[Jammy2211]

Closes #497. Likelihood lifecycle phase 3, day-zero write 1/4: the producer side of the 39 designations recorded in PyAutoLabs/PyAutoPulse#84.

Heart RED override: invoked by the human 2026-10-31 for this task (development only). Tier: judge, so merge is by human /prm only.

What changes

• catalogue/pulse_designations.json mirrors PyAutoPulse decisions.yaml kind: designation (39 rows). It is written by build_catalogue.py --sync-designations ../PyAutoPulse. CI's build_catalogue.py --check --validate-with ../PyAutoPulse now also fails if the mirror drifts from Pulse, and it resolves every decision_id through pulse.ingest.unresolved_decisions.
• Classes (catalogue_lifecycle.record_class). Hardware classes map as hpc_ral_cpu = cpu_many_core and hpc_a100 = gpu_a100.
  • The designated optimal setup on its likelihood × standard regime × hardware class → class: optimal + decision_id.
  • A named secondary → class: secondary + decision_id.
  • Both apply to current and superseded rows.
  • Human rulings 2026-10-31:
    • Any other setup on a designated key → experimental.
    • local_* hosts and source_pixels_<n> regimes inherit the role as secondary, with an undesignated_regime_or_host limitation and no decision_id. Optimal is only ever decision-backed.
• catalogue/designation_holds.json keeps the held datacube/delaunay hpc_a100 group provisional (secondary + provisional_optimal: true; 24 records). It also holds the planning metadata kept from the retired file: GPU hosts, regime-only routes, and the R1(a) dense-NUFFT-at-sma rule.
• catalogue/provisional_designations.json is deleted. planned_setups, planned_cells and coverage_matrix.py now derive from the designations plus the holds: 272 → 232 planned cells (148 optimal, 84 secondary), all current.
• Frozen phase-2 campaign tests: each manifest measured every designated cell. Each extra cell is a setup its designation dropped, which is asserted.

Result (dashboard/catalogue.json, 1,548 records)

class	current	superseded
optimal (with decision_id)	283	253
secondary with decision_id	177	191
secondary without (inherited / held)	37	408
experimental	115	84

All 39 decision_ids are cited, every designation has a current optimal record, and all 230 evidence ids are present.

Checks run locally

• build_catalogue.py --check --validate-with ../PyAutoPulse: current. The mirror equals Pulse, and Pulse classify returns ok with every decision_id resolved.
• coverage_matrix.py --check, build_dashboard.py --check, build_readme.py --check, check_wiki.py --check, list_timing_assertions.py --check, wall/check_submits.py --check: OK.
• ruff check and ruff format --check: clean.
• pytest scripts/misc/*/test: no new failures against main. The sandbox lacks jax, scipy and the PyAuto libraries, so the modules needing them were not run here; CI runs them.

Not in this PR (tasks 2–4)

• The day_zero stamp and Pulse snapshot.
• The wiki/campaigns/<dataset_type>/<variant>/day_zero.md rule pages.
• The setup-baseline close.
• The runtime_estimates export.

---

Generated by Claude Code