devlyn-cli 2.3.0 → 2.3.2

package/benchmark/auto-resolve/scripts/full-pipeline-pair-gate.py

@@ -3,54 +3,298 @@
 
 This is stricter than headroom-gate.py. Headroom only says a candidate set is
 worth measuring. This gate says the measured L2 arm is usable evidence:
-bare and solo leave headroom, l2_gated is clean, gated pair actually fired, and
-the blind judge scores l2_gated materially above solo_claude.
+bare and solo leave headroom with complete comparable artifacts, the selected
+pair arm is evidence-clean, pair mode actually fired for a canonical trigger
+reason, and the blind judge scores the selected pair arm materially above
+solo_claude.
 """
 from __future__ import annotations
 
 import argparse
 import json
+import os
 import pathlib
+import re
 import sys
 from typing import Any
 
+SCRIPT_DIR = pathlib.Path(__file__).resolve().parent
+if str(SCRIPT_DIR) not in sys.path:
+    sys.path.insert(0, str(SCRIPT_DIR))
+FIXTURES_ROOT = SCRIPT_DIR.parent / "fixtures"
 
-def load_json(path: pathlib.Path) -> dict[str, Any] | None:
+from pair_evidence_contract import (
+    ALLOWED_PAIR_ARMS,
+    all_known_pair_trigger_reasons,
+    has_canonical_pair_trigger_reason,
+    has_known_pair_trigger_reason,
+    is_score,
+    is_strict_number,
+    loads_strict_json_object,
+    path_has_actionable_solo_headroom_hypothesis,
+)
+
+KNOWN_ARMS = {"bare", "solo_claude"} | ALLOWED_PAIR_ARMS
+PASS_VERDICTS = {"PASS", "PASS_WITH_ISSUES"}
+REJECTED_REGISTRY = pathlib.Path(__file__).with_name("pair-rejected-fixtures.sh")
+
+
+def load_json(path: pathlib.Path) -> tuple[dict[str, Any] | None, str | None]:
     if not path.is_file():
-        return None
-    return json.loads(path.read_text())
+        return None, "missing"
+    try:
+        data = loads_strict_json_object(path.read_text())
+    except (ValueError, json.JSONDecodeError):
+        return None, "malformed"
+    return data, None
+
+
+def fixture_short(name: str) -> str:
+    return name.split("-", 1)[0] if "-" in name else name
+
+
+def rejected_registry_path() -> pathlib.Path:
+    override = os.environ.get("PAIR_REJECTED_FIXTURES_REGISTRY")
+    return pathlib.Path(override) if override else REJECTED_REGISTRY
+
+
+def load_rejected_short_ids(path: pathlib.Path) -> set[str]:
+    if not path.is_file():
+        raise ValueError(f"rejected fixture registry missing: {path}")
+    rejected = set()
+    for line in path.read_text().splitlines():
+        match = re.match(r"\s*([FS]\d+)-\*\|([FS]\d+)\)", line)
+        if match and match.group(1) == match.group(2):
+            rejected.add(match.group(1))
+    if not rejected:
+        raise ValueError(f"rejected fixture registry has no fixture entries: {path}")
+    return rejected
 
 
 def score_for(judge: dict[str, Any], arm: str) -> int | None:
-    value = (judge.get("scores_by_arm") or {}).get(arm)
-    return value if isinstance(value, int) else None
+    mapping = judge.get("_blind_mapping")
+    if not isinstance(mapping, dict):
+        return None
+    if arm not in {mapped for slot, mapped in mapping.items() if slot in {"A", "B", "C"}}:
+        return None
+    raw_scores = judge.get("scores_by_arm")
+    scores = raw_scores if isinstance(raw_scores, dict) else {}
+    value = scores.get(arm)
+    return value if is_score(value) else None
 
