Background and Context. Modern advanced topics in computer science like LLMs and diffusion models are multi-foundational: they draw on multiple distinct fields and have complex prerequisite structures. This makes instructional sequencing particularly challenging, yet no systematic method exists for designing presentation orders for such topics.

Objectives. Drawing on knowledge space theory and cognitive load theory, we introduce PATHOS (Path Covering for Optimal Sequencing), an instructional sequencing strategy that is designed to minimize estimated intrinsic and extraneous cognitive load for multi-foundational topics.

Method. We evaluated PATHOS in a preregistered, randomized controlled trial (N=209) in a university AI course, comparing a PATHOS-derived sequence for teaching diffusion models against a control sequence following the discovery order of the field’s development. To complement our results, we also administered a survey on cognitive load, and conducted post-intervention interviews with students.

Findings. Students who received the PATHOS sequence scored significantly higher on a proctored learning assessment (p=0.011, Cohen’s d=0.32) and reported significantly lower intrinsic and extraneous cognitive load. Response-process evidence gathered from student interviews enriched these findings by showing how sequencing shaped students’ learning experience.

Implications. PATHOS demonstrates that sequencing alone can meaningfully improve learning outcomes for the multi-foundational topic of diffusion models, and provides instructors with a systematic, theory-grounded procedure for determining presentation order that is applicable beyond diffusion models.