Module 06 Activity

Scenario

Everything with a writable rule is currently being improvised by a model. This activity moves it into code.

What you build

A flow owning the sequence and routing, a crew owning only the judgement steps, and a fingerprint that makes runs reproducible.

Steps

1. Sort every step into rule, judgement, or action. Count how many genuinely need judgement.
2. Move every writable rule into the flow. Where a case is ambiguous, add an explicit tie-break rather than delegating it.
3. Build the flow with @start() and @listen(), calling a crew only for the judgement steps.
4. Fingerprint every run input including prompt version, policy version, model and flow version.
5. Run the same fingerprint twice and report what differed.
6. List your system's claims and mark each provable or measurable. Move one claim across the line by moving a step into the flow.

Evidence to hand in

• The three-way sort with the judgement count.
• The flow, with the routing rules and any tie-breaks visible.
• The fingerprint and what it covers.
• Two same-fingerprint runs with the differences reported.
• The claims table, before and after moving one step.

Review checklist

• Ambiguous routing cases got an explicit tie-break, not a model.
• The fingerprint includes prompt and policy versions.
• The number of model steps is stated and is small.
• The claims table separates proven from measured.
• At least one claim moved from measured to proven, with the step that moved.

Mark complete