Take a Rust crate from a "works on my machine" laptop build to a production-grade artifact that ships through a gate no one can bypass. The course walks through a real 3-crate ETL workspace — etl-core, etl-cli, and etl-bench — built around clap derive macros, a tuned `[profile.release]`, and a multi-stage Dockerfile that drops a 1.8 GB rust:latest image to a 6 MB scratch+musl container with no shell to attack. You then wire pmat, bashrs, forjar, and pv onto the standard fmt + clippy + test + 100% coverage + audit + deny stack — because a green build badge is misleading when an agent wrote half the code. A five-job GitHub Actions matrix runs stable, beta, and Minimum Supported Rust Version on Linux, macOS, and Windows in parallel, with cached cold builds reduced to two-minute incremental ones. You close with criterion — warmup, statistical sampling, the HTML report — and a regression gate that fails the build on a measurable throughput drop. Every concept is tied to runnable code and Provable contracts in the case-study repo.