observability

A prompt assembly failure can send a naked question to an AI chat and still pass as grounded synthesis. A cheap input-side guard keeps the chain honest.

TL;DR: The pipeline sent a naked question to the AI with zero context, yet the generic answer passed all validation because checks only ran on output, never input. The fix is a cheap context_present() gate requiring either 400 characters or a URL, tagging results as missing or ok. It warns without blocking, since harsh guards get deleted.

Two numbers for one server: risk score and health score point in opposite directions, and my sidebar was still reading the old one.

TL;DR: A dashboard showed conflicting scores for the same server because the sidebar still read the old risk-based fields (100 means critical) while everything else used the new health-based ones (100 means healthy). The lesson: invert metric direction once at the API boundary, never per component. Treat opposite-direction metric renames as data contract migrations, not simple renames.