 
-def clean_failures(fixture_dir: pathlib.Path, judge: dict[str, Any], arm: str) -> list[str]:
-    failures: list[str] = []
-    result = load_json(fixture_dir / arm / "result.json")
-    verify = load_json(fixture_dir / arm / "verify.json")
+def verify_score_clean(payload: dict[str, Any] | None) -> bool:
+    if payload is None:
+        return False
+    value = payload.get("verify_score")
+    return is_strict_number(value) and value >= 1.0
+
+
+def bool_flag_failure(value: Any, true_reason: str, malformed_reason: str) -> str | None:
+    if value is True:
+        return true_reason
+    if value is False or value is None:
+        return None
+    return malformed_reason
+
+
+def pair_trigger_failures(result: dict[str, Any] | None, arm: str) -> list[str]:
+    if result is None:
+        return []
+    trigger = result.get("pair_trigger")
+    if not isinstance(trigger, dict):
+        return [f"{arm} pair_trigger missing or malformed"]
+    eligible = trigger.get("eligible")
+    reasons = trigger.get("reasons")
+    skipped_reason = trigger.get("skipped_reason")
+    if not isinstance(eligible, bool):
+        return [f"{arm} pair_trigger.eligible malformed"]
+    if not isinstance(reasons, list) or not all(isinstance(reason, str) for reason in reasons):
+        return [f"{arm} pair_trigger.reasons malformed"]
+    if skipped_reason is not None and not isinstance(skipped_reason, str):
+        return [f"{arm} pair_trigger.skipped_reason malformed"]
+    if eligible is not True:
+        return [f"{arm} pair_trigger not eligible"]
+    if not reasons:
+        return [f"{arm} pair_trigger eligible with empty reasons"]
+    if not has_known_pair_trigger_reason(reasons):
+        return [f"{arm} pair_trigger reasons missing known trigger reason"]
+    if not all_known_pair_trigger_reasons(reasons):
+        return [f"{arm} pair_trigger reasons contain unknown trigger reason"]
+    if not has_canonical_pair_trigger_reason(reasons):
+        return [f"{arm} pair_trigger reasons missing canonical trigger reason"]
+    if skipped_reason is not None:
+        return [f"{arm} pair_trigger eligible with skipped_reason"]
+    return []
+
+
+def pair_trigger_eligible(result: dict[str, Any] | None) -> bool:
+    if result is None:
+        return False
+    trigger = result.get("pair_trigger")
+    return (
+        isinstance(trigger, dict)
+        and trigger.get("eligible") is True
+        and isinstance(trigger.get("reasons"), list)
+        and bool(trigger.get("reasons"))
+        and all(isinstance(reason, str) for reason in trigger.get("reasons", []))
+        and all_known_pair_trigger_reasons(trigger.get("reasons", []))
+        and has_canonical_pair_trigger_reason(trigger.get("reasons", []))
+        and trigger.get("skipped_reason") is None
+    )
+
+
+def pair_trigger_reasons(result: dict[str, Any] | None) -> list[str]:
     if result is None:
-        failures.append(f"{arm} result.json missing")
-    if verify is None:
-        failures.append(f"{arm} verify.json missing")
+        return []
+    trigger = result.get("pair_trigger")
+    if not isinstance(trigger, dict):
+        return []
+    reasons = trigger.get("reasons")
+    if not isinstance(reasons, list) or not all(isinstance(reason, str) for reason in reasons):
+        return []
+    return reasons
+
+
+def fixture_spec_has_solo_headroom_hypothesis(fixture: str) -> bool:
+    return path_has_actionable_solo_headroom_hypothesis(FIXTURES_ROOT / fixture / "spec.md")
+
+
+def skill_verdict_failures(result: dict[str, Any] | None, arm: str) -> list[str]:
+    if result is None or arm == "bare":
+        return []
+    failures: list[str] = []
+    terminal = result.get("terminal_verdict")
+    verify = result.get("verify_verdict")
+    if terminal not in PASS_VERDICTS:
+        failures.append(f"{arm} terminal verdict not pass")
+    if verify not in PASS_VERDICTS:
+        failures.append(f"{arm} verify verdict not pass")
+    return failures
+
 
