Four measures for every controlled trial
01
ICP acceptance
Use a labeled test set
02
Evidence completeness
Track source and retrieval context
03
Human review time
Compare equal tasks
04
Deliverability outcomes
Observe by domain and segment
These are evaluation dimensions, not performance claims. Verification does not guarantee inbox placement, replies, compliance, or commercial results.
A defensible trial freezes its ICP definition, labels a small reference set, records provider and runtime conditions, and reviews false positives as carefully as accepted accounts. Teams should preserve uncertain fields, measure the effect of source freshness, and document every human override before comparing one workflow with another. Repeat the method by territory because data availability, consent standards, and channel rules vary.