Metrx verification artifact
Sample — synthetic datagpt-4o-mini vs gpt-4o
Verified as of
This is a synthetic sample showing what a purchased verification artifact looks like. Every number on this page is fabricated. A real artifact is computed from your own production traffic.
On this traffic, the candidate model held quality within the non-inferiority margin and the cost difference cleared the threshold. The evidence supports switching.
Observed cost difference
38.2%
95% CI 31.0% – 44.9%
Mean quality delta
-0.012
95% CI -0.041 – 0.018
Paired samples
42
Quality held (non-inferior)
Yes
gpt-4o cost (window)
$412.53
gpt-4o-mini cost (window)
$254.94
Method: holdout_confirmed paired comparison judged by claude-sonnet-4-5.
Evidence window: May 31, 2026 → June 30, 2026
This artifact is an immutable snapshot issued at verification time. Its numbers never change after issuance.