A controlled, experiment-first testbed that measures the performance and resource cost of observability instrumentation in a containerized microservice system. It compares five additive observability modes (none, metrics, +logs, +traces, full OpenTelemetry pipeline) against a true uninstrumented baseline with statistical rigor (open-model load, bootstrap confidence intervals, non-parametric tests, effect sizes), measures objective failure detection (mean-time-to-detect via Prometheus alert firing), and includes 1M-row PostgreSQL/MongoDB indexing experiments and a Docker Compose-vs-Swarm-vs-Kubernetes comparison. Code is MIT-licensed; experimental data and figures are CC-BY-4.0.