devlyn-cli 2.3.0 → 2.3.2

package/benchmark/auto-resolve/scripts/test-audit-pair-evidence.sh

@@ -0,0 +1,1672 @@
+#!/usr/bin/env bash
+# Regression tests for audit-pair-evidence.py.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+SCRIPT="$SCRIPT_DIR/audit-pair-evidence.py"
+TMP_DIR="$(mktemp -d /tmp/audit-pair-evidence-test.XXXXXX)"
+trap 'rm -rf "$TMP_DIR"' EXIT
+
+expect_fail_contains() {
+  local label="$1"
+  local needle="$2"
+  shift 2
+  local out="$TMP_DIR/$label.out"
+  if "$@" > "$out" 2>&1; then
+    echo "expected failure for $label" >&2
+    cat "$out" >&2
+    exit 1
+  fi
+  if ! grep -Fq -- "$needle" "$out"; then
+    echo "missing expected text for $label: $needle" >&2
+    cat "$out" >&2
+    exit 1
+  fi
+}
+
+fixtures="$TMP_DIR/fixtures"
+results="$TMP_DIR/results"
+registry="$TMP_DIR/pair-rejected-fixtures.sh"
+mkdir -p "$fixtures/F16-cli-quote-tax-rules" \
+  "$fixtures/F21-cli-scheduler-priority" \
+  "$fixtures/F34-cli-rejected-candidate" \
+  "$results/pair-pass" \
+  "$results/pair-pass-2" \
+  "$results/rejected-headroom"
+
+cat > "$registry" <<'SH'
+rejected_pair_fixture_reason() {
+  local fid="$1"
+  case "$fid" in
+    F34-*|F34)
+      echo "measured solo ceiling"
+      ;;
+    *)
+      return 1
+      ;;
+  esac
+}
+SH
+
+cat > "$results/pair-pass/full-pipeline-pair-gate.json" <<'JSON'
+{
+  "run_id": "pair-pass",
+  "verdict": "PASS",
+  "pair_arm": "l2_risk_probes",
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "PASS",
+      "bare_score": 50,
+      "solo_score": 75,
+      "pair_score": 96,
+      "pair_margin": 21,
+      "pair_mode": true,
+      "pair_trigger_eligible": true,
+      "pair_trigger_reasons": ["complexity.high"],
+      "pair_trigger_has_canonical_reason": true,
+      "pair_solo_wall_ratio": 1.28
+    }
+  ]
+}
+JSON
+
+cat > "$results/rejected-headroom/headroom-gate.json" <<'JSON'
+{
+  "run_id": "rejected-headroom",
+  "verdict": "FAIL",
+  "rows": [
+    {
+      "fixture": "F34-cli-rejected-candidate",
+      "status": "FAIL",
+      "bare_score": 33,
+      "solo_score": 98,
+      "reason": "solo_claude score 98 > 80"
+    }
+  ]
+}
+JSON
+
+expect_fail_contains unmeasured "unmeasured candidate fixture(s): F21-cli-scheduler-priority" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --out-dir "$TMP_DIR/out-fail"
+
+expect_fail_contains bad-min-pair-evidence "--min-pair-evidence must be >= 1" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --min-pair-evidence 0
+
+expect_fail_contains bad-min-pair-margin "--min-pair-margin must be >= 1" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --min-pair-margin 0
+
+expect_fail_contains bad-max-wall-ratio "--max-pair-solo-wall-ratio must be finite and > 0" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --max-pair-solo-wall-ratio 0
+
+expect_fail_contains nan-max-wall-ratio "--max-pair-solo-wall-ratio must be finite and > 0" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --max-pair-solo-wall-ratio NaN
+
+cat > "$results/pair-pass-2/full-pipeline-pair-gate.json" <<'JSON'
+{
+  "run_id": "pair-pass-2",
+  "verdict": "PASS",
+  "pair_arm": "l2_risk_probes",
+  "rows": [
+    {
+      "fixture": "F21-cli-scheduler-priority",
+      "status": "PASS",
+      "bare_score": 33,
+      "solo_score": 66,
+      "pair_score": 99,
+      "pair_margin": 33,
+      "pair_mode": true,
+      "pair_trigger_eligible": true,
+      "pair_trigger_reasons": ["complexity.high", "risk_profile.high_risk"],
+      "pair_trigger_has_canonical_reason": true,
+      "pair_solo_wall_ratio": 1.47
+    }
+  ]
+}
+JSON
+
+expect_fail_contains pair-evidence-hypotheses "pair evidence hypotheses missing for fixture(s): F16-cli-quote-tax-rules, F21-cli-scheduler-priority" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --min-pair-evidence 2 \
+    --out-dir "$TMP_DIR/out-hypothesis-fail"
+
+for fixture in F16-cli-quote-tax-rules F21-cli-scheduler-priority; do
+  cat > "$fixtures/$fixture/spec.md" <<'EOF'
+# Spec
+
+## Verification
+
+- Visible pair-evidence fixture.
