as did verify_install_release.

Example root cause (autolens_workspace / imaging leg)

The imaging leg ran 102 scripts; all passed except two, which failed the whole leg:

• scripts/imaging/features/advanced/double_einstein_ring/chaining.py → TIMEOUT (300 s) — a heavy multi-search chaining pipeline exceeding the per-script cap under the release profile.
• scripts/imaging/features/advanced/shapelets/modeling.py → FAIL (~77 s) — a JAX runtime exception ("JAX has removed its internal frames from the traceback … Set JAX_TRACEBACK_FILTERING=off").

The pattern (a small number of long-running / JAX-path scripts per leg) likely explains most of the other 13 legs, but only the autolens/imaging leg was drilled into for specific script names — the per-script failures for the remaining legs are in the run's per-job logs (results-scripts-<project>-<directory> artifacts + each run_scripts job log).

Suggested next steps

1. For each failing leg, open its run_scripts job log in run 28784914443 and identify the specific failing/timed-out scripts.
2. Triage: genuine regression vs. release-profile timeout (the 300 s cap) vs. JAX-path issue on the packaged wheels.
3. Fix in the owning workspace/library repo; consider splitting per-workspace sub-issues if ownership differs.
4. Re-run the release-validation once fixed (Brain Release Agent: pyauto-brain release validate) to confirm readiness returns to GREEN.

Provenance

Surfaced by the PyAutoBrain health conductor + release conductor driving the full Stages 0–3 release-validation pipeline end-to-end (rehearsal → integrate → ingest → re-judge). No workspace/library code was modified — this issue only reports the regressions.