Evaluate: did the primary metric move meaningfully, are guardrail metrics unharmed, is the result statistically significant, and are the tradeoffs acceptable at the intended rollout scale?

Decision framework: (1) Statistical significance - enough data to trust the result. (2) Practical significance - effect size is large enough to matter. (3) Guardrail checks - no critical regression. (4) Long-term outlook - does the metric change hold over the experiment window (novelty effect?). (5) Launch risk - reversibility, blast radius. A "launch" decision is not permanent; staged rollouts with monitoring allow course-correction.