Skip to content

feat: campaign evidence readers, evidence-graded plan and pilot (hpc-campaign p3) #1745

Closed

Description

@Jammy2211

Overview

Phase 3 of the hpc-campaign epic. Before a campaign is submitted, python -m autofit.campaign plan proposes resources, wall time, ETA and disk from measured evidence (PyAutoPulse/Insight profiling and inference JSON) or a pilot, with an evidence grade on every number and no invented 'optimal'.

Plan

  • Readers for profiling-summary@3, walltime-validation@1, inference-summary@2 by path or URL; stages declare opaque match keys under evidence:
  • Evidence grades on every number: measured-current / provisional / analogue / superseded / estimated / unsupported
  • campaign plan → plan.md + plan.json (capacity, p50/p95/max wall + RSS, proposed CPUs / mem_per_cpu / walltime = p95+25 % per size bin / throttle, ETA, disk, energy placeholder for phase 4)
  • campaign pilot → stratified 1–2 % manifest; never submitted in this task
  • Witness W3: graded 5,000-lens Euclid VIS plan in the human scratch folder
  • Tier: judge — merge mode: human /prm

Heart RED at the door (2026-10-30): release validation FAILED (stage integrate); PyAutoHeart/workflows/Release Integrate failure; PyAutoNerves test_fitsable.py::test__output_to_fits red; Unit-test timing: 1 regression (>3× baseline), 14 slow; autogalaxy red; Release validation NOT release_ready (v2026.10.30.1.dev85101); autolens_profiling red; Worktree drift 1 dirty. Human override for Heart RED (development only) authorized live by the human for hpc-campaign phases 3 and 4 ("Override, proceed"). Merge still needs green CI and a separate /prm; no release.

Detailed implementation plan

Branch survey: ./fit/PyAutoFit main (7d0fe9d) clean. Suggested branch: feature/hpc-campaign-p3-plan-evidence.

Files (all new unless noted): autofit/campaign/evidence.py (loaders + grade logic + matcher), plan.py, pilot.py, __main__.py (add plan/pilot), spec.py (accept evidence: per stage). Tests: fixture JSON for each schema; grade matrix; DR1 repack reproduced from a sacct fixture (12 GB/4 CPU vis_lp, 16 GB vis_pix, 8 GB SED) within tolerance; 'optimal' never emitted without a designation record. Shares only __main__.py with phase 4; p3 merges first.

Original Prompt

Click to expand starting prompt

Issued: 2026-10-30
Epic: hpc-campaign
Phase: 3
Blocked-by: phase 2 (COMPLETE 2026-10-30)
Concurrent-with: hpc-campaign phase 4 (owns campaign/carbon.py, report.py); this task owns campaign/evidence.py, plan.py, pilot.py; shared campaign/main.py — p3 merges first
Consequence: judge
Autonomy: human-required
Witness: W3 — graded 5,000-lens Euclid VIS plan (vis_lp → vis_pix → SED) from live Pulse/Insight JSON, in Workspace/Science/scratch/hpc_campaign_p3_w3/

HPC campaign phase 3 — evidence readers, plan, pilot

Original request (human, 2026-10-30): "do both" (hpc-campaign phases 3 and 4, in parallel), after
"Ctoninue the PyAutoLEns II HPC runs and Results".

Lane prompt: Workspace/Science/scratch/pyautolens_ii_campaign_results_scoping_2026-10-28/PROMPT_hpc_campaign.md §Phase 3.
@PyAutoFit: readers for profiling-summary@3, walltime-validation@1, inference-summary@2 (path/URL; opaque
match keys under evidence:); evidence grades on every number; never "optimal" without a Pulse designation
record; campaign plan (plan.md + plan.json); campaign pilot (stratified 1–2 % manifest, never submitted
here). Done: DR1 repack numbers reproduced from a sacct fixture; W3.

Activity

  1. Jammy2211 commented on Oct 30, 2026

    @Jammy2211
    CollaboratorAuthor

    Library PR(s) open under the human-authorized Heart RED development override (RED reasons quoted in the issue body). Branch gates passed: full PyAutoFit suite (p3 4,120 / p4 4,143), smoke autofit 10/0, independent review FINDINGS all fixed with tests. PRs: #1747 (p3) → #1748 (p4, stacked) + PyAutoNerves#209. Merge needs green CI and /prm.

  2. Jammy2211 commented on Oct 30, 2026

    @Jammy2211
    CollaboratorAuthor

    Shipped

    hpc-campaign phase 3 merged to main (pending release): #1747 (merge 70ab26c) — evidence readers with grades, campaign plan, campaign pilot. All CI legs green; full suite 4,120; smoke autofit green; independent review findings fixed. Witness W3 in the human scratch folder. Merged under the human-authorized Heart RED development override with a separate /prm.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions