Let an AI clear out your false positives without letting it hide a real bug

Short summary

Synapse uses a two-model consensus system to triage false positives in CI security scans without letting AI delete real findings. A proposer model flags suspected false positives with a typed verdict, a different verifier model must independently agree at ≥75 confidence, and nothing is ever deleted—findings are only exempt from the gate's exit code. Deterministic pre-filtering strips obvious noise before any model runs, keeping token costs down.

• •Two-model consensus: proposer flags false positives, separate verifier must independently agree at ≥75 confidence
• •Nothing is ever deleted—refuted findings stay in reports and compliance tables, only exempt from gate exit code
• •Deterministic scope classifier pre-filters noise before model calls to save token costs

Generated with AI, which can make mistakes.