eduKateSG Learning Node Series · 0116
A student gets a mathematics question wrong.
What exactly did they fail to know?
“Algebra” is too large. “Simultaneous equations” may still be too large. The first broken piece could be substitution, distributive expansion, preserving equality, interpreting variables, arithmetic with negatives, or choosing which equation to rearrange.
Learning science needs a way to talk about those smaller pieces without pretending that complex performance is nothing more than a pile of fragments.
One useful idea is the knowledge component.
A knowledge component is a smaller unit of cognitive capability inferred from performance across related tasks.
In the Knowledge–Learning–Instruction framework developed by Kenneth Koedinger, Albert Corbett and Charles Perfetti, knowledge components can include concepts, principles, facts, skills, schemas, production rules, facets and even misconceptions. The point is not to force every kind of learning into one identical box. The point is to reason at a level small enough to connect what is known, how it changes, and what instruction should do next.
The 50-Second Read
- A complex task usually depends on several smaller cognitive capabilities.
- Knowledge components are inferred units such as facts, rules, concepts, skills, schemas or misconceptions that help explain performance across related tasks.
- The same question can require multiple knowledge components at once.
- The same knowledge component can appear across many different questions.
- Mapping tasks to knowledge components can improve diagnosis, practice design and adaptive learning.
- A weak component can create errors across apparently unrelated questions.
- Overly coarse components hide the first weak link; overly fine components create an unusable map.
- Knowledge tracing often depends on a model of which components each item exercises.
- Task analysis and knowledge-component modelling are related but different: one decomposes observable performance; the other models the cognitive capabilities underlying that performance.
- The system must eventually reconnect components into whole, fluent performance. Decomposition is a diagnostic and design tool, not the final form of expertise.
Canonical Owner Boundary
This Learning Node owns the cognitive decomposition of learning into smaller knowledge units inferred from related task performance. How Task Analysis Works owns decomposition of complex performance into observable steps, decisions and subskills for instruction. How Knowledge Tracing Works owns estimation of learner knowledge from sequences of attempts. How Studying Works | Learning Granularity owns how the size of the learning unit changes understanding, practice and measurement. How Learning Dependencies Work owns prerequisite order. This page owns the hidden middle layer: what cognitive pieces a task is actually exercising.
1. Subjects Are Too Large for Diagnosis
“Weak at mathematics” is not an actionable diagnosis.
Neither is “bad at English,” “doesn’t understand science,” or “struggles with comprehension.”
These labels identify a territory, not the mechanism of failure.
Instruction becomes more precise when the territory is decomposed into capabilities that can be observed indirectly through tasks.
2. The KLI Framework
Koedinger, Corbett and Perfetti’s 2012 Knowledge–Learning–Instruction framework was designed to connect three levels that education often studies separately:
- Knowledge: what cognitive components are being learned?
- Learning: what mental processes change those components?
- Instruction: what experiences or designs trigger those learning processes?
The framework defines a knowledge component broadly as an acquired unit of cognitive function or structure that can be inferred from performance on related tasks.
That broad definition is deliberate. Different domains require different kinds of knowledge units.
3. A Knowledge Component Is Not Necessarily a Fact
Some components are declarative facts: the chemical symbol for sodium is Na.
Others are procedural: distribute multiplication across addition.
Others are conceptual: understand that equality represents a relationship, not an instruction to compute.
Others can be schemas, principles, perceptual discriminations or stable misconceptions.
The useful level depends on the instructional question.
4. One Task Can Require Many Components
Consider a word problem requiring simultaneous equations.
The learner may need to:
- identify the unknown quantities;
- translate verbal relationships into equations;
- preserve equality;
- choose substitution or elimination;
- expand brackets;
- operate with signed numbers;
- solve a linear equation;
- substitute back;
- interpret the numerical answer in context;
- and check whether the solution satisfies both original relationships.
A final wrong answer does not tell us which component failed.
5. One Component Can Appear Across Many Tasks
The reverse is equally important.
Understanding equivalence appears in algebra, fractions, ratios, unit conversion and formula manipulation. Evidence-to-claim reasoning appears in science, history, English and geography. Proportional reasoning appears across percentages, rates, scale drawings, chemistry concentrations and physics relationships.
A knowledge component can therefore explain a pattern of errors that cuts across textbook chapters.
6. The Item Is Not the Skill
A common educational mistake is to treat each question as a separate capability.
“Student got Question 7 wrong.”
But Question 7 is an observation surface. Underneath it may sit several reusable components.