-    dq_by_arm = judge.get("disqualifiers_by_arm") or {}
-    if bool((dq_by_arm.get(arm) or {}).get("disqualifier")):
-        failures.append(f"{arm} judge disqualifier")
+def axis_validation_counts(judge: dict[str, Any]) -> tuple[dict[str, int], int]:
+    raw_mapping = judge.get("_blind_mapping")
+    mapping = raw_mapping if isinstance(raw_mapping, dict) else {}
+    raw_validation = judge.get("_axis_validation")
+    validation = raw_validation if isinstance(raw_validation, dict) else {}
+    cells = validation.get("out_of_range_cells") or []
+    declared_count = validation.get("out_of_range_count")
+    total_invalid = max(
+        declared_count if isinstance(declared_count, int) else 0,
+        len(cells) if isinstance(cells, list) else 0,
+    )
+    breakdown_to_letter = {
+        "a_breakdown": "A",
+        "b_breakdown": "B",
+        "c_breakdown": "C",
+    }
+    counts: dict[str, int] = {}
+    mapped_count = 0
+    if not isinstance(cells, list):
+        return counts, total_invalid
+    for cell in cells:
+        if not isinstance(cell, dict):
+            continue
+        letter = breakdown_to_letter.get(cell.get("breakdown"))
+        arm = mapping.get(letter) if letter else None
+        if arm in KNOWN_ARMS:
+            counts[arm] = counts.get(arm, 0) + 1
+            mapped_count += 1
+    return counts, max(0, total_invalid - mapped_count)
+
+
+def axis_invalid_count(judge: dict[str, Any], arm: str) -> int:
+    counts, _ = axis_validation_counts(judge)
+    return counts.get(arm, 0)
+
+
+def axis_unmapped_invalid_count(judge: dict[str, Any]) -> int:
+    _, unmapped = axis_validation_counts(judge)
+    return unmapped
+
+
+def blind_mapping_failures(judge: dict[str, Any], required_arms: set[str]) -> list[str]:
+    mapping = judge.get("_blind_mapping")
+    if not isinstance(mapping, dict):
+        return ["judge blind mapping missing"]
+    mapped_arms = {arm for key, arm in mapping.items() if key in {"A", "B", "C"}}
+    missing = sorted(required_arms - mapped_arms)
+    if missing:
+        return [f"judge blind mapping missing arm(s): {', '.join(missing)}"]
+    return []
+
+
+def clean_failures(
+    fixture_dir: pathlib.Path,
+    judge: dict[str, Any],
+    arm: str,
+    *,
+    require_correctness: bool,
+) -> list[str]:
+    failures: list[str] = []
+    result, result_error = load_json(fixture_dir / arm / "result.json")
+    verify, verify_error = load_json(fixture_dir / arm / "verify.json")
+    diff = fixture_dir / arm / "diff.patch"
+    if result_error:
+        failures.append(f"{arm} result.json {result_error}")
+    if verify_error:
+        failures.append(f"{arm} verify.json {verify_error}")
+    if not diff.is_file():
+        failures.append(f"{arm} diff.patch missing")
+
+    raw_dq_by_arm = judge.get("disqualifiers_by_arm")
+    dq_by_arm = raw_dq_by_arm if isinstance(raw_dq_by_arm, dict) else {}
+    dq_entry = dq_by_arm.get(arm)
+    dq_value = dq_entry.get("disqualifier") if isinstance(dq_entry, dict) else dq_entry
+    judge_dq_failure = bool_flag_failure(
+        dq_value,
+        f"{arm} judge disqualifier",
+        f"{arm} judge disqualifier malformed",
+    )
+    if judge_dq_failure:
+        failures.append(judge_dq_failure)
+    axis_invalid = axis_invalid_count(judge, arm)
+    if axis_invalid > 0:
+        failures.append(f"{arm} judge axis-invalid ({axis_invalid})")
     if result is not None:
-        if bool(result.get("disqualifier")):
-            failures.append(f"{arm} result disqualifier")
-        if bool(result.get("timed_out")):
-            failures.append(f"{arm} timed out")
-        if bool(result.get("invoke_failure")):
+        for field, true_reason in (
+            ("disqualifier", f"{arm} result disqualifier"),
+            ("timed_out", f"{arm} timed out"),
+            ("environment_contamination", f"{arm} environment contamination"),
+        ):
+            failure = bool_flag_failure(
+                result.get(field),
+                true_reason,
+                f"{arm} result {field} malformed",
+            )
+            if failure:
+                failures.append(failure)
+        invoke_failure = bool_flag_failure(
+            result.get("invoke_failure"),
+            f"{arm} invoke failure",
+            f"{arm} result invoke_failure malformed",
+        )
+        if invoke_failure == f"{arm} invoke failure":
             reason = result.get("invoke_failure_reason")
             if isinstance(reason, str) and reason:
                 failures.append(f"{arm} invoke failure ({reason})")
             else:
-                failures.append(f"{arm} invoke failure")
-    if verify is not None and bool(verify.get("disqualifier")):
-        failures.append(f"{arm} verify disqualifier")
+                failures.append(invoke_failure)
+        elif invoke_failure:
+            failures.append(invoke_failure)
+        if require_correctness:
+            failures.extend(skill_verdict_failures(result, arm))
+    if verify is not None:
+        verify_dq_failure = bool_flag_failure(
+            verify.get("disqualifier"),
+            f"{arm} verify disqualifier",
+            f"{arm} verify disqualifier malformed",
+        )
+        if verify_dq_failure:
+            failures.append(verify_dq_failure)
+    if require_correctness and verify is not None and not verify_score_clean(verify):
+        failures.append(f"{arm} verify_score < 1.0")
     return failures
 
