Observability

Five-step progressive architecture (context enhancement, guardrails, router/gateway, caching, agents); model drift detection; orchestration frameworks; user feedback systems (explicit vs implicit, edit-as-preference, degenerate feedback loops, sycophancy)

Black-box vs white-box monitoring; metrics and pre-aggregation; SLIs/SLOs; burn rate alerting; three-layer ML monitoring taxonomy (golden signals/generic ML signals/domain-specific quality); four actuals cases; drift detection (PSI, KL divergence, Wasserstein); ML SLOs and privacy in monitoring

Observability as superset of monitoring; three telemetry sources (metrics, logs, traces); structured event logs (one event per work unit, request ID, sampling); distributed tracing (trace ID propagation, spans, Zipkin/X-Ray); metrics and traces as derived views of event logs; Newman: log aggregation first, correlation IDs, Jaeger, synthetic transactions

East–west traffic governance: mTLS, service authorisation, evolution from libraries to eBPF; sidecar as reuse mechanism for operational concerns; orthogonal coupling; data-sinking sidecar for EDM legacy integration

*Understanding Distributed Systems* — Roberto Vitillo