+
+## Solo-headroom hypothesis
+
+A capable solo_claude baseline is expected to miss the ordering interaction;
+observable command `node "$BENCH_FIXTURE_DIR/verifiers/visible.js"` exposes the miss.
+EOF
+  cat > "$fixtures/$fixture/NOTES.md" <<'EOF'
+# Notes
+EOF
+  cat > "$fixtures/$fixture/expected.json" <<'EOF'
+{
+  "verification_commands": [
+    {
+      "cmd": "node \"$BENCH_FIXTURE_DIR/verifiers/visible.js\"",
+      "exit_code": 0
+    }
+  ]
+}
+EOF
+done
+
+python3 "$SCRIPT" \
+  --fixtures-root "$fixtures" \
+  --registry "$registry" \
+  --results-root "$results" \
+  --min-pair-evidence 2 \
+  --out-dir "$TMP_DIR/out-pass" \
+  > "$TMP_DIR/pass.out"
+grep -Fq 'PASS audit-pair-evidence' "$TMP_DIR/pass.out"
+test -f "$TMP_DIR/out-pass/frontier.json"
+test -f "$TMP_DIR/out-pass/frontier.stdout"
+test -f "$TMP_DIR/out-pass/frontier.stderr"
+test -f "$TMP_DIR/out-pass/headroom-audit.json"
+test -f "$TMP_DIR/out-pass/headroom-rejections.stdout"
+test -f "$TMP_DIR/out-pass/headroom-rejections.stderr"
+test -f "$TMP_DIR/out-pass/audit.json"
+grep -Fq 'F16-cli-quote-tax-rules: bare=50 solo_claude=75 pair=96 arm=l2_risk_probes margin=+21' "$TMP_DIR/out-pass/frontier.stdout"
+grep -Fq 'pair_margin_avg=+27.00 pair_margin_min=+21 wall_avg=1.38x wall_max=1.47x' "$TMP_DIR/out-pass/frontier.stdout"
+grep -Fq 'verdict=pair_evidence_passed' "$TMP_DIR/out-pass/frontier.stdout"
+grep -Fq 'PASS pair-candidate-frontier' "$TMP_DIR/out-pass/frontier.stdout"
+grep -Fq 'headroom_rejections=PASS verdict=PASS unrecorded=0 unsupported=0' "$TMP_DIR/pass.out"
+grep -Fq 'pair_evidence_quality=PASS min_pair_margin_actual=+21 min_pair_margin_required=+5 max_wall_actual=1.47x max_wall_allowed=3.00x' "$TMP_DIR/pass.out"
+grep -Fq 'pair_trigger_reasons=PASS canonical=2 historical_alias=1 exposed=2 total=2 summary=2 rows_match=true' "$TMP_DIR/pass.out"
+grep -Fq 'pair_trigger_historical_aliases=F21-cli-scheduler-priority=risk_profile.high_risk' "$TMP_DIR/pass.out"
+grep -Fq 'pair_evidence_hypotheses=PASS documented=2 total=2' "$TMP_DIR/pass.out"
+grep -Fq 'pair_evidence_hypothesis_triggers=WARN matched=0 documented=2 total=2' "$TMP_DIR/pass.out"
+grep -Fq 'pair_evidence_hypothesis_trigger_gaps=F16-cli-quote-tax-rules=complexity.high;F21-cli-scheduler-priority=complexity.high,risk_profile.high_risk' "$TMP_DIR/pass.out"
+python3 - "$TMP_DIR/out-pass/audit.json" <<'PY'
+import json
+import sys
+
+report = json.load(open(sys.argv[1], encoding="utf8"))
+assert report["verdict"] == "PASS"
+assert report["min_pair_evidence"] == 2
+assert report["min_pair_margin"] == 5
+assert report["max_pair_solo_wall_ratio"] == 3.0
+assert report["frontier_summary"]["min_pair_margin"] == 5
+assert report["frontier_summary"]["max_pair_solo_wall_ratio"] == 3.0
+assert report["frontier_summary"]["fixtures_total"] == 3
+assert report["frontier_summary"]["candidate_count"] == 2
+assert report["frontier_summary"]["pair_evidence_count"] == 2
+assert report["frontier_summary"]["unmeasured_count"] == 0
+assert report["frontier_summary"]["pair_margin_avg"] == 27
+assert report["frontier_summary"]["pair_margin_min"] == 21
+assert report["frontier_summary"]["pair_solo_wall_ratio_avg"] == 1.38
+assert report["frontier_summary"]["pair_solo_wall_ratio_max"] == 1.47
+assert report["pair_evidence_rows"] == [
+    {
+        "fixture": "F16-cli-quote-tax-rules",
+        "verdict": "pair_evidence_passed",
+        "run_id": "pair-pass",
+        "pair_arm": "l2_risk_probes",
+        "bare_score": 50,
+        "solo_score": 75,
+        "pair_score": 96,
+        "pair_margin": 21,
+        "pair_mode": True,
+        "pair_trigger_eligible": True,
+        "pair_trigger_reasons": ["complexity.high"],
+        "pair_trigger_has_canonical_reason": True,
+        "pair_trigger_has_hypothesis_reason": False,
+        "pair_solo_wall_ratio": 1.28,
+    },
+    {
+        "fixture": "F21-cli-scheduler-priority",
+        "verdict": "pair_evidence_passed",
+        "run_id": "pair-pass-2",
+        "pair_arm": "l2_risk_probes",
+        "bare_score": 33,
+        "solo_score": 66,
+        "pair_score": 99,
+        "pair_margin": 33,
+        "pair_mode": True,
+        "pair_trigger_eligible": True,
+        "pair_trigger_reasons": ["complexity.high", "risk_profile.high_risk"],
+        "pair_trigger_has_canonical_reason": True,
+        "pair_trigger_has_hypothesis_reason": False,
+        "pair_solo_wall_ratio": 1.47,
+    },
+]
+assert report["checks"]["frontier"]["status"] == "PASS"
+assert report["checks"]["headroom_rejections"]["status"] == "PASS"
+assert report["checks"]["headroom_rejections"]["exit_code"] == 0
+assert report["checks"]["headroom_rejections"]["report_check_exit_code"] == 0
+assert report["checks"]["headroom_rejections"]["verdict"] == "PASS"
+assert report["checks"]["headroom_rejections"]["unrecorded_failure_count"] == 0
+assert report["checks"]["headroom_rejections"]["unsupported_registry_rejection_count"] == 0
+assert report["checks"]["frontier_report"]["status"] == "PASS"