 
@@ -59,9 +303,7 @@ def elapsed_ratio(pair_result: dict[str, Any] | None, solo_result: dict[str, Any
         return None
     pair_elapsed = pair_result.get("elapsed_seconds")
     solo_elapsed = solo_result.get("elapsed_seconds")
-    if not isinstance(pair_elapsed, (int, float)) or not isinstance(solo_elapsed, (int, float)):
-        return None
-    if solo_elapsed <= 0:
+    if not is_strict_number(pair_elapsed) or not is_strict_number(solo_elapsed):
         return None
     return pair_elapsed / solo_elapsed
 
@@ -73,53 +315,78 @@ def provider_limited(result: dict[str, Any] | None) -> bool:
 def evaluate_fixture(
     fixture_dir: pathlib.Path,
     *,
+    rejected_short_ids: set[str],
     pair_arm: str,
     bare_max: int,
     solo_max: int,
+    min_bare_headroom: int,
+    min_solo_headroom: int,
     min_pair_margin: int,
     max_pair_solo_wall_ratio: float | None,
+    require_hypothesis_trigger: bool,
 ) -> dict[str, Any]:
-    judge = load_json(fixture_dir / "judge.json")
+    judge, judge_error = load_json(fixture_dir / "judge.json")
     if judge is None:
         return {
             "fixture": fixture_dir.name,
             "status": "FAIL",
-            "reason": "judge.json missing",
+            "reason": f"judge.json {judge_error}",
         }
 
     bare = score_for(judge, "bare")
     solo = score_for(judge, "solo_claude")
     pair = score_for(judge, pair_arm)
-    solo_result = load_json(fixture_dir / "solo_claude" / "result.json")
-    pair_result = load_json(fixture_dir / pair_arm / "result.json")
+    bare_headroom = bare_max - bare if isinstance(bare, int) else None
+    solo_headroom = solo_max - solo if isinstance(solo, int) else None
+    solo_result, _ = load_json(fixture_dir / "solo_claude" / "result.json")
+    pair_result, _ = load_json(fixture_dir / pair_arm / "result.json")
     ratio = elapsed_ratio(pair_result, solo_result)
     pair_provider_limited = provider_limited(pair_result)
     if pair_provider_limited:
         ratio = None
 
     reasons: list[str] = []
+    if fixture_short(fixture_dir.name) in rejected_short_ids:
+        reasons.append("fixture rejected for pair-candidate runs")
     if bare is None:
         reasons.append("bare score missing")
     elif bare > bare_max:
         reasons.append(f"bare score {bare} > {bare_max}")
+    elif bare_headroom is not None and bare_headroom < min_bare_headroom:
+        reasons.append(f"bare headroom {bare_headroom} < {min_bare_headroom}")
     if solo is None:
         reasons.append("solo_claude score missing")
     elif solo > solo_max:
         reasons.append(f"solo_claude score {solo} > {solo_max}")
+    elif solo_headroom is not None and solo_headroom < min_solo_headroom:
+        reasons.append(f"solo_claude headroom {solo_headroom} < {min_solo_headroom}")
     if pair_provider_limited:
         pass
     elif pair is None:
         reasons.append(f"{pair_arm} score missing")
     elif solo is not None and pair - solo < min_pair_margin:
         reasons.append(f"{pair_arm} margin {pair - solo:+d} < +{min_pair_margin}")
+    unmapped_axis_invalid = axis_unmapped_invalid_count(judge)
+    if unmapped_axis_invalid > 0:
+        reasons.append(f"judge axis-invalid unmapped ({unmapped_axis_invalid})")
+    reasons.extend(blind_mapping_failures(judge, {"bare", "solo_claude", pair_arm}))
 
