Findings are grouped when they share the same file, line range (±2 lines), and category.
Categories: security, performance, architecture, correctness, style.
Groups reaching the promotion threshold get an agreement badge; others are reported individually.
Selected per pipeline via consensus: { strategy: <id> }:
| Strategy | Rule | Threshold |
|---|---|---|
overlap-v1 |
Absolute | ≥ requireAgreement (default 2) |
majority-v1 |
Strict majority | > n/2, min 2 |
severity-aware-v1 |
Severity-scaled | crit/high→1, med→2, low/info→3 |
Finding A B C overlap-v1 majority-v1 severity-aware-v1
auth:42 (crit) ✓ ✓ · promoted promoted promoted (≥1)
auth:45 (med) · ✓ ✓ — promoted promoted (≥2)
db:10 (low) ✓ · · — — — (needs ≥3)
Text-similarity grouping with optional contradiction detection. Configured via:
consensus:
strategy: semantic-v2
similarityThreshold: 0.78
enableContradictions: trueSee Architecture for the consensus engine internals, ConsensusStrategy interface, and ConsensusResult types.