+assert report["checks"]["frontier_report"]["verdict"] == "PASS"
+assert report["checks"]["frontier_report"]["unmeasured_count"] == 0
+assert report["checks"]["frontier_stdout"]["status"] == "PASS"
+assert report["checks"]["frontier_stdout"]["summary_rows"] == 1
+assert report["checks"]["frontier_stdout"]["aggregate_rows"] == 1
+assert report["checks"]["frontier_stdout"]["final_verdict_rows"] == 1
+assert report["checks"]["frontier_stdout"]["expected_rows"] == 2
+assert report["checks"]["frontier_stdout"]["stdout_rows"] == 2
+assert report["checks"]["frontier_stdout"]["trigger_rows"] == 2
+assert report["checks"]["frontier_stdout"]["hypothesis_trigger_rows"] == 2
+assert report["checks"]["frontier_stdout"]["rows_match_count"] is True
+assert report["checks"]["frontier_stdout"]["trigger_rows_match_count"] is True
+assert report["checks"]["frontier_stdout"]["hypothesis_trigger_rows_match_count"] is True
+assert report["checks"]["min_pair_evidence"]["status"] == "PASS"
+assert report["checks"]["min_pair_evidence"]["required"] == 2
+assert report["checks"]["min_pair_evidence"]["actual"] == 2
+assert report["checks"]["min_pair_evidence"]["actual_rows"] == 2
+assert report["checks"]["min_pair_evidence"]["rows_match_count"] is True
+assert report["checks"]["pair_evidence_quality"]["status"] == "PASS"
+assert report["checks"]["pair_evidence_quality"]["min_pair_margin_required"] == 5
+assert report["checks"]["pair_evidence_quality"]["min_pair_margin_actual"] == 21
+assert report["checks"]["pair_evidence_quality"]["max_pair_solo_wall_ratio_allowed"] == 3.0
+assert report["checks"]["pair_evidence_quality"]["max_pair_solo_wall_ratio_actual"] == 1.47
+assert report["checks"]["pair_evidence_quality"]["summary_min_pair_margin"] == 21
+assert report["checks"]["pair_evidence_quality"]["summary_max_pair_solo_wall_ratio"] == 1.47
+assert report["checks"]["pair_trigger_reasons"]["status"] == "PASS"
+assert report["checks"]["pair_trigger_reasons"]["summary_pair_evidence_count"] == 2
+assert report["checks"]["pair_trigger_reasons"]["canonical_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["historical_alias_rows"] == 1
+assert report["checks"]["pair_trigger_reasons"]["historical_alias_details"] == [
+    {"fixture": "F21-cli-scheduler-priority", "aliases": ["risk_profile.high_risk"]}
+]
+assert report["checks"]["pair_trigger_reasons"]["exposed_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["total_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["rows_match_count"] is True
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["status"] == "WARN"
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["exit_code"] == 0
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["required"] is False
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["matched_rows"] == 0
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["documented_rows"] == 2
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["total_rows"] == 2
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["gap_details"] == [
+    {
+        "fixture": "F16-cli-quote-tax-rules",
+        "pair_trigger_reasons": ["complexity.high"],
+    },
+    {
+        "fixture": "F21-cli-scheduler-priority",
+        "pair_trigger_reasons": ["complexity.high", "risk_profile.high_risk"],
+    },
+]
+assert report["artifacts"] == {
+    "frontier_json": "frontier.json",
+    "frontier_stdout": "frontier.stdout",
+    "frontier_stderr": "frontier.stderr",
+    "headroom_audit_json": "headroom-audit.json",
+    "headroom_rejections_stdout": "headroom-rejections.stdout",
+    "headroom_rejections_stderr": "headroom-rejections.stderr",
+    "audit_json": "audit.json",
+}
+PY
+
+if python3 "$SCRIPT" \
+  --fixtures-root "$fixtures" \
+  --registry "$registry" \
+  --results-root "$results" \
+  --min-pair-evidence 2 \
+  --require-hypothesis-trigger \
+  --out-dir "$TMP_DIR/out-strict-trigger" \
+  > "$TMP_DIR/strict-trigger.out" 2>&1; then
+  echo "audit must fail when --require-hypothesis-trigger sees trigger gaps" >&2
+  cat "$TMP_DIR/strict-trigger.out" >&2
+  exit 1
+fi
+grep -Fq 'pair evidence hypothesis triggers missing for fixture(s): F16-cli-quote-tax-rules, F21-cli-scheduler-priority' "$TMP_DIR/strict-trigger.out"
+grep -Fq 'pair_evidence_hypothesis_triggers=FAIL matched=0 documented=2 total=2' "$TMP_DIR/strict-trigger.out"
+grep -Fq 'pair_evidence_hypothesis_trigger_gaps=F16-cli-quote-tax-rules=complexity.high;F21-cli-scheduler-priority=complexity.high,risk_profile.high_risk' "$TMP_DIR/strict-trigger.out"
+grep -Fq 'FAIL audit-pair-evidence' "$TMP_DIR/strict-trigger.out"
+python3 - "$TMP_DIR/out-strict-trigger/audit.json" <<'PY'
+import json
+import sys
+
+report = json.load(open(sys.argv[1], encoding="utf8"))
+assert report["verdict"] == "FAIL"
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["status"] == "FAIL"
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["exit_code"] == 1
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["required"] is True