-    reasons.extend(clean_failures(fixture_dir, judge, "bare"))
-    reasons.extend(clean_failures(fixture_dir, judge, "solo_claude"))
-    reasons.extend(clean_failures(fixture_dir, judge, pair_arm))
+    reasons.extend(clean_failures(fixture_dir, judge, "bare", require_correctness=False))
+    reasons.extend(clean_failures(fixture_dir, judge, "solo_claude", require_correctness=False))
+    reasons.extend(clean_failures(fixture_dir, judge, pair_arm, require_correctness=True))
 
     pair_mode = None if pair_result is None else pair_result.get("pair_mode")
     if pair_mode is not True and not pair_provider_limited:
         reasons.append(f"{pair_arm} pair_mode not true")
+    if not pair_provider_limited:
+        reasons.extend(pair_trigger_failures(pair_result, pair_arm))
+        if (
+            require_hypothesis_trigger
+            and
+            fixture_spec_has_solo_headroom_hypothesis(fixture_dir.name)
+            and "spec.solo_headroom_hypothesis" not in pair_trigger_reasons(pair_result)
+        ):
+            reasons.append(f"{pair_arm} pair_trigger missing spec.solo_headroom_hypothesis")
 
     if max_pair_solo_wall_ratio is not None and not pair_provider_limited:
         if ratio is None:
@@ -131,7 +398,9 @@ def evaluate_fixture(
         "fixture": fixture_dir.name,
         "status": "PASS" if not reasons else "FAIL",
         "bare_score": bare,
+        "bare_headroom": bare_headroom,
         "solo_score": solo,
+        "solo_headroom": solo_headroom,
         "pair_score": pair,
         "pair_margin": (
             None if pair_provider_limited
@@ -139,6 +408,14 @@ def evaluate_fixture(
             else None
         ),
         "pair_mode": pair_mode,
+        "pair_trigger_eligible": pair_trigger_eligible(pair_result),
+        "pair_trigger_reasons": pair_trigger_reasons(pair_result),
+        "pair_trigger_has_canonical_reason": has_canonical_pair_trigger_reason(
+            pair_trigger_reasons(pair_result)
+        ),
+        "pair_trigger_has_hypothesis_reason": (
+            "spec.solo_headroom_hypothesis" in pair_trigger_reasons(pair_result)
+        ),
         "pair_solo_wall_ratio": ratio,
         "reason": "; ".join(reasons),
     }
@@ -148,27 +425,50 @@ def fmt_ratio(value: Any) -> str:
     return f"{value:.2f}x" if isinstance(value, (int, float)) else "n/a"
 
