How the
practice works.
In the guided bakery introduction, you help Hani investigate a booking mistake. Follow the request, change the assistant’s rules and use the event record to explain what improved.
The learning loop
Observe
Follow a collection request and notice where the assistant’s promise stops matching the booking record.
Predict
Say what you expect to happen before a fresh case, so you have something to compare with the result.
Build
Configure instructions and tool boundaries, such as confirming only after a successful reservation.
Run
Try the assistant’s rules in the simulated bakery. The booking tool can succeed, reject a slot or fail.
Inspect
Read the request, tool call and returned result. Check the booking record against what the assistant said.
Improve
Change the rule, replay the original conditions, then test an unfamiliar request or a tool outage.
Tomorrowville uses local scripted rules. Configuring instructions, supplying information and training model weights are different activities; this experience practises configuration and verification.
A booking mistake you can investigate
The assistant says a collection is booked, but the tool returned slot_unavailable. Check the ledger: no reservation was created.
Require a successful reservation before confirming. Replay the same request, then try an outage to check when the assistant should ask Hani for help.
Play the demo“You’re booked for 09:10.”
slot_unavailableNo reservation was created.
A statement is not a completed action. The repair is to confirm only after a successful reservation.
Inspect it in the mission