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["gap_details"] == [
+    {
+        "fixture": "F16-cli-quote-tax-rules",
+        "pair_trigger_reasons": ["complexity.high"],
+    },
+    {
+        "fixture": "F21-cli-scheduler-priority",
+        "pair_trigger_reasons": ["complexity.high", "risk_profile.high_risk"],
+    },
+]
+PY
+
+python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/out-pass/frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/missing-trigger-reasons.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains missing-trigger-reasons "pair trigger reason rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/missing-trigger-reasons.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_trigger_reasons(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/malformed-trigger-reason-rows.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": [],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains malformed-trigger-reason-rows "pair trigger reason rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/malformed-trigger-reason-rows.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_trigger_reasons(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/mixed-unknown-trigger-reason-rows.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high", "looks-hard"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains mixed-unknown-trigger-reason-rows "pair trigger reason rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/mixed-unknown-trigger-reason-rows.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_trigger_reasons(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/normalized-canonical-trigger-reason-rows.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["risk high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains normalized-canonical-trigger-reason-rows "pair trigger reason rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/normalized-canonical-trigger-reason-rows.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_trigger_reasons(pathlib.Path(sys.argv[2])))
+PY
+
+grep -Fv 'PASS pair-candidate-frontier' "$TMP_DIR/out-pass/frontier.stdout" \
+  > "$TMP_DIR/missing-final-verdict-frontier.stdout"
+expect_fail_contains missing-final-frontier-verdict "frontier stdout final verdict row count is not exactly 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/missing-final-verdict-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cp "$TMP_DIR/out-pass/frontier.stdout" "$TMP_DIR/duplicate-final-verdict-frontier.stdout"
+printf 'PASS pair-candidate-frontier\n' >> "$TMP_DIR/duplicate-final-verdict-frontier.stdout"
+expect_fail_contains duplicate-final-frontier-verdict "frontier stdout final verdict row count is not exactly 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/duplicate-final-verdict-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+printf 'fixtures=3 rejected=1 candidates=2 pair_evidence=2 unmeasured=0 verdict=PASS\n' \
+  > "$TMP_DIR/bad-frontier.stdout"
+printf 'pair_margin_avg=+27.00 pair_margin_min=+21 wall_avg=1.38x wall_max=1.47x\n' \
+  >> "$TMP_DIR/bad-frontier.stdout"
+printf 'F16-cli-quote-tax-rules: bare=50 solo_claude=75 pair=95 arm=l2_risk_probes margin=+20 wall=1.28x run=pair-pass verdict=pair_evidence_passed\n' \
+  >> "$TMP_DIR/bad-frontier.stdout"
+printf 'F21-cli-scheduler-priority: bare=33 solo_claude=66 pair=99 arm=l2_risk_probes margin=+33 wall=1.47x run=pair-pass-2 verdict=pair_evidence_passed\n' \
+  >> "$TMP_DIR/bad-frontier.stdout"
+expect_fail_contains missing-frontier-score-row "frontier stdout missing score row for F16-cli-quote-tax-rules" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/bad-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+sed -E 's/ triggers=[^[:space:]]+//' "$TMP_DIR/out-pass/frontier.stdout" \
+  > "$TMP_DIR/no-trigger-frontier.stdout"
+expect_fail_contains missing-frontier-triggers "frontier stdout missing score row for F16-cli-quote-tax-rules" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/no-trigger-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/no-aggregate-frontier.stdout" <<'OUT'
+fixtures=3 rejected=1 candidates=2 pair_evidence=2 unmeasured=0 verdict=PASS
+F16-cli-quote-tax-rules: bare=50 solo_claude=75 pair=96 arm=l2_risk_probes margin=+21 wall=1.28x run=pair-pass verdict=pair_evidence_passed
+F21-cli-scheduler-priority: bare=33 solo_claude=66 pair=99 arm=l2_risk_probes margin=+33 wall=1.47x run=pair-pass-2 verdict=pair_evidence_passed
+OUT
+expect_fail_contains missing-frontier-aggregate-row "frontier stdout aggregate score row count is not exactly 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/no-aggregate-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cp "$TMP_DIR/out-pass/frontier.stdout" "$TMP_DIR/duplicate-summary-frontier.stdout"
+printf 'fixtures=3 rejected=1 candidates=2 pair_evidence=2 unmeasured=0 verdict=PASS\n' \
+  >> "$TMP_DIR/duplicate-summary-frontier.stdout"