 
+def fmt_margin(value: Any) -> str:
+    return f"{value:+.1f}" if isinstance(value, (int, float)) else "n/a"
+
+
+def fmt_trigger_reasons(value: Any) -> str:
+    if not isinstance(value, list) or not all(isinstance(item, str) for item in value):
+        return ""
+    return ",".join(value)
+
+
 def write_md(path: pathlib.Path, report: dict[str, Any]) -> None:
     lines = [
         f"# Full-Pipeline Pair Gate - {report['run_id']}",
         "",
         f"Verdict: **{report['verdict']}**",
         "",
+        f"Fixtures passed: {report['fixtures_passed']}/{report['fixtures_total']} "
+        f"(minimum required: {report['min_fixtures']})",
+        "",
         f"Rule: at least {report['min_fixtures']} fixtures; bare <= {report['bare_max']}; "
-        f"solo_claude <= {report['solo_max']}; {report['pair_arm']} clean; pair_mode true; "
+        f"bare headroom >= {report['min_bare_headroom_required']}; "
+        f"solo_claude <= {report['solo_max']}; "
+        f"solo_claude headroom >= {report['min_solo_headroom_required']}; "
+        f"{report['pair_arm']} evidence-clean; pair_mode true; "
+        "pair_trigger eligible with canonical reason; "
         f"{report['pair_arm']} - solo_claude >= {report['min_pair_margin']}.",
-        f"Max pair/solo wall ratio: {fmt_ratio(report['max_pair_solo_wall_ratio'])}",
+        f"Average pair margin: {fmt_margin(report['avg_pair_margin'])}",
+        f"Allowed pair/solo wall ratio: {fmt_ratio(report['max_pair_solo_wall_ratio'])}",
+        f"Maximum observed pair/solo wall ratio: {fmt_ratio(report['max_observed_pair_solo_wall_ratio'])}",
         f"Average pair/solo wall ratio: {fmt_ratio(report['avg_pair_solo_wall_ratio'])}",
+        f"Hypothesis trigger required: {str(report['require_hypothesis_trigger']).lower()}",
         "",
-        "| Fixture | Bare | Solo | Pair | Margin | Pair mode | Wall ratio | Status | Reason |",
-        "|---|---:|---:|---:|---:|---|---:|---|---|",
+        "| Fixture | Bare | Bare headroom | Solo_claude | Solo_claude headroom | Pair | Margin | Pair mode | Hypothesis trigger | Triggers | Wall ratio | Status | Reason |",
+        "|---|---:|---:|---:|---:|---:|---:|---|---|---|---:|---|---|",
     ]
     for row in report["rows"]:
         margin = row.get("pair_margin")
         margin_text = f"{margin:+d}" if isinstance(margin, int) else "n/a"
         lines.append(
-            f"| {row['fixture']} | {row.get('bare_score')} | {row.get('solo_score')} | "
+            f"| {row['fixture']} | {row.get('bare_score')} | {row.get('bare_headroom')} | "
+            f"{row.get('solo_score')} | {row.get('solo_headroom')} | "
             f"{row.get('pair_score')} | {margin_text} | {str(row.get('pair_mode')).lower()} | "
+            f"{str(row.get('pair_trigger_has_hypothesis_reason')).lower()} | "
+            f"{fmt_trigger_reasons(row.get('pair_trigger_reasons'))} | "
             f"{fmt_ratio(row.get('pair_solo_wall_ratio'))} | {row['status']} | {row.get('reason', '')} |"
         )
     lines.append("")
@@ -177,38 +477,82 @@ def write_md(path: pathlib.Path, report: dict[str, Any]) -> None:
 
 def positive_float(value: str) -> float:
     parsed = float(value)
+    if not is_strict_number(parsed):
+        raise argparse.ArgumentTypeError("value must be finite and > 0")
+    return parsed
+
+
+def positive_int(value: str) -> int:
+    parsed = int(value)
     if parsed <= 0:
         raise argparse.ArgumentTypeError("value must be > 0")
     return parsed
 
 
+def non_negative_int(value: str) -> int:
+    parsed = int(value)
+    if parsed < 0:
+        raise argparse.ArgumentTypeError("value must be >= 0")
+    return parsed
+
+
 def main() -> int:
     parser = argparse.ArgumentParser()
     parser.add_argument("--run-id", required=True)
     parser.add_argument("--results-root", default="benchmark/auto-resolve/results", type=pathlib.Path)
     parser.add_argument("--bare-max", type=int, default=60)
     parser.add_argument("--solo-max", type=int, default=80)
-    parser.add_argument("--min-pair-margin", type=int, default=5)
-    parser.add_argument("--min-fixtures", type=int, default=2)
-    parser.add_argument("--pair-arm", default="l2_gated")
-    parser.add_argument("--max-pair-solo-wall-ratio", type=positive_float)
+    parser.add_argument("--min-bare-headroom", type=non_negative_int, default=5)
+    parser.add_argument("--min-solo-headroom", type=non_negative_int, default=5)
+    parser.add_argument("--min-pair-margin", type=positive_int, default=5)
+    parser.add_argument("--min-fixtures", type=positive_int, default=2)
+    parser.add_argument("--pair-arm", default="l2_risk_probes")
+    parser.add_argument("--max-pair-solo-wall-ratio", type=positive_float, default=3.0)
+    parser.add_argument(
+        "--require-hypothesis-trigger",
+        action="store_true",
+        help="require fixtures with actionable solo-headroom hypotheses to expose spec.solo_headroom_hypothesis in pair_trigger.reasons",
+    )
     parser.add_argument("--out-json", type=pathlib.Path)
     parser.add_argument("--out-md", type=pathlib.Path)
     args = parser.parse_args()
 
