The iterative design cycle is the structural core of learning engineering practice — the sequence of phases through which a learning solution is planned, built, tested, analyzed, and revised repeatedly until it achieves target learning outcomes or meets defined performance criteria. Unlike linear development models that treat design, development, and evaluation as sequential phases separated by formal reviews, the iterative design cycle treats these as continuously recurring activities whose outputs feed directly into the next iteration. Each cycle is simultaneously a development step and an experiment: it produces an improved learning solution and generates evidence about what works, what fails, and why. [LE-LS-GL-007]
Planning and Initial Design
Every iteration of the design cycle begins with a planning phase in which the team specifies what learning outcomes are targeted, what design hypotheses will be tested, and how success will be measured. This phase is more demanding than it first appears: specifying measurable learning outcomes requires decisions about what counts as evidence of learning (not just course completion or satisfaction ratings), and specifying design hypotheses requires the team to articulate specific, testable claims about how particular design choices will affect particular outcomes. Ann Brown's design-based research framework established that this upfront specification of what is being tested and why is essential for ensuring that iterative cycles generate genuine design knowledge rather than merely accumulating anecdotal experience. [LE-LS-AP-003]
The initial design phase translates planning specifications into concrete learning solution components: content, practice activities, assessments, adaptive sequencing logic, and data instrumentation. Cognitive Load Theory provides a key constraint on initial design: the cognitive demands of the designed solution on learners' working memory must be estimated and managed, with worked examples, scaffolding, and information integration deployed where extraneous load would otherwise exceed capacity. [LE-LS-AP-008] Kenneth Koedinger's research on learning curve analysis — examining how error rates decline across practice trials for individual knowledge components — provides quantitative methods for evaluating whether initial design decisions produced appropriate practice difficulty, or whether components are too easy (revealing insufficient challenge) or too hard (revealing unmet prerequisites). [LE-LS-PP-005]
Development, Testing, and Analysis
The development phase produces deployable learning solution components alongside the instrumentation infrastructure needed to collect analysis data. This pairing is non-negotiable in learning engineering practice: developing content without simultaneously designing the data collection that will enable iteration is like building a prototype with no way to test it. ASSISTments, developed at Worcester Polytechnic Institute by Neil Heffernan and colleagues, operationalized this principle by enabling teachers to author content within a platform that simultaneously ran controlled experiments comparing instructional variants — making every deployment both a learning experience for students and a test of design hypotheses for researchers and designers. [LE-LS-PP-011]
Testing includes both technical validation (ensuring the system functions as specified) and empirical validation (testing the learning solution with representative learners and collecting the data needed to evaluate design hypotheses). The analysis phase interprets this data to identify which design elements are working, which are failing, and what specific changes are most likely to improve performance in the next cycle. Bayesian Knowledge Tracing provides one template for this analysis at the level of individual knowledge components: by examining how mastery probability estimates evolve across practice for different learner subpopulations, designers can identify components with anomalous learning curves that indicate design problems. [LE-LS-AP-011] At larger scale, platform telemetry analysis — of the kind used in the doer effect research — can reveal cross-course design principles that should inform all subsequent iterations. [LE-LS-AP-012] The Learning Engineering Toolkit insists that analysis outputs be translated into documented revision decisions — specific changes with stated rationales and expected outcomes — before the next iteration begins. [LE-LS-GL-007]