Method

How we test

This build describes the evidence contract before publishing outcome claims. A full wave uses 12 scripted calls for every service, with the same scripts for KaiCalls and every other candidate.

The 12 calls

1. After-hours appointment booking
2. A published price question
3. An unpublished price question
4. Emergency escalation
5. A caller asking for a human
6. A Spanish-language request
7. A transfer request
8. Message capture with tricky spellings
9. An interruption and off-script question
10. A spam-pattern call
11. Change or cancel the first booking
12. A repeat caller context check

What gets recorded

• Whether the job finished correctly
• Caller facts captured correctly
• Invented facts, if any
• Time to resolution
• Whether escalation followed the vendor documentation

There are no published call outcomes in this first build. A service that cannot be called is labeled clearly rather than treated as tested.