Debrief

The debrief and compressed course close share the final 30 minutes. Stop coding when they begin, even when targets are still red. Use the full course retrospective after the workshop when no additional time is available.

The trace and revision (8 minutes)

• Show the worst decision trace of the afternoon. What did you change to eliminate it, where does that live in the types, and what did it cost on the latency target?
• What policy amendment came from the trace, and what changed in the measured result?
• Which corner of the noise / false-alarm / latency triangle did you concede, and who at work should be making that call?

The compact safety case (10 minutes)

Each pair reads one row from SAFETY-NOTES.md:

• claim;
• evidence;
• assumption;
• known limitation; and
• next verification action.

Ask whether the evidence actually speaks to the claim. A green command is not an answer unless the pair can explain what it measured.

The test suite and metrics (9 minutes)

• Each pair: count your suite by layer (example, property, statistical). What did the agent’s first suite look like by the same count?
• Show the best property of the day. What bug does it catch that no example test could?
• Your before/after metrics: trial count, replayed seed, and remaining miss. Which statement is only a measured rate, and what further work would be required to call it a confidence-supported claim?

Closing question (3 minutes)

The acceptance table judged guardian in aggregate. The processes that judge your real systems work the same way, and increasingly so do the ones that judge LLM-produced code. What changes in your team’s definition of “tested” on Monday?