consensus-validation

name	consensus-validation
description	This skill should be used when the user asks to "check consensus", "validate agreement", "evaluate review threshold", "check autonomy mode", "verify consensus gate", or mentions consensus validation, agreement threshold, 75% gate, autonomy modes, or ConsensusGate evaluation.

Consensus Validation

ConsensusGate (from src/mde/consensus.py)

The consensus gate evaluates agreement across multi-model review results.

Threshold

• 75% agreement required for consensus to pass
• Ported from claude-octopus v9.13.0 quality.sh evaluate_quality_branch()

Autonomy Modes

Mode	Behavior
FULL	Auto-commit if consensus passes and quality gate passes
SUPERVISED	Auto-commit with notification to human
GATED	Require human approval before commit
MANUAL	Human must review and commit manually

Consensus Decision Flow

1. Collect structured review results from all models
2. Count agreements on each finding (severity + category match)
3. Calculate agreement percentage per finding
4. Apply 75% threshold to determine consensus
5. Map consensus result to autonomy mode action

Using the Python Module

from mde.consensus import ConsensusGate, ConsensusDecision, AutonomyMode

gate = ConsensusGate(threshold=0.75)
decision = gate.evaluate(review_results)

if decision.passed:
    # Consensus reached — apply autonomy mode
    ...
else:
    # Escalate to human review
    ...

Debate Integrity Integration

Before evaluating consensus, debate integrity rules must pass:

• DebateIntegrityRule checks from src/mde/consensus.py
• Circular reasoning detection
• Severity inflation detection
• Rubber-stamp detection

Rules

• 75% threshold is mandatory — never lower it
• Debate integrity must pass before consensus evaluation
• Failed consensus always escalates to human review
• Log consensus decisions with full rationale for audit trail
• Never auto-commit in GATED or MANUAL autonomy modes