Module 08 Summary

What this module established

Six dimensions, and permission violations is a count with a threshold of zero rather than a rate. Build the case set from real requests and include the ones where answering would be wrong - if every case expects an answer, refusal and escalation are untested.

Carry forward

• Score the path alongside the outcome with mechanical checks: redundant calls and tools outside the expected set.
• Include the cases that lower the score, especially messy real input. A set that always passes is not a measurement.
• Store per-case results. The broken list is the decision; the total hides regressions in exactly the behaviour that was hardest to build.

Before moving on

Move on when most of your cases expect something other than an answer, and one change has been judged on its broken list rather than its total.

Mark complete