+expect_fail_contains duplicate-frontier-summary-row "frontier stdout summary score row count is not exactly 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/duplicate-summary-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cp "$TMP_DIR/out-pass/frontier.stdout" "$TMP_DIR/duplicate-aggregate-frontier.stdout"
+printf 'pair_margin_avg=+27.00 pair_margin_min=+21 wall_avg=1.38x wall_max=1.47x\n' \
+  >> "$TMP_DIR/duplicate-aggregate-frontier.stdout"
+expect_fail_contains duplicate-frontier-aggregate-row "frontier stdout aggregate score row count is not exactly 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/duplicate-aggregate-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/partial-frontier.stdout" <<'OUT'
+fixtures=3 rejected=1 candidates=2 pair_evidence=2 unmeasured=0 verdict=PASS
+pair_margin_avg=+27.00 pair_margin_min=+21 wall_avg=1.38x wall_max=1.47x
+F16-cli-quote-tax-rules: bare=50 solo_claude=75 pair=96 arm=l2_risk_probes margin=+21 verdict=pair_evidence_passed
+F21-cli-scheduler-priority: bare=33 solo_claude=66 pair=99 arm=l2_risk_probes margin=+33 wall=1.47x run=pair-pass-2 verdict=pair_evidence_passed
+OUT
+expect_fail_contains partial-frontier-score-row "frontier stdout missing score row for F16-cli-quote-tax-rules" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/partial-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cp "$TMP_DIR/out-pass/frontier.stdout" "$TMP_DIR/extra-frontier.stdout"
+printf 'F99-stale-fixture: bare=1 solo_claude=2 pair=3 arm=l2_risk_probes margin=+1 wall=1.00x run=stale verdict=pair_evidence_passed\n' \
+  >> "$TMP_DIR/extra-frontier.stdout"
+expect_fail_contains extra-frontier-score-row "frontier stdout score row count 3 does not match frontier evidence row count 2" \
+  python3 - "$SCRIPT" "$TMP_DIR/out-pass/frontier.json" "$TMP_DIR/extra-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/malformed-frontier-summary.json" <<'JSON'
+{
+  "verdict": "PASS"
+}
+JSON
+expect_fail_contains malformed-frontier-stdout-summary "frontier stdout check missing summary fields" \
+  python3 - "$SCRIPT" "$TMP_DIR/malformed-frontier-summary.json" "$TMP_DIR/bad-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/malformed-frontier-count.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "fixtures_total": 3,
+  "rejected_count": 1,
+  "candidate_count": 2,
+  "pair_evidence_count": "2",
+  "unmeasured_count": 0
+}
+JSON
+expect_fail_contains malformed-frontier-stdout-counts "frontier stdout summary counts malformed" \
+  python3 - "$SCRIPT" "$TMP_DIR/malformed-frontier-count.json" "$TMP_DIR/bad-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/malformed-frontier-aggregate.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "fixtures_total": 3,
+  "rejected_count": 1,
+  "candidate_count": 2,
+  "pair_evidence_count": 2,
+  "unmeasured_count": 0,
+  "pair_margin_avg": "27",
+  "pair_margin_min": 21,
+  "pair_solo_wall_ratio_avg": 1.38,
+  "pair_solo_wall_ratio_max": 1.47,
+  "rows": []
+}
+JSON
+expect_fail_contains malformed-frontier-stdout-aggregate "frontier stdout aggregate fields malformed" \
+  python3 - "$SCRIPT" "$TMP_DIR/malformed-frontier-aggregate.json" "$TMP_DIR/bad-frontier.stdout" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_stdout(pathlib.Path(sys.argv[2]), pathlib.Path(sys.argv[3])))
+PY
+
+cat > "$TMP_DIR/frontier-fail-verdict.json" <<'JSON'
+{
+  "verdict": "FAIL",
+  "unmeasured_count": 1,
+  "pair_evidence_count": 1,
+  "rows": []
+}
+JSON
+expect_fail_contains frontier-fail-verdict "frontier verdict 'FAIL' is not PASS" \
+  python3 - "$SCRIPT" "$TMP_DIR/frontier-fail-verdict.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_report(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/frontier-unmeasured.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "unmeasured_count": 1,
+  "pair_evidence_count": 1,
+  "rows": []
+}
+JSON
+expect_fail_contains frontier-unmeasured "frontier has 1 unmeasured candidate fixture(s)" \
+  python3 - "$SCRIPT" "$TMP_DIR/frontier-unmeasured.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_report(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/frontier-malformed-unmeasured.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "unmeasured_count": true,
+  "pair_evidence_count": 1,
+  "rows": []
+}
+JSON
+expect_fail_contains frontier-malformed-unmeasured "frontier unmeasured count missing or malformed" \
+  python3 - "$SCRIPT" "$TMP_DIR/frontier-malformed-unmeasured.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_frontier_report(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/headroom-fail-verdict.json" <<'JSON'
+{
+  "verdict": "FAIL",
+  "unrecorded_failures": [],
+  "unsupported_registry_rejections": []
+}
+JSON
+expect_fail_contains headroom-fail-verdict "headroom audit verdict 'FAIL' is not PASS" \
+  python3 - "$SCRIPT" "$TMP_DIR/headroom-fail-verdict.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_headroom_audit_report(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/headroom-missing-unsupported.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "unrecorded_failures": []
