Current educational discourse holds that effective pedagogy requires engagement through active student participation with subject matter relating to them. The lack of testing of lessons in series is recognized as a potential weakness in the evidence base, not least because standard parallel designs cannot capture serial interaction effects (cf. drug interactions). However, logistic issues make large-scale replicated in situ assessment of serial designs challenging. The recent introduction of evolution into the UK primary school curriculum presents a rare opportunity to overcome this. We implemented a randomised control 2 × 2 design with four inexpensive schemes of work, comparable to drug interaction trials. This involved an initial test phase (N = 1152) with replication (N = 1505), delivered by teachers, after training, in their classrooms with quantitative before-after-retention testing. Utilising the "genetics-first" approach, the schemes comprised four lessons taught in the same order. Lessons 1 (variation) and 3 (deep-time) were invariant. Lesson 2 (selection) was either student-centred or teacher-centred, with subject organism constant, while lesson 4 (homology) was either human-centred or not, with learning mode constant. All four schemes were effective in replicate, even for lower ability students. Unexpectedly, the teacher-focused/non-human centred scheme was the most successful in both test and replicate, in part owing to a replicable interaction effect but also because it enabled engagement. These results highlight the importance of testing lessons in sequence and indicate that there are many routes to effective engagement with no "one-size fits all" solution in education.