Low participation has long shadowed mathematics instruction in Indonesian vocational high schools, where teacher-led delivery leaves learners watching rather than thinking. This study tested whether delivering the demonstration phase of the Direct Instruction (DI) model through purpose-built videos raises learning engagement and mathematics achievement, and explored how the two outcomes are related. A quasi-experiment with a non-equivalent pre-test-post-test control-group design was conducted with 58 tenth-grade students at a vocational school in West Sumatra, allocated to video-assisted DI (n = 30) or conventional DI (n = 28). Videos followed a documented design specification, and implementation fidelity, viewing compliance, and instructional time were recorded in both classes. Because a non-significant baseline test is not evidence of equivalence, the primary analysis was an analysis of covariance on post-test achievement with the pre-test as covariate, with familywise error controlled across both confirmatory outcomes. Engagement was higher in the experimental class by 10.92 points, 95% CI [7.90, 13.94], t(56) = 7.25, Hedges' g = 1.88. Adjusted post-test achievement was higher by 13.16 points, 95% CI [9.66, 16.67], F(1, 55) = 56.73, partial η² = .51, and remained stable across separate-slopes, Welch, and rank-based analyses. Engagement and achievement were strongly associated, r = .74, 95% CI [.60, .84], though this fell to .43 and .62 within classes. An exploratory decomposition attributed roughly half the adjusted group difference to engagement; because engagement was measured once, concurrently with the post-test, this is a statistical partition rather than a causal mechanism. Findings from a single school require replication.