+    if args.pair_arm == "l2_forced":
+        print(
+            "pair-arm l2_forced is retired: it leaks pair-awareness before IMPLEMENT; "
+            "use l2_risk_probes for current proof runs or l2_gated for diagnostics.",
+            file=sys.stderr,
+        )
+        return 2
+    if args.pair_arm not in ALLOWED_PAIR_ARMS:
+        print(
+            f"pair-arm must be one of {sorted(ALLOWED_PAIR_ARMS)}: {args.pair_arm}",
+            file=sys.stderr,
+        )
+        return 2
+
     run_root = args.results_root / args.run_id
     if not run_root.is_dir():
         print(f"no results dir: {run_root}", file=sys.stderr)
         return 2
 
+    try:
+        rejected_short_ids = load_rejected_short_ids(rejected_registry_path())
+    except ValueError as exc:
+        print(str(exc), file=sys.stderr)
+        return 2
     rows = [
         evaluate_fixture(
             fixture_dir,
+            rejected_short_ids=rejected_short_ids,
             pair_arm=args.pair_arm,
             bare_max=args.bare_max,
             solo_max=args.solo_max,
+            min_bare_headroom=args.min_bare_headroom,
+            min_solo_headroom=args.min_solo_headroom,
             min_pair_margin=args.min_pair_margin,
             max_pair_solo_wall_ratio=args.max_pair_solo_wall_ratio,
+            require_hypothesis_trigger=args.require_hypothesis_trigger,
         )
         for fixture_dir in sorted(p for p in run_root.iterdir() if p.is_dir())
     ]
@@ -218,11 +562,24 @@ def main() -> int:
     ratios = [
         row["pair_solo_wall_ratio"]
         for row in rows
-        if isinstance(row.get("pair_solo_wall_ratio"), (int, float))
+        if is_strict_number(row.get("pair_solo_wall_ratio"))
+    ]
+    margins = [
+        row["pair_margin"]
+        for row in rows
+        if isinstance(row.get("pair_margin"), int)
     ]
+    rule = (
+        "headroom candidates only; "
+        f"bare headroom >= {args.min_bare_headroom}; "
+        f"solo_claude headroom >= {args.min_solo_headroom}; "
+        f"{args.pair_arm} must be evidence-clean, pair_mode true, "
+        "pair_trigger eligible with a canonical reason, and beat solo_claude "
+        "by the configured margin"
+    )
     report = {
         "run_id": args.run_id,
-        "rule": "headroom candidates only; l2_gated must be clean, pair_mode true, and beat solo_claude by the configured margin",
+        "rule": rule,
         "verdict": verdict,
         "fixtures_total": len(rows),
         "fixtures_passed": pass_count,
@@ -230,9 +587,14 @@ def main() -> int:
         "fixture_count_ok": fixture_count_ok,
         "bare_max": args.bare_max,
         "solo_max": args.solo_max,
+        "min_bare_headroom_required": args.min_bare_headroom,
+        "min_solo_headroom_required": args.min_solo_headroom,
         "min_pair_margin": args.min_pair_margin,
         "pair_arm": args.pair_arm,
+        "require_hypothesis_trigger": args.require_hypothesis_trigger,
         "max_pair_solo_wall_ratio": args.max_pair_solo_wall_ratio,
+        "max_observed_pair_solo_wall_ratio": max(ratios) if ratios else None,
+        "avg_pair_margin": (sum(margins) / len(margins)) if margins else None,
         "avg_pair_solo_wall_ratio": (sum(ratios) / len(ratios)) if ratios else None,
         "rows": rows,
     }