Build garageTry it → read → next · ~10 min

Tutorials · Chapter D (4/4) · ~10 min

Eval gates in code

Try it → see it → read → next

Block deploys when **golden tasks fail** — eval is code, not a spreadsheet you forget.

On paths: Builder · Student

Optional on: Professional

Try tools: Agents & automation

Try yourself

Playground

Eval gate before ship

No agent or RAG change reaches prod without a golden task set.

10 golden tasks defined
Eval runs in CI on prompt change
Pass threshold documented (e.g. 8/10)
Deploy blocked on eval failure

Recap

What you just did

You checked four ship gates: golden set, CI hook, pass threshold, deploy block.

Read

Minimal gate

JSON file of { input, must_include, must_not_include } tasks
Script scores model output on staging
Exit code 1 if score < threshold
CI runs script on PR touching prompts/tools