Merge pull request #12 from SharedIntellect/feature/wire-tester-phase3

v0.7.0: Wire TesterCritic into pipeline as Phase 3 (DEC-020)

Author: dacervera

.gitignore

@@ -3,3 +3,5 @@ ports/
 __pycache__/
 .pytest_cache/
 .hypothesis/
+.DS_Store
+quorum-runs/

README.md

@@ -5,7 +5,7 @@
 <p align="center">
   <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-2ba4c8.svg" alt="MIT License"></a>
   <img src="https://img.shields.io/badge/platform-OpenClaw-2ba4c8" alt="Platform: OpenClaw">
-  <img src="https://img.shields.io/badge/status-v0.6.0-2ba4c8" alt="Status: v0.6.0">
+  <img src="https://img.shields.io/badge/status-v0.7.0-2ba4c8" alt="Status: v0.7.0">
   <a href="https://clawhub.ai/dacervera/quorum"><img src="https://img.shields.io/badge/ClawHub-dacervera%2Fquorum-2ba4c8" alt="Available on ClawHub"></a>
 </p>
 

critic-status.yaml

@@ -2,7 +2,7 @@
 # Used by tools/validate-docs.py to detect stale documentation.
 # Update this file FIRST when shipping a critic or feature.
 
-version: "0.6.1"
+version: "0.7.0"
 spec_version: "3.0"
 
 critics:
@@ -34,10 +34,13 @@ critics:
     note: "Activated with --relationships flag; not part of default depth panel"
   tester:
     status: shipped
-    callable: false
+    callable: true
     since: "0.6.0"
     description: "Finding verification — L1 deterministic + L2 LLM claim checking"
-    note: "Code and tests shipped; not yet wired into pipeline (not in CRITIC_REGISTRY or depth configs)"
+    note: "Phase 3 in pipeline. L1 at standard depth, L1+L2 at thorough. Quick: skipped."
+    depth_configs:
+      - standard
+      - thorough
   architecture:
     status: planned
     callable: false

docs/CHANGELOG.md

@@ -4,6 +4,23 @@ All notable changes to Quorum will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [0.7.0] — 2026-03-12
+
+### Added
+- **TesterCritic wired into pipeline as Phase 3** (DEC-020). The Tester verifies other critics' findings after Phase 1 (single-file critics) and Phase 2 (cross-consistency), before the Aggregator produces the verdict.
+  - **L1 contradictions auto-excluded**: If a critic cites a file/line that doesn't exist or content that doesn't match, the finding is silently removed from verdict calculation. Preserved in tester output for audit.
+  - **L2 contradictions annotated**: If the Tester's LLM judges a claim unsupported by evidence, the finding gets a `[L2-CONTRADICTED]` annotation but remains in the verdict. Human decides.
+  - **Depth-based activation**: Quick = skip, Standard = L1 only (zero API cost), Thorough = L1 + L2.
+- **Tester stats in run manifest**: `verified`, `unverified`, `contradicted`, `verification_rate`, `runtime_ms`.
+- **Verification section in report.md** when tester results are present: table of verified/unverified/contradicted counts with per-finding exclusion details.
+- 12 new integration tests (`test_tester_pipeline_integration.py`). Total: 879 passed, 6 skipped.
+
+### Changed
+- `AggregatorAgent.run()` accepts optional `tester_result` parameter. Applies DEC-020 policy before deduplication.
+- `AggregatedReport` model gains `tester_result` and `l1_excluded_count` fields.
+- Verdict reasoning now includes excluded finding count when tester ran.
+- Removed unnecessary `asyncio.CancelledError` isinstance guards (CancelledError is BaseException in Python 3.9+, never caught by `except Exception`).
+
 ## [0.6.1] — 2026-03-11
 
 ### Changed

reference-implementation/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "quorum-validator"
-version = "0.6.1"
+version = "0.7.0"
 description = "Multi-critic quality validation for agents, research, and configurations"
 authors = [
     {name = "Daniel Cervera", email = "daniel@sharedintellect.com"},

reference-implementation/quorum/agents/aggregator.py

@@ -28,8 +28,10 @@
     CriticResult,
     Finding,
     Severity,
+    TesterResult,
     Verdict,
     VerdictStatus,
+    VerificationStatus,
 )
 from quorum.providers.base import BaseProvider
 
@@ -52,17 +54,26 @@ def __init__(self, provider: BaseProvider, config: QuorumConfig):
         self.provider = provider
         self.config = config
 
-    def run(self, critic_results: list[CriticResult]) -> Verdict:
+    def run(self, critic_results: list[CriticResult], tester_result: TesterResult | None = None) -> Verdict:
         """
         Main entry point.
 
