Phase 1 · LLM proposes, textbook order checks

Analytics Concept Network

Knowledge components extracted from a five-chapter interactive business-analytics textbook, typed as code, concept, or judgment; tagged to every section, cell, and exercise; and linked by proposed prerequisite edges. Click a node to see the passage it was extracted from and where the book teaches, assumes, or tests it.

Solid arrows = proposed prerequisite edges (thickness = prior); dotted = is-a. Node size = number of book units tagged. Drag to move, scroll to zoom.

Leap warnings

Unreviewed, rule-generated. A KC whose first tagged occurrence is assumes or exercises with no earlier teaches. A single missed tag manufactures one, so this is an upper bound on real defects. never taught in the book · taught later.

Proposed prerequisite edges

Prior = mean of textbook order, counterfactual judgment, and zero-shot scores (0–1). The last column is the book's own teaching order: supports (prerequisite taught first), reversed, or same unit — co-introduced, which neither supports nor contradicts the edge.

fromtopriororder

Heaviest exercises

Number of distinct KCs an exercise requires (Mihaylova 2026's KC-load metric).

unitkindsectionKCs

Instructor knowledge base

Hand-written edges are coarse (one concept spans 7–90 leaves). A finer leaf edge running the other way is a level mismatch, not a contradiction: both can be true.

fromtoleaf pairsoutcome

Method: ontology-guided tagging (Micheli 2025 Python ontology as the seed for chapters 0–2; LLM-proposed typed hierarchy for chapters 3–4), one annotation pass per unit by Claude, score ≥ 4 counts as present; roles teaches / assumes / exercises; KC load follows Mihaylova 2026; the teaches/assumes/exercises roles and the leap rule are this project's own. Edges carry a prior, not a verdict. Nothing here is validated against student data, and no expert has audited the tags. Coverage: 227/246 chapter 3–4 units tagged (92.3%); separately, all 127 chapter 3–4 vocabulary leaves were used at least once.