[Jammy2211]

Summary

Files one profiling task (measurement-tools campaign): tasks/numba_num_threads_pin_and_delaunay_nn_remeasure.md plus its campaigns.yaml row and the regenerated board.

autolens_profiling/_production_config.pin_thread_env pins BLAS/OMP threads but not NUMBA_NUM_THREADS, and numba ignores OMP_NUM_THREADS. So the day-zero optimal cell imaging/delaunay_nn/hst numba_cpu_sparse_fp64@hpc_ral_cpu ran its then-always-parallel Sibson kernel on up to 8 cores. On the laptop at the profiled size, Sibson takes 14 ms with 8 threads and 56 ms with 1, about 3–10% of the 1.3 s call. Production workers are single-core and, since PyAutoArray#677, always serial.

Steps:

1. Pin NUMBA_NUM_THREADS (no compute).
2. Re-measure after the release carrying #677 (RAL go needed).
3. A human decision on the day-zero designation.

No other numba cell is affected; the task lists why.

Test Plan

• bin/pyauto-pulse board --offline (board gains one open-task line only)
• bin/pyauto-pulse check --offline: OK

🤖 Generated with Claude Code