Observe, evaluate, ship

Trace every run. Scope evals in CI that fail when a gate goes missing.

What you will build

By the end you will have nested traces for every run, behavior-scoped evals that state what must and must not happen, and a CI gate that fails the build the moment a safety gate goes missing.

Why it matters

Safety code is easy to delete by accident. A refactor drops the confirm check, every unit test still passes because none of them asserted it, and the next deploy issues refunds without approval. Evals that check behavior, wired into CI, turn that silent regression into a red build.

Inside this module

1. 1.Trace every run
   Record nested spans with timing and attributes for every step of a run.
2. 2.An agent under test
   Make the safety gate a config switch so evals can prove it matters.
3. 3.Evals scoped to behavior
   Each eval states what must and must not happen, and some are marked safety.
4. 4.A CI gate
   Fail the build on any safety failure or a pass rate below baseline.

Every step comes with the exact code diff, a run command, the expected output, a common mistake, a production note and a checkpoint. Then 2 graded exercises (MORE_CASES, slowest) checked by npm test, a solution branch, and a 3-question quiz. TypeScript, scripted model, no API key.