Catchup 19: Fix tooling per solver

[jmsexton03]

Summary

• Catchup context: slice 19 on branch catchup_19_fix_tooling_per_solver.
• Ordered split wave objective: preserve parity/paper cutoff lineage by landing slices in ascending order.
• What was added/changed in this slice:
  • updated derive wave per-solver summary labels from target case path
  • align hierarchical baseline and FAISS scoring expectations with tuned behavior
  • updated revert "fix: align hierarchical baseline candidate pool and FAISS semantic scoring"
  • align hierarchical baseline candidate pool and FAISS semantic scoring
  • add --strategy flag to benchmark runner and update wave script for single-strategy-per-process isolation
  • add wave orchestration script with explicit human tuning checkpoint
  • updated document non-ERF sanity placeholder behavior in runner
  • updated isolate benchmark strategy output roots and nest input writer runs
• Workstreams (topic-level):
  • benchmark workflow/retry behavior updates
  • paper/reproduction workflow scripting
  • node orchestration flow updates
  • service contract/behavior updates
  • unit regression coverage updates
  • retry/timeout handling adjustments
  • FAISS/manifest compatibility handling
  • paper/wave workflow-path handling
  • ERF execution/fallback behavior handling
• Slice metadata:
  • Commit range: 2c2d60017707..1dad54f399ff (source apply_stack_slice_119 -> canonical fix_stack_main)
  • Findings profile (P0/P1/P2/P3): 1/2/0/2 (total 5)
• Fix implementation note: findings are reconciled/resolved on canonical stacked branch fix_stack_main at 73f37cf9e86d.

Related or overlapping functionality / DRY guidance

• Overlap is expected with stacked fix lineage (fix_stack_main); avoid duplicating logic that is already hardened in shared services/nodes.
• Keep node/state contract compatibility aligned with src/models/graph_state_canonical.py and tests/contracts/* when touching shared flows.
• Evidence artifacts for cross-slice decisions: artifacts/integration/findings_reconciliation.json and artifacts/integration/fix_branch_remap_impact.md.
• This embeds a significant architectural decision that needs an ADR.
  • If checked, add an ADR under docs/adr/ (one short file describing context, decision, consequences).

Impact checklist

• fixes a bug or incorrect behavior
• adds new capabilities
• changes answers in the test suite to more than roundoff level
• likely affects downstream users or results
• includes docs updates (code/docs), if appropriate
• none of the above

Tests run (CI runs: pytest tests/unit, pytest tests/quality, pytest tests/integration -m "integration_l1 or integration_l2 or integration_l3 or integration_l4 or integration_full")

• tests/unit: pytest tests/unit
• tests/quality: pytest tests/quality
• integration ladder (CI): pytest tests/integration -m "integration_l1 or integration_l2 or integration_l3 or integration_l4 or integration_full"
• other (list): final closure validation on canonical fix_stack_main
• Output/summary:
  • per-slice branch-head run in this phase: not executed
  • canonical closure branch used for validation: fix_stack_main (73f37cf9e86d)
  • canonical unit: 1663 passed, 31 skipped, 3 warnings (coverage 56.82%)
  • canonical full: 1813 passed, 78 skipped, 10 xfailed, 11 warnings (coverage 58.63%)
  • canonical quality: 20 passed, 1 skipped, 5 warnings
  • canonical integration ladder: 47 passed, 36 skipped, 92 deselected, 1 xfailed, 7 warnings
  • canonical junit evidence: artifacts/integration/reports/fix_stack_main_20260318_034847/unit.junit.xml, artifacts/integration/reports/fix_stack_main_20260318_034847/full.junit.xml
• If tests require repos/schemas/indices or real services, note markers used.
• requires_solver(...) implies repo + schema + default indices are available locally.
• Use -k pelec|erf|amrex|warpx to filter solver-specific tests.

Tests not run in CI (required if any)

• CI runs tests/unit, tests/quality, and tests/integration with integration_l1..l4 + integration_full markers via micromamba; list anything else not covered by CI here.
• None
• tests/e2e
• other (list): per-slice branch-local test reruns
• Reason for skip: this phase validated closure on canonical stacked branch (fix_stack_main) rather than re-running each catchup branch independently.
• Risk/mitigation: parity/open-PR coverage gate rerun is explicitly queued in the handoff for network-enabled execution.

Notes (optional)

• Manual output / logs (short):
  • Validation bundle: artifacts/integration/reports/fix_stack_main_20260318_034847
  • Reconciliation totals: total_findings=109, addressed=109
  • Remap artifact: artifacts/integration/fix_branch_remap_impact.md
• Known limitations:
  • Catchup PRs are split for ordering/parity traceability; final integrated evidence remains anchored on fix_stack_main artifacts.

Labels (optional)

• Not applicable for these ordered catchup PRs; label hygiene is deferred to maintainer-side triage.

[jmsexton03]

note these choices reflect an important shift in src