[Jammy2211]

Summary

Records the human decision of 2026-11-06 on the critical-curves campaign: move toward the JAX zero-contour engine for critical curves and caustics drawn during modeling-run visualization, starting with a feasibility phase. The campaign goes needs-decision → active.

It files tasks/critical_curves_jax_modeling_viz_feasibility.md (status ready, research only, no default change). The study covers:

• correctness on phase-4 posterior draws, per scale, against the pre-registered two-pixel rule;
• cold and warm cost per visualization call, on CPU and A100;
• cheap detection of incomplete curves, with a marching-squares fallback;
• a GO/NO-GO per scale.

It also notes a docs drift: PyAutoGalaxy plot_utils calls zero_contour the default, but the config default is marching_squares.

No measurement, compute or default change.

Test Plan

• pytest tests (380 passed)
• bin/pyauto-pulse board --offline / check --offline: OK

🤖 Generated with Claude Code