Research & References

How it informed the tool: Teacher-predicting-student framing produces better profile alignment than direct role-play. The KLI (Knowledge-Learning-Instruction) framework informed how we structure knowledge state dimensions.

How it informed the tool: Identified the "hyper-accuracy distortion" — LLMs struggle to be wrong on purpose and default to correct answers. This motivated our emphasis on detailed misconception descriptions in the knowledge state section.

How it informed the tool: Introduced deep, surface, and lazy learning profiles with distinct behaviors. Showed that surface learners fail on "trap questions" (same structure, different context) and that self-efficacy should evolve dynamically within a tutoring session.

How it informed the tool: Comprehensive survey of cognitive modeling approaches for simulated students. Reinforced working memory and metacognitive awareness as key simulation dimensions alongside knowledge state.

How it informed the tool: Proposed a Turing-like evaluation framework for simulated learners. Emphasized that realistic communication patterns — not just correct knowledge modeling — are essential for passing interaction validity tests.

Working memory has limited capacity. When learners must hold too many steps in mind simultaneously, learning breaks down — not because they lack ability, but because the processing demands exceed capacity.

A learner's belief about their own capability directly predicts their effort, persistence, and how they interpret failure. Low self-efficacy learners may give up before trying or change correct answers under pressure.

Mastery-oriented learners seek deep understanding and persist through difficulty. Performance-oriented (grade-seeking) learners optimize for correct answers and may skip conceptual understanding entirely.

The ability to monitor one's own understanding — knowing what you know and what you don't — is a key predictor of learning outcomes. Metacognitively unaware learners don't experience confusion where they should.

Learners who are confident but unaware of their gaps won't seek help — they don't know they need it. This interaction between confidence and metacognitive awareness creates a distinct behavioral pattern in the simulation.

Intrinsic motivation arises from autonomy, competence, and relatedness. Disengaged or resistant learners aren't being difficult — they often lack one or more of these needs in their learning context.

How students communicate — their questions, hesitations, and responses to correction — is the primary channel through which tutors diagnose understanding. A learner's communication style directly determines what a tutor can observe, making it critical for realistic simulated interactions.