version: "1.0.2"
Do not skip trigger testing. A good-sounding description is not evidence; the standard requires ≥80% trigger rate and the checklist targets ≥90% activation.
Create should-trigger and should-not-trigger queries with synonyms, abbreviations, adjacent domains, and file/keyword cues. Run the evaluation, inspect false positives and negatives, revise the What + When description with surgical triggers, and rerun until the measured result meets target without unacceptable over-triggering. Record the query set and result.
“Sounds good enough,” “the name is obvious,” and “we can test later” are red flags. Stop and restart from the labeled query set; do not claim completion from prose review alone.