# FDE Professional AI reliability and assurance lab

Run `python 11-Blog-Programme/labs/fde-pro-reliability/reliability_cases.py` from the repository root. The verifier executes twenty deterministic checks for task SLOs, retry amplification, circuit breaking, model release gates, grader calibration, production sampling, cache isolation, threat modelling, egress allowlists, secret rotation, queue metrics, load tests, model routing, client cost attribution, rollback thresholds, incident exercises, control evidence, inference deployment choices, workflow-state recovery and assurance reporting.

All requests, tenants, incidents, costs, model scores, identities and controls are authored fixtures. No model API, network, secret manager, cloud resource, production trace or client system is used. Several scenarios deliberately block release or require follow-up so a passing verifier does not imply that the fictional AI service passed every business gate.

The verifier must pass all 20 checks and save `reliability-verification.json`. Human calibration, security, legal, operational and editorial review remains pending.