Optimizing the Delivery of Shamiri: A Series of Rapid A/B Tests (2024-2025)

Across two program years (2024–2025), we embedded six A/B tests within routine implementation, each holding the four-session intervention content constant and varying a single component of its delivery scaffold — provider recruitment, group size, supervisor caseload, training, or supervision structure. Adaptations meant to lower cost, widen reach, or simplify operations were framed as non-inferiority questions; the one adaptation that added cost was framed as a superiority question. A parallel routine quality-assurance (QA) stream of standard, unmodified delivery anchored the comparisons and was benchmarked against five prior Shamiri studies. Co-primary outcomes were depression (PHQ-8) and anxiety (GAD-7) at a four-week endpoint, analyzed by intention-to-treat with multiple imputation in the full sample and the primary clinically-elevated subsample (PHQ-8 ≥ 10 / GAD-7 ≥ 10); longitudinal mixed models estimated between-arm differences in change for the A/B tests against pre-specified margins and within-arm change for the QA stream. The analytic sample comprised 5,714 students.