Gauntlet
Evidence-first autonomous QA and candidate-bound verification for websites, AI discovery, agent actions, and coding-agent output.
The thesis
A model may propose an action or finding; deterministic policy, reproducible evidence, and exact candidate identity decide whether it can block or ship.
Evidence state
working. See the proof page for exactly what that state means and, more importantly, what it does not mean.
What exists
- Product contract, architecture, threat model, and schemas
- Minimal Playwright capture runner and evidence ledger
- Candidate-bound verification kernel
- Seeded defect benchmark foundation
Limits and boundaries
- Foundation status, not a claim of complete autonomous coverage
- Production is read-only unless an explicit policy grants a narrow action
- Missing dependencies and corrupt evidence produce BLOCKED, never PASS