Module 09 Activity

Scenario

The crew works. This activity is about what it costs when it does not.

What you build

Source-based checking, both loop detectors, a three-ceiling budget, and a failure report that ends in a regression test.

Steps

1. Give every checking agent the SOURCE rather than the previous agent's output, with a rule approving nothing the source does not state.
2. Construct the agreement failure deliberately: plant a wrong fact early and confirm your reviewer now catches it.
3. Add a hop ceiling and confirm a loop terminates. Report the cost with and without.
4. Reject unregistered tool names with the real list in the error, and measure your invention rate.
5. Implement a budget with token, money and time ceilings, checked after every step. Size it from the worst legitimate run.
6. Write a failure report ending in a regression test, and add the test.

Evidence to hand in

• The checking agent's source and rule.
• The planted wrong fact and evidence the reviewer caught it.
• Loop cost with and without a ceiling.
• The invention rate and what it says about goal versus tools.
• The three-ceiling budget with its sizing rationale.
• The failure report and the regression test now in your suite.

Review checklist

• The reviewer checks against a source, not the previous agent's output.
• The agreement failure was constructed and caught.
• All three budget ceilings exist and are checked after every step.
• Budget sizing comes from the worst legitimate run, not the typical one.
• The regression test was actually added, not merely described.

Mark complete