205: Testing with Agents
Key question: What does a test suite prove when an agent writes it and a model runs in it?
This Chapter's ACIs
| ACI | Name | Takeaway |
|---|---|---|
| ACI-044 | The Suite Has a Blast Radius | A suite reaches past its assertions. Force the offline, deterministic setting in the test helper, and run the tests the change can reach. |
| ACI-045 | Measuring with a Stochastic Instrument | A model that plays or grades is an instrument with sampling error. Print the sample, commit the design before the data, and give every ablation a red path. |
| ACI-046 | The Seam That Takes Model Output | Put each guarantee where the model cannot get past it: a schema, a clamp with one home, a parse that fails closed. |
| ACI-047 | Model and Cost Are Configuration and Measurement | Bind each model version once, behind a role and a fence with a floor. State every size and cost with its convention. |