+}
+JSON
+expect_fail_contains headroom-missing-unsupported "headroom audit unsupported registry rejection count missing or malformed" \
+  python3 - "$SCRIPT" "$TMP_DIR/headroom-missing-unsupported.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_headroom_audit_report(pathlib.Path(sys.argv[2])))
+PY
+
+cat > "$TMP_DIR/headroom-unsupported.json" <<'JSON'
+{
+  "verdict": "PASS",
+  "unrecorded_failures": [],
+  "unsupported_registry_rejections": [{"fixture": "F36-unsupported-rejection"}]
+}
+JSON
+expect_fail_contains headroom-unsupported "headroom audit has 1 unsupported registry rejection(s)" \
+  python3 - "$SCRIPT" "$TMP_DIR/headroom-unsupported.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_headroom_audit_report(pathlib.Path(sys.argv[2])))
+PY
+
+python3 - "$SCRIPT" "$TMP_DIR/headroom-unsupported.json" > "$TMP_DIR/headroom-summary.out" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+module.print_headroom_rejections_summary(pathlib.Path(sys.argv[2]), status=1)
+PY
+grep -Fq 'headroom_rejections=FAIL verdict=PASS unrecorded=0 unsupported=1' "$TMP_DIR/headroom-summary.out"
+
+cat > "$TMP_DIR/frontier-incomplete-best.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "higher-incomplete",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 98,
+          "pair_margin": 23,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.32
+        },
+        {
+          "run_id": "lower-complete",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+python3 - "$SCRIPT" "$TMP_DIR/frontier-incomplete-best.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+rows = module.load_pair_evidence_rows(pathlib.Path(sys.argv[2]))
+assert rows == [
+    {
+        "fixture": "F16-cli-quote-tax-rules",
+        "verdict": "pair_evidence_passed",
+        "run_id": "lower-complete",
+        "pair_arm": "l2_risk_probes",
+        "bare_score": 50,
+        "solo_score": 75,
+        "pair_score": 96,
+        "pair_margin": 21,
+        "pair_mode": True,
+        "pair_trigger_eligible": True,
+        "pair_trigger_reasons": ["complexity.high"],
+        "pair_trigger_has_canonical_reason": True,
+        "pair_trigger_has_hypothesis_reason": False,
+        "pair_solo_wall_ratio": 1.28,
+    }
+]
+PY
+
+cat > "$TMP_DIR/bad-frontier-rows.json" <<'JSON'
+{
+  "pair_evidence_count": 2,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": []
+    },
+    {
+      "fixture": "F21-cli-scheduler-priority",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": "malformed"
+    }
+  ]
+}
+JSON
+expect_fail_contains missing-pair-evidence-rows "pair evidence rows 0 do not match summary count 2" \
+  python3 - "$SCRIPT" "$TMP_DIR/bad-frontier-rows.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 2))
+PY
+
+cat > "$TMP_DIR/bad-frontier-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 2,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": null,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    },
+    {
+      "fixture": "F21-cli-scheduler-priority",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass-2",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 33,
+          "solo_score": 66,
+          "pair_score": 99,
+          "pair_margin": 33,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": true
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains malformed-pair-evidence-row-fields "pair evidence rows 0 do not match summary count 2" \
+  python3 - "$SCRIPT" "$TMP_DIR/bad-frontier-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 2))
+PY
+
+cat > "$TMP_DIR/nan-frontier-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "nan-wall-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": NaN
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains nan-pair-evidence-row-fields "pair evidence count missing or malformed from frontier report" \
+  python3 - "$SCRIPT" "$TMP_DIR/nan-frontier-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/mismatched-margin-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "inflated-margin-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 76,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains mismatched-margin-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/mismatched-margin-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/overrange-score-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "overrange-score-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 101,
+          "pair_margin": 26,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains overrange-score-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/overrange-score-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/invalid-pair-arm-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "invalid-arm-run",
