Captures every model substitution as an official governance event, recording the requested model, actual model used, substitution reason (cost, availability, or policy), timestamp, agent ID, and session ID.

Compares the context window capability profile of requested versus actual models and flags material capability downgrades (>50% context reduction) as High or Critical risk.

Evaluates agent compliance policies against approved model whitelists. Substitutions landing outside approved bounds automatically trigger a compliance flag requiring review.

Computes historical compliance exposure across past substitution logs, auditing total requests served by unapproved models and high-risk exposure ratios.

Repository of model capability specifications sourced directly from official model provider documentation (OpenAI, Anthropic, Google, Meta, DeepSeek, Mistral, Cohere, xAI, Microsoft).