check-in 2026-11-06: all campaigns

checkins/2026-11-06.md

@@ -0,0 +1,50 @@
+# Profiling check-in: 2026-11-06 (all campaigns)
+
+Reviewed on 2026-11-06. This is the first all-campaign sweep since 2026-10-17; the 2026-10-23 check-in covered cluster-pointsolver only, and the likelihood-lifecycle row was refreshed on 2026-11-05 (PyAutoPulse#91). Review time and board generation are not measurement times. The check-in authorizes no compute, default change, resubmission, scientific acceptance, release or merge.
+
+## Verified availability
+
+- **GitHub:** autolens_profiling merged 52 PRs since 2026-10-17 (#418–#520). It has no open PRs; the open issues are #517, #445, #166, #128, #103 and #50. autofit_profiling: #7 (CI) only.
+- **Mind:** completion records since 2026-10-17 were read for the point-solver, cluster, interferometer, lifecycle and numba work.
+- **Library state:** PyAutoLens#800/#801/#808, PyAutoFit#1715, PyAutoArray#655/#677. The latest release is 2026.11.5.1, which does **not** contain PyAutoArray#677.
+- **RAL:** reachable (`euclid_jump`). The queue holds 11 of the human's jobs, all likelihood-lifecycle phase 4 (datacube 414148, interferometer 414156_1, multi_dataset 414158, multi_galaxy 414159/414161/414392). No collection or compute was run for this check-in.
+
+## Campaign review
+
+| Campaign | Changes since last review | Current state | Blockers | Next bounded step |
+|---|---|---|---|---|
+| Likelihood lifecycle | Phase 4: GPU legs done; 4e ingest started (autolens_inference#59, branch with 103 rows / 293 walltime records). Cores sweep complete (parallel efficiency 0.68/0.62/0.55/0.36/0.28 at 2–32 cores). Resumed-stage walls are invalid (Nautilus time across a kill and resume reads 3–4× high), so 414362 is rerun fresh as 414392. Wall-estimate bug filed. | active | 6 running legs; parity rulings (human) | Last legs → 4e PR → Pulse `lens-walltime` refresh |
+| Cluster PointSolver | Spurious-core filter shipped (PyAutoLens#800, default 10.0) plus the guide (#801, workspace #600/#601, HowToLens#101) | active | — | Workspace validation check (Mind draft, not started) |
+| Measurement tools | Numba pin done differently than filed (#519/#520: worker route for the DelaunayNN cells only); phase 6 overlaps `gradient_cost_probe` | active | Re-measure needs a release carrying #677 | Human: supersede `gradient_cost_probe`? |
+| DelaunayNN CPU | #677 serial Sibson in workers; #519/#520 worker-route cells; serial kernel ≈ 1.25× prange@1 (Mind bug draft) | parked | — | Unchanged (parked) |
+| Interferometer | Aris SIGSEGV root cause plus fixes shipped 2026-10-24 (PyAutoFit#1715 warm-up, PyAutoArray#655 finufft) | parked | — | Human: close `interferometer_large_mask_aris`? |
+| Point-source image · CPU | No measurement; adjacent PyAutoLens#800/#801 | active | — | Epic closure is a human decision |
+| Point-source image · A100 | Unchanged | parked | — | — |
+| Imaging | Lifecycle 2a standard-size rows plus the day_zero stamp are now the baseline; over-sampling phase 2 still held | active | Human hold | — |
+| Linear solver | Unchanged; the project wiki still says the decision is pending (stale) | active | — | Wiki correction when next touched |
+| PyAutoFit bootstrap | CI token wiring only (autofit_profiling#7) | active | Human scope for epic 2 | — |
+| Critical curves | No measurement; adjacent PyAutoLens#808 | needs-decision | Human scoping | — |
+| Certified solver | Unchanged | parked | — | — |
+| Setup baseline, point-source source plane | Closed earlier (2026-11-01, 2026-10-17) | complete | — | — |
+
+## Task changes
+
+| Task | State | Note |
+|---|---|---|
+| numba_num_threads_pin_and_delaunay_nn_remeasure | ready → **blocked** | Step 1 done via #519/#520; step 2 waits for a release carrying PyAutoArray#677 |
+| interferometer_large_mask_aris | active → **needs-decision** | Root cause fixed and shipped; whether to close is the human's call |
+| gradient_cost_probe | ready → **needs-decision** | Phase 6 covered its scope; supersede, or keep for the PyAutoLens#768 test-timing question |
+| cluster_pointsolver_speed | active | Spurious-core and guide shipped; validation check next |
+| delaunay_nn_cpu_imaging | parked | #677 / #519 / #520 and the serial-kernel bug draft recorded |
+| All other open tasks | unchanged | Reviewed; no PR, issue, Mind record or measurement since the last review |
+
+## Unavailable or unverified
+
+- Phase-4 legs still running were not collected. The 4e branch holds only rows from COMPLETED legs.
+- No profiling measurement was run, and no project result was re-validated. Statements about the phase-6 overlap and the Aris closure are candidates for human decision, not verdicts.
+
+## Human decisions (same day)
+
+- `interferometer_large_mask_aris`: closed (complete), "1. yes".
+- `gradient_cost_probe`: superseded by likelihood-lifecycle phase 6, "2. yes".
+- Phase-4 parity rulings: wait until the 4e PR, "3. ok wait until 4e".