Test check_query_denominator; record measured harness coverage (#819)

[joshfactorial]

Steps 1 and 2 of the #819 testing audit. Test and docs only; no change to eidolon or to what the pipeline does on Delta.

Step 1: #[ignore] tests (no change needed)

release-gates runs on every PR into develop. The log of run 37473380770 (#821) shows all 8 ignored tests executing and passing, each in 7–30 s: five realigned SV gates, the transition-matrix round trip, the golden-BAM NM check and the validate differential. Each one fails with a message if its tool is missing; none can skip silently. The workflow header said "six" and named only the gates and the differential. It now lists them without a count.

Step 2: harness coverage of sv_pipeline.sbatch

All 18 suites in scripts/delta/tests/ pass, and all 232 of their mutations are caught locally. CI already runs both modes and fails a suite that reports no mutation summary.

Measured: 25 of 29 functions are exercised, counting functions reached through a tested caller (run_truvari_both → run_truvari → comp_sample, check_denominator). This replaces CLAUDE.md's rule-6 example, which still said 2 of 14 and listed check_denominator as untested.

Untested: score_caller, build_bnd_spans, convert_manta_inversions, index_and_align. These are reported on #819, not fixed here.

The defect fixed

test_scorer_calibration.sh never extracted check_query_denominator. Every run_truvari in the suite therefore called an undefined command, and || QUERY_FILTERED+=(…) flagged each stratum. Output went to /dev/null and nothing asserted QUERY_FILTERED, so the suite reported 132 passed. Shown before the fix: check_query_denominator: command not found, then QUERY_FILTERED=regression regression_allcalls.

Fix:

• The function is extracted.
• The filter stub writes TP-comp, as truvari does. Without it, every case took the "UNVERIFIED" branch.
• New tests:
  • a summary missing TP-comp/FP is reported UNVERIFIED and not flagged;
  • a query with none of its 4 records scored is flagged, counted, and blamed on --passonly (4 non-PASS);
  • a fully scored query is not flagged (must-not-fire);
  • through run_truvari_both, an all-non-PASS query flags exactly the PASS-only pass, and an all-PASS query flags nothing (must-not-fire).
• Mutations, all caught: reading missing keys as dropped; run_truvari ignoring the result; counting no non-PASS records.
• bash -x shows no command not found. 142 passed, 0 mutations survived.

Not verified

• Only the stubs were exercised; no real truvari ran. The stub's TP-comp is derived the way truvari defines it, but it has not been compared against a real truvari summary.

Refs #819.

🤖 Generated with Claude Code