The better diagnostic question is: which component does this item provide evidence about?
7. Knowledge Components Create a Q-Matrix-Like View
In educational modelling, tasks can be mapped to the knowledge components they require.
Imagine a table where rows are questions and columns are components. A mark indicates that a question requires that component.
Now repeated attempts can provide evidence about individual components rather than only total scores.
This idea underlies many forms of cognitive diagnosis and adaptive learning, though the exact models differ.
8. The Map Is a Hypothesis
A knowledge-component map is not reality carved into stone.
It is a model of what cognitive pieces appear to explain performance.
If students who supposedly know a component keep failing certain item types, the map may be wrong. Perhaps those items require an additional distinction. Perhaps one component should be split into two. Perhaps two components should be merged.
Good learning systems revise the model when the data do not fit.
9. Too Coarse: “Fractions”
Suppose the component is simply labelled “fractions.”
A learner gets two fraction questions right and one wrong. What should the teacher do?
The component is too coarse to tell us whether the issue is equivalence, magnitude, operation choice, common denominators, whole identification or interpretation.
Coarse components are easy to manage but weak for diagnosis.
10. Too Fine: Every Micro-Move Becomes Its Own Component
The opposite failure creates hundreds of tiny labels for every possible action.
The model becomes impossible to interpret, impossible to teach against and too sparse for reliable estimation.
The useful granularity is the level at which different instructional responses become meaningful.
11. The Instructional Test for Granularity
Ask: if two learners differ on this component, would I teach them differently?
If yes, the distinction may be instructionally useful.
If no, the decomposition may be too fine for the current purpose.
This keeps the map tied to decisions rather than taxonomy for its own sake.
12. Misconceptions Can Be Modelled as Components Too
Not all cognitive structure is correct knowledge.
A learner may have a stable rule such as “a larger denominator means a larger fraction” or “multiplying always makes a number bigger.”
These are not empty spaces. They are active models producing predictable answers.
Treating misconceptions as structured components can improve diagnosis because the repair is different from initial teaching.
13. Mathematics: Equality as a Knowledge Component
Many students learn the equals sign as “the answer comes next.”
That interpretation may work for 3 + 4 = 7 and then fail for 3 + 4 = __ + 5 or algebraic manipulation.
Relational understanding of equality is a component that supports later algebraic learning across many tasks.
A knowledge-component view explains why an early conceptual weakness can echo years later.
14. Mathematics: Method Selection Is Its Own Component
A student may execute factorisation perfectly when told to factorise but fail to recognise when factorisation is the right method inside a mixed problem set.
Execution and selection are different capabilities.
Interleaving exposes this because the learner must choose the method. A knowledge-component model should represent the decision, not only the procedure.
15. English: Inference Is Not One Skill
“Inference” may look like one component, but classroom evidence can reveal subcomponents.
A learner may infer emotion from action but struggle to infer motive from conflicting evidence. Another may make plausible inferences but fail to anchor them in text.
The right decomposition depends on the patterns we need to explain and teach.
16. English: Writing Quality Contains Reusable Components
Argument writing may require claim control, evidence selection, explanation, counterargument, paragraph cohesion and audience calibration.
A weak essay can therefore contain a strong component profile in some dimensions and weak components in others.
This explains why “write more essays” can be inefficient. The learner may need targeted work on one reusable component followed by reintegration into full writing.
17. Science: Mechanism Versus Vocabulary
A learner may correctly name “diffusion” but fail to explain particle movement down a concentration gradient.
Word recognition and mechanism understanding are different components.
Assessment that scores only the term can overestimate the learner’s scientific model.
18. Science: Evidence Evaluation Is Cross-Cutting
Evaluating whether evidence supports a claim appears across physics, biology, chemistry and practical work.
A student may know content facts while lacking this cross-cutting component.
The component view can therefore reveal skills that textbook chapters hide because they recur across many topics.
19. Knowledge Components and Learning Progressions
Knowledge components describe pieces. Learning progressions describe states and routes across development.
The two can connect. A progression may show how components accumulate, reorganise or become coordinated as expertise grows.
For example, arithmetic components may later become subroutines inside algebraic reasoning rather than disappearing.
20. Knowledge Components and Prerequisites
Some components depend on others.
Solving quadratic equations by factorisation requires factorisation knowledge. Interpreting a velocity–time graph requires graph reading plus a concept of velocity and change over time.
A dependency graph built over components can make the first weak link visible.
21. Knowledge Components and Knowledge Tracing
Knowledge tