AI engineering · interactive portfolio work

Model Rollout Gate

Compare paired model evaluation evidence across global and required segments, with latency limits and reviewable pass / hold / block outcomes.

Browser-only demonstration
Model release review
A pass is evidence for a human—not a deploy button

Inspect paired champion/challenger scores, a bootstrap interval, required segments and p95 latency before a rollout decision.

Evaluation summary
Held-out rows—
Champion Brier—
Challenger Brier—
Paired delta · 95% CI—
Review controls

Adjust thresholds to explore the policy. This view reevaluates saved demo reports; it does not rerun bootstrap sampling.

Verdict
—

Latency evidence
— ratio

This reads saved synthetic reports from the source repository; no model is trained or served here. Bootstrap intervals are illustrative and not a formal equivalence proof. Keep holdout data private; a gate pass does not authorize deployment or replace production monitoring.

Companion demo for the source repository. The page evaluates checked-in synthetic reports; it does not deploy a model.