llm-observability

Our incident report was a vendor blog post

Anthropic's invisible-guardrails apology misses the point: production agents need output contracts, audit ledgers, and sentinel checks, not model-default trust.