         Args:
             critic_results: List of CriticResult from each critic
+            tester_result:  Optional TesterResult from Phase 3 verification
 
         Returns:
             Final Verdict with AggregatedReport attached
         """
         all_findings = self._collect_findings(critic_results)
+
+        # DEC-020: Apply tester results before deduplication
+        l1_excluded_count = 0
+        if tester_result is not None:
+            all_findings, l1_excluded_count = self._apply_tester_results(
+                all_findings, tester_result,
+            )
+
         deduped_findings, conflicts_resolved = self._deduplicate(all_findings)
         confidence = self._calculate_confidence(critic_results, deduped_findings)
 
@@ -71,16 +82,72 @@ def run(self, critic_results: list[CriticResult]) -> Verdict:
             confidence=confidence,
             conflicts_resolved=conflicts_resolved,
             critic_results=critic_results,
+            tester_result=tester_result,
+            l1_excluded_count=l1_excluded_count,
         )
 
-        verdict = self._assign_verdict(report)
+        verdict = self._assign_verdict(report, l1_excluded_count=l1_excluded_count)
         logger.info(
             "Aggregator: %d findings → %d deduped → verdict=%s (confidence=%.2f)",
             len(all_findings), len(deduped_findings), verdict.status.value, confidence,
         )
+        if l1_excluded_count > 0:
+            logger.info(
+                "Aggregator: %d finding(s) excluded by Tester (L1 contradicted)",
+                l1_excluded_count,
+            )
 
         return verdict
 
+    def _apply_tester_results(
+        self,
+        findings: list[Finding],
+        tester_result: TesterResult,
+    ) -> tuple[list[Finding], int]:
+        """Apply DEC-020: auto-exclude L1 contradictions, annotate L2 contradictions.
+
+        Returns (filtered_findings, l1_excluded_count).
+        """
+        # Build mapping: finding_id → VerificationResult
+        verification_map = {
+            vr.original_finding_id: vr
+            for vr in tester_result.verification_results
+        }
+
+        filtered: list[Finding] = []
+        l1_excluded = 0
+
+        for finding in findings:
+            vr = verification_map.get(finding.id)
+            if vr is None:
+                filtered.append(finding)
+                continue
+
+            if vr.status == VerificationStatus.CONTRADICTED and vr.level == 1:
+                # L1 CONTRADICTED: auto-exclude from verdict
+                l1_excluded += 1
+                logger.debug(
+                    "DEC-020: Excluding finding %s (L1 contradicted): %s",
+                    finding.id, vr.explanation,
+                )
+                continue
+
+            if vr.status == VerificationStatus.CONTRADICTED and vr.level == 2:
+                # L2 CONTRADICTED: annotate only, keep in verdict
+                annotated = finding.model_copy(
+                    update={"description": f"[L2-CONTRADICTED] {finding.description}"}
+                )
+                filtered.append(annotated)
+                logger.debug(
+                    "DEC-020: Annotating finding %s (L2 contradicted): %s",
+                    finding.id, vr.explanation,
+                )
+                continue
+
+            filtered.append(finding)
+
+        return filtered, l1_excluded
+
     def _collect_findings(self, results: list[CriticResult]) -> list[Finding]:
         """Flatten all findings from all critics into a single list."""
         findings = []
@@ -168,7 +235,7 @@ def _calculate_confidence(
 
         return round(evaluated_criteria / total_criteria, 3)
 
-    def _assign_verdict(self, report: AggregatedReport) -> Verdict:
+    def _assign_verdict(self, report: AggregatedReport, l1_excluded_count: int = 0) -> Verdict:
         """
         Deterministic verdict assignment based on findings severity.
 
@@ -222,6 +289,9 @@ def _assign_verdict(self, report: AggregatedReport) -> Verdict:
         if counts:
             reasoning += f" Issues: {', '.join(counts)}."
 
+        if l1_excluded_count > 0:
+            reasoning += f" {l1_excluded_count} finding(s) excluded by Tester (L1 contradicted)."
+
         return Verdict(
             status=status,
             reasoning=reasoning,

reference-implementation/quorum/configs/quick.yaml

@@ -1,5 +1,6 @@
 # Quick depth profile — fast feedback for low-stakes validation
 # Critics: correctness + completeness (2 critics, 0 fix loops)
+# Tester: skipped at quick depth
 # Use for: draft review, pre-commit checks, iterative development
 
 critics:

reference-implementation/quorum/configs/standard.yaml

@@ -1,5 +1,6 @@
 # Standard depth profile — balanced rigor for most work
 # Critics: correctness + completeness + security (3 critics)
+# Tester: L1 (deterministic) verification active — zero API cost
 # Use for: PR reviews, config changes, research submissions
 
 critics:

reference-implementation/quorum/configs/thorough.yaml

@@ -1,5 +1,6 @@
 # Thorough depth profile — maximum rigor for critical decisions
 # Critics: all 4 implemented critics (correctness, completeness, security, code_hygiene)
+# Tester: L1 + L2 (LLM claim check) verification active
 # Use for: production releases, security review, compliance artifacts
 # Add --relationships for cross-artifact consistency (Phase 2)
 

reference-implementation/quorum/models.py

@@ -125,6 +125,8 @@ class AggregatedReport(BaseModel):
     confidence: float = Field(ge=0.0, le=1.0)
     conflicts_resolved: int = Field(default=0)
     critic_results: list[CriticResult] = Field(default_factory=list)
+    tester_result: Optional["TesterResult"] = None
+    l1_excluded_count: int = Field(default=0, description="Findings excluded by Tester L1 contradictions")
 
     @property
     def critical_count(self) -> int: