[Jammy2211]

Summary

This PR writes the 2026-10-15 interferometer CPU profiling (Aris SPT2319-55, 125×125 and 512×512 masks) into the Pulse ledger, following Pulse's own standard. Campaign anchor: PyAutoPulse#62, task interferometer_large_mask_aris.

• Knowledge (2026-10-15) in tasks/interferometer_large_mask_aris.md, written for whoever plans the next interferometer CPU campaign:
  • (a) cost structure after the three fixes. At 125×125: 0.63 s at 1,405 vertices (direct_conv 53%, solve 18%, Cholesky 13%) and 2.0–2.4 s at 2,947 (18/40/31%). At 512×512 the curvature build is 96–98% of the call.
  • (b) the #630 memory fix.
  • (c) the #631 log-det root cause: zero-weight on-vertex mappings, plus the unfiled jitter residual.
  • (d) the #632 cost model.
  • (e) the phase-2 six-cell table from autolens_profiling#420. That PR is at PR-open, not merged. Both of its stated limits are kept.
  • (f) the ranked levers. L1 Cholesky reuse (−20% / −40–45%) is in development as PyAutoArray#633. L2 direct_conv symmetry (2–2.5× at 512, unconfirmed) is filed. The fixed-mapper preload is an existing task. The not-worth-it list keeps its numbers.
  • (g) collaborator runtime settings.
• Two human decision records, indexed in decisions.yaml and rendered in Decision History:
  • decisions/adaptsplit-zero-weight-delaunay-only.md: the human's AskUserQuestion answer "Delaunay only (Recommended by review)", PyAutoArray#627 issuecomment-6200058449.
  • decisions/interferometer-route-gate-cost-model.md: plan approved under --auto; #632 merged unchanged by human /prm.
  • The human explicitly directed both records this session ("Yes, record it as my decision").
• campaigns.yaml: the task and campaign next: lines are refreshed, and the board is re-rendered.

This is a ledger-only change. No library, compute, default or release change. Timings are laptop ratios (load 2–14) and qualify no baseline. last_checkin is not stamped.

Scripts Changed

• tasks/interferometer_large_mask_aris.md
• decisions/adaptsplit-zero-weight-delaunay-only.md (new)
• decisions/interferometer-route-gate-cost-model.md (new)
• decisions.yaml, campaigns.yaml
• dashboard.md, dashboard.html (re-rendered)

Test Plan

• bin/pyauto-pulse board --offline: rendered; dashboard.md lists both decision ids
• bin/pyauto-pulse check --offline: OK
• python3 -m pytest -q tests: 207 passed

Heart RED development override

Heart is RED (organism-scope reasons as listed on PyAutoArray#626, unrelated to this ledger change). The live human authorized a development-only override this session (session 3e3dea38, AskUserQuestion "Yes, override (Recommended)", 2026-10-15 late), and it is recorded in PyAutoMind active.md under pulse-interferometer-knowledge. The override permits commit, push and PR-open. Merging is a separate human /prm. This PR does not claim to fix Heart.

Campaign: PyAutoPulse#62 stays open as the anchor.

🤖 Generated with Claude Code