Module 06 Summary
What this module established
Correct output can come from a path that cost three times as much or acted before it checked. For agents the path is part of the behaviour you are guaranteeing, so score the trajectory as well as the answer.
Carry forward
- Approval violations are a count with a threshold of zero. A rate hides a single breach, and a single breach of an approval gate is an incident.
- Bound steps, tokens and time independently and record which fired. Ceiling stops are failures and belong in your failure rate.
- Score trajectories with booleans computed from recorded data. A model judging whether a path was reasonable is not reproducible.
Before moving on
Move on when your approval violation count is zero, all three ceilings exist, and two people scoring the same trajectory get identical numbers.