+          "pair_arm": "bare",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains invalid-pair-arm-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/invalid-pair-arm-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/false-pair-mode-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "false-pair-mode-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": false,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains false-pair-mode-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/false-pair-mode-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/missing-pair-trigger-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "stale-gate-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains missing-pair-trigger-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/missing-pair-trigger-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/zero-wall-row-fields.json" <<'JSON'
+{
+  "pair_evidence_count": 1,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "zero-wall-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 0
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains zero-wall-row-fields "pair evidence rows 0 do not match summary count 1" \
+  python3 - "$SCRIPT" "$TMP_DIR/zero-wall-row-fields.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/bool-frontier-count.json" <<'JSON'
+{
+  "pair_evidence_count": true,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains malformed-pair-evidence-count "pair evidence count missing or malformed from frontier report" \
+  python3 - "$SCRIPT" "$TMP_DIR/bool-frontier-count.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+cat > "$TMP_DIR/mismatched-frontier-rows.json" <<'JSON'
+{
+  "pair_evidence_count": 2,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "pair-pass",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.28
+        }
+      ]
+    },
+    {
+      "fixture": "F21-cli-scheduler-priority",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "incomplete-row",
+          "bare_score": 33,
+          "solo_score": 66,
+          "pair_score": 99,
+          "pair_margin": 33,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_solo_wall_ratio": 1.47
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains mismatched-pair-evidence-rows "pair evidence rows 1 do not match summary count 2" \
+  python3 - "$SCRIPT" "$TMP_DIR/mismatched-frontier-rows.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_min_pair_evidence(pathlib.Path(sys.argv[2]), 1))
+PY
+
+expect_fail_contains min-pair-evidence "pair evidence count 2 below required minimum 4" \
+  python3 "$SCRIPT" \
+    --fixtures-root "$fixtures" \
+    --registry "$registry" \
+    --results-root "$results" \
+    --out-dir "$TMP_DIR/out-low-evidence"
+grep -Fq 'FAIL audit-pair-evidence' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'headroom_rejections=PASS verdict=PASS unrecorded=0 unsupported=0' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'pair_evidence_quality=PASS min_pair_margin_actual=+21 min_pair_margin_required=+5 max_wall_actual=1.47x max_wall_allowed=3.00x' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'pair_trigger_reasons=PASS canonical=2 historical_alias=1 exposed=2 total=2 summary=2 rows_match=true' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'pair_trigger_historical_aliases=F21-cli-scheduler-priority=risk_profile.high_risk' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'pair_evidence_hypothesis_triggers=WARN matched=0 documented=2 total=2' "$TMP_DIR/min-pair-evidence.out"
+grep -Fq 'pair_evidence_hypothesis_trigger_gaps=F16-cli-quote-tax-rules=complexity.high;F21-cli-scheduler-priority=complexity.high,risk_profile.high_risk' "$TMP_DIR/min-pair-evidence.out"
+python3 - "$TMP_DIR/out-low-evidence/audit.json" <<'PY'
+import json
+import sys
+
+report = json.load(open(sys.argv[1], encoding="utf8"))
+assert report["verdict"] == "FAIL"
+assert report["checks"]["frontier"]["status"] == "PASS"
+assert report["checks"]["headroom_rejections"]["status"] == "PASS"
+assert report["checks"]["headroom_rejections"]["report_check_exit_code"] == 0
+assert report["checks"]["headroom_rejections"]["verdict"] == "PASS"
+assert report["checks"]["headroom_rejections"]["unrecorded_failure_count"] == 0
+assert report["checks"]["headroom_rejections"]["unsupported_registry_rejection_count"] == 0
+assert report["checks"]["min_pair_evidence"]["status"] == "FAIL"
+assert report["checks"]["min_pair_evidence"]["required"] == 4
+assert report["checks"]["min_pair_evidence"]["actual_rows"] == 2
+assert report["checks"]["pair_evidence_quality"]["status"] == "PASS"
+assert report["checks"]["pair_evidence_quality"]["min_pair_margin_actual"] == 21
+assert report["checks"]["pair_evidence_quality"]["max_pair_solo_wall_ratio_actual"] == 1.47
+assert report["checks"]["pair_trigger_reasons"]["status"] == "PASS"
+assert report["checks"]["pair_trigger_reasons"]["summary_pair_evidence_count"] == 2
+assert report["checks"]["pair_trigger_reasons"]["canonical_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["historical_alias_rows"] == 1
+assert report["checks"]["pair_trigger_reasons"]["historical_alias_details"] == [
+    {"fixture": "F21-cli-scheduler-priority", "aliases": ["risk_profile.high_risk"]}
+]
+assert report["checks"]["pair_trigger_reasons"]["exposed_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["total_rows"] == 2
+assert report["checks"]["pair_trigger_reasons"]["rows_match_count"] is True
+assert report["checks"]["pair_evidence_hypotheses"]["status"] == "PASS"
+assert report["checks"]["pair_evidence_hypotheses"]["documented_rows"] == 2
+assert report["checks"]["pair_evidence_hypotheses"]["total_rows"] == 2
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["status"] == "WARN"
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["exit_code"] == 0
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["required"] is False
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["matched_rows"] == 0
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["documented_rows"] == 2
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["total_rows"] == 2
+assert report["checks"]["pair_evidence_hypothesis_triggers"]["gap_details"] == [
+    {
+        "fixture": "F16-cli-quote-tax-rules",
+        "pair_trigger_reasons": ["complexity.high"],
+    },
+    {
+        "fixture": "F21-cli-scheduler-priority",
+        "pair_trigger_reasons": ["complexity.high", "risk_profile.high_risk"],
+    },
+]
+PY
+
+cat > "$TMP_DIR/low-quality-frontier.json" <<'JSON'
+{
+  "pair_margin_min": 4,
+  "pair_solo_wall_ratio_max": 1.2,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "low-quality-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 79,
+          "pair_margin": 4,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.2
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains low-quality-pair-evidence "pair evidence margin below minimum for fixture(s): F16-cli-quote-tax-rules" \
+  python3 - "$SCRIPT" "$TMP_DIR/low-quality-frontier.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+))
+PY
+python3 - "$SCRIPT" "$TMP_DIR/low-quality-frontier.json" > "$TMP_DIR/low-quality-quality-row.out" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+module.print_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+    status=1,
+)
+PY
+grep -Fq 'pair_evidence_quality=FAIL min_pair_margin_actual=+4 min_pair_margin_required=+5 max_wall_actual=1.20x max_wall_allowed=3.00x' "$TMP_DIR/low-quality-quality-row.out"
+
+cat > "$TMP_DIR/no-quality-rows-frontier.json" <<'JSON'
+{
+  "pair_margin_min": 21,
+  "pair_solo_wall_ratio_max": 1.2,
+  "rows": []
+}
+JSON
+expect_fail_contains no-quality-rows "pair evidence quality check has no complete rows" \
+  python3 - "$SCRIPT" "$TMP_DIR/no-quality-rows-frontier.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+))
+PY
+
+cat > "$TMP_DIR/high-wall-frontier.json" <<'JSON'
+{
+  "pair_margin_min": 21,
+  "pair_solo_wall_ratio_max": 3.5,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "high-wall-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 3.5
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains high-wall-pair-evidence "pair evidence wall ratio above maximum for fixture(s): F16-cli-quote-tax-rules" \
+  python3 - "$SCRIPT" "$TMP_DIR/high-wall-frontier.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+))
+PY
+
+cat > "$TMP_DIR/summary-mismatch-frontier.json" <<'JSON'
+{
+  "pair_margin_min": 22,
+  "pair_solo_wall_ratio_max": 1.2,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "summary-mismatch-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.2
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains summary-margin-mismatch "frontier pair_margin_min does not match pair evidence rows" \
+  python3 - "$SCRIPT" "$TMP_DIR/summary-mismatch-frontier.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+))
+PY
+
+cat > "$TMP_DIR/summary-wall-mismatch-frontier.json" <<'JSON'
+{
+  "pair_margin_min": 21,
+  "pair_solo_wall_ratio_max": 1.3,
+  "rows": [
+    {
+      "fixture": "F16-cli-quote-tax-rules",
+      "status": "pair_evidence_passed",
+      "passing_pair_evidence": [
+        {
+          "run_id": "summary-wall-mismatch-run",
+          "pair_arm": "l2_risk_probes",
+          "bare_score": 50,
+          "solo_score": 75,
+          "pair_score": 96,
+          "pair_margin": 21,
+          "pair_mode": true,
+          "pair_trigger_eligible": true,
+          "pair_trigger_reasons": ["complexity.high"],
+          "pair_trigger_has_canonical_reason": true,
+          "pair_solo_wall_ratio": 1.2
+        }
+      ]
+    }
+  ]
+}
+JSON
+expect_fail_contains summary-wall-mismatch "frontier pair_solo_wall_ratio_max does not match pair evidence rows" \
+  python3 - "$SCRIPT" "$TMP_DIR/summary-wall-mismatch-frontier.json" <<'PY'
+import importlib.util
+import pathlib
+import sys
+
+spec = importlib.util.spec_from_file_location("audit_pair_evidence", sys.argv[1])
+module = importlib.util.module_from_spec(spec)
+assert spec.loader is not None
+spec.loader.exec_module(module)
+sys.exit(module.check_pair_evidence_quality(
+    pathlib.Path(sys.argv[2]),
+    min_pair_margin=5,
+    max_pair_solo_wall_ratio=3.0,
+))
+PY
+
+echo "PASS test-audit-pair-evidence"