Prerequisite Knowledge Graphs: Finding the Root of a Gap

Tell a learner they're "weak at math" and you've described a feeling, not a problem you can solve. Decades of learning science suggest the label is usually wrong anyway. More often, a learner who struggles across many topics is missing one or two foundations that sit beneath all of them — and the way to see that is to stop looking at topics in isolation and start mapping how they depend on one another.

What is a prerequisite knowledge graph?

A prerequisite knowledge graph is a map of order. It records which skills have to be in place before another can be learned — fractions before ratios, ratios before proportional reasoning, and so on. Knowledge isn't a flat list of topics; it has a dependency structure, and that structure decides what a learner can pick up next. The idea was put on a formal footing by Jean-Paul Doignon and Jean-Claude Falmagne in 1985, whose knowledge space theory describes the states a learner can actually be in as constrained by these prerequisite relations.¹ The practical upshot is simple but powerful. Learning follows feasible paths, not arbitrary ones. If you know the graph, you know not just what a learner is missing, but what becomes reachable once a particular gap is closed — which turns a flat list of topics into a route through them.

Why does a learner who "looks weak at math" often have just one missing foundation?

Because math is a network, not a checklist. Skills sit on top of other skills, so a single weak foundation doesn't stay contained — it leaks upward into everything built on it. Take fractions. A learner who never solidified fractions will struggle with ratios, with proportion, with rates, and with parts of algebra and geometry, because each of those quietly depends on fractional reasoning. From the outside, that looks like a learner who is "bad at math" across the board. On the graph, it looks like one thing: a missing root, with a cluster of topics failing downstream of it. The grade flattens all of that into a single word. The graph tells a more accurate — and more hopeful — story: the learner is not globally weak; one foundation is missing, and it is holding up a whole branch above it. That distinction changes what you do next.

How do prerequisite graphs turn scattered mistakes into a pattern?

Looked at one by one, a learner's errors can seem random: a bad day on ratios, a stumble on proportion, a wrong turn in an algebra word problem. Three separate problems, three separate fixes — or so it appears. Place those same topics on a prerequisite graph and something clicks. They share a parent. The scattered failures line up beneath one upstream skill, and the pattern points straight at the root. This isn't only theory. Annalies Vuong and colleagues showed in 2011 that prerequisite relationships can be recovered from learner performance data itself, by comparing how learners who had a skill fared against those who didn't.³ The graph is what converts a pile of individual mistakes into a diagnosis. Instead of a list of weak topics, you get a single, actionable cause — and a clear first thing to fix.

What does the research say about learning hierarchies?

The pedagogical version of this idea is older still. In 1968, Robert Gagné described what he called learning hierarchies: complex capabilities decompose into subordinate skills arranged in an ordered relationship, where mastering the lower skills produces "positive transfer" to the higher ones.² Read the other way, the claim is stark. If a subordinate skill is missing, the skills above it cannot transfer — the foundation simply isn't there to build on. That is exactly why one gap can cap an entire branch of a subject rather than a single topic. Gagné's hierarchies and Doignon and Falmagne's knowledge spaces reach the same truth from different directions, one from instructional theory and one from mathematics. Both say learning has an order. And both imply the same practical rule: find the lowest missing skill, because everything above it is waiting on it.

Why does fixing the root change everything?

Here is the part that surprises people. The topics failing above a missing prerequisite usually aren't broken in their own right — they were starved. The learner couldn't get traction on ratios because the fractional reasoning underneath them wasn't there. Repair that root, and the dependent skills don't each need to be re-taught from scratch. Many of them become learnable almost immediately, because the thing that was blocking them is gone. This is why diagnosing to the root matters so much. Re-teaching the whole subject treats every symptom equally and wastes most of the effort. Fixing the one foundation beneath the cluster is a single, targeted intervention that unlocks the rest. The leverage was never spread evenly across the topics a learner failed. It was concentrated in the root — and that is the one place worth spending time first.

What does this mean for teaching?

For a teacher, the shift is from treating symptoms to finding the cause. A learner failing several topics doesn't need several separate rescue plans. They need the right first move — the foundational gap that, once closed, lets the rest fall into place. The hard part is seeing it, because the root often sits a layer or two below where the visible failures are. Mapping skills onto their prerequisites is what makes that root visible, turning "weak at math" into "missing fractions, which is why ratios, proportion and parts of algebra aren't landing." This is the principle CogniTrace is built around — not just flagging where a learner struggles, but tracing it to the foundation underneath and prescribing that as the next step. Fix the root, and you change the whole branch above it — with far less effort than re-teaching everything.