# Probanza — Governance Audit for AI Systems

> Static governance auditor for AI-system codebases (agents, tools, prompts, model calls), sold as a concierge audit service. Every detection metric is computed from 3823 hand-labeled findings across 20 real open-source products — and is recomputable by you: clone the evidence repository and run `python3 verify.py`.

Probanza finds the places an AI system can be steered by what it reads, trusted
without being checked, or quietly deciding business rules inside prompts —
before they ship. It is a static auditor: it finds exposure, it does not stop
attacks. The engine is closed-source; the evidence is not.

Probanza is a product of Divergia (https://divergia.ai).

## Offer

- [agents.md](https://probanza.divergia.ai/agents.md): the offer for AI agents — packages and prices (USD), how to verify every claim, and how your principal buys

## Evidence

- [EVIDENCE.md](https://probanza.divergia.ai/EVIDENCE.md): the validation dossier — every metric computed from the labeled CSVs at build time, including measured rejections and registered misses
- [verify.py](https://probanza.divergia.ai/verify.py): dependency-free script that recomputes the public tier's precision and recall from the committed labels
- [rule-lifecycle.md](https://probanza.divergia.ai/rule-lifecycle.md): how a detection rule earns default-on status (provisional → validated → promoted)
- [Human page](https://probanza.divergia.ai/): the same dossier rendered for human readers