[Jammy2211]

Summary

Profiling side of PyAutoArray#505 — the numba CPU-path speed-up of the curvature
matrix F at HST resolution. Adds per-sub-block timing of F to the numba
likelihood-breakdown harnesses, re-baselines the breakdown artifacts (they were
stale since #462), and records the full before/after method and results.

The harness split F into mapper x mapper, mapper x linear-func and
linear-func x linear-func steps, which is what identified the dense
sliding-window correlation in the mapper x linear-func block as 72 % of F and
55 % of the whole HST evaluation — the finding the library change acts on.

Headline (s per evaluation, OMP_NUM_THREADS=1, memo off, n_repeats 10; the
before column is a fresh re-measurement taken back-to-back with the after run,
since this host carries 20-30 % session-to-session variance):

cell	eval	F total	mapper x mapper	mapper x l-func
hst, rectangular bilinear	1.562 -> 0.595	1.195 -> 0.359	0.295 -> 0.277	0.858 -> 0.054
euclid, rectangular bilinear	0.349 -> 0.249	0.256 -> 0.097	0.060 -> 0.065	0.182 -> 0.023
hst, Delaunay-1250	1.367 -> 0.758	1.077 -> 0.184	0.122 -> 0.123	0.854 -> 0.052
hst, rectangular RTU	(8.501) -> 7.479	(1.250) -> 0.334	(0.304) -> 0.249	(0.929) -> 0.061

The notes also record a Nautilus pool run: the same workspace fit at
number_of_cores=8 against canonical PyAutoArray main and against the feature
branch, plus a serial control, confirming the single-thread win survives
multiprocessing (pool +9 %, serial +7 %, parallel speed-up ratio flat at
2.52x -> 2.57x on 8 cores — no FFT/BLAS thread oversubscription).

Only this task's claimed file sets are touched (scripts/imaging/likelihood_breakdown/,
results/breakdown/imaging/, results/notes/), committed with explicit
pathspecs per the parallel claim with nuts-warm-start-driver-and-a100-probe.

Upstream PR

PyAutoLabs/PyAutoArray#506 — PyAutoLabs/PyAutoArray#506

Library-first merge gate: this PR must NOT merge until PyAutoArray#506 is
MERGED. The re-baselined artifacts and the "after" columns were measured
against that branch.

Scripts Changed

• scripts/imaging/likelihood_breakdown/pixelization_numba.py — split the F
  step into mapper x mapper, mapper x linear-func and
  linear-func x linear-func sub-steps. The mapper x linear-func label was
  [dense conv] and is now [FFT conv + scatter], since that is what it now
  times. Step list, residual design and pins otherwise unchanged.
• scripts/imaging/likelihood_breakdown/delaunay_numba.py — same sub-block
  split for the Delaunay cell.

Results regenerated

• results/breakdown/imaging/pixelization_numba_breakdown_hst_v2026.8.17.1.{json,png}
• results/breakdown/imaging/pixelization_numba_breakdown_hst_v2026.8.17.1_rtu.{json,png}
• results/breakdown/imaging/pixelization_numba_breakdown_euclid_v2026.8.17.1.{json,png}
• results/breakdown/imaging/delaunay_numba_breakdown_hst_v2026.8.17.1.{json,png}
• results/notes/numba_curvature_matrix_f_split.md — method, the step-0
  baseline, the paired before/after table, the "where the time went" analysis,
  the correctness evidence, and the pool-run section.

Test Plan

• All 3 pinned log-likelihoods PASSED on the re-baselined harness: hst
  bilinear 27661.91013366411 (pin 27661.910133665442), hst RTU
  27180.70471569685 (pin 27180.704715698186), hst Delaunay
  29090.527210448134 (pin 29090.527192092646). euclid has no pin; it
  measured 6213.306873885871, unchanged to every recorded digit.
• F agrees with the pre-change sliding-window result to 3e-18 relative
  on euclid and hst.
• Pool run reproduced twice per configuration with the output directory
  wiped between runs, autoarray.__file__ asserted on both sides.
• Reviewer: the "after" numbers only reproduce with PyAutoArray#506 checked
  out; against main the harness will report the "before" column.

Heart readiness — human RED override

pyauto-heart readiness graded RED score 45 at ship time (ts
2026-08-28T15:02:11Z). The diff repairs none of the named reasons, so this is a
plain human override, not the corrective-PR exception. Verbatim:

red_reasons:
  "release validation FAILED (stage integrate)"
yellow_reasons:
  "workspace validation not passing (2 failed, cloud#33179766004: autolens_test scripts/imaging/rectangular_mge.py, autolens_test scripts/imaging/rectangular_mge_rtu.py)"
  "manifest drift: session-start hooks (generated) — 32 mismatch(es) vs PyAutoMind/repos.yaml"

Human authorisation, 2026-08-28, verbatim:

I authorize things to override the heart RED.

Scope: push + PR-open for this task only. Merge and release stay human, and
the override does not extend to any other task. The RED is pre-existing on
main and unrelated — every failing script is on the JAX likelihood path or in
autofit, in other repos.

Generated by the PyAutoLabs agent workflow.