Validation gate
Can realistically-collectable signals recover known high-fit customers and reject lookalikes?
Positives in sample
48
Hard negatives
8
Holdouts loaded
10
Derived from
scored companies
Gate metrics (auto-derived)
| Metric | Value | Gate |
|---|---|---|
| Evidence availability on known positives w/ public footprint | 33% | red |
| Discoverable recall | 2% | red |
| False positives on hard negatives | 13% | green |
Holdout inventory
Positive
4
Target Like
2
Hard Negative
3
Public Competitor
1
Rep-accepted, non-obvious-accepted, and cost-per-lead metrics come from the manual 120-company spike and rep dispositions — they are not auto-derivable. Run the spike first (see docs/VALIDATION-SPIKE.md). The build only proceeds on a green / strong-yellow gate.