I review one real AI, inference, or agent workflow using redacted traces or exports. You receive a baseline of reliability, latency, retries, tool failures, authorization signals, and cost, plus a prioritized next-step memo and one measured improvement opportunity. Read-only, evidence-gated, and designed to finish in one week. The diagnostic is $1,500 and can be credited toward a separate implementation sprint if the evidence supports it. Public proof: https://jtabsbm.github.io/guardrail-lens/