[base76]

Happy to answer questions from anyone testing it.

  The core loop is simple: every request through the gateway gets a trace_id, a trust score, and a signed decision context. The audit trail stays in your own infrastructure — no external  
  calls.                                                                                                                                                                                    

  One thing we're actively working on: calibrating trust thresholds per risk category under EU AI Act Article 13. If anyone's done similar work on confidence scoring for high-risk AI
  systems, I'd genuinely like to compare notes.

  Repo + 5-minute quickstart: https://github.com/base76-research-lab/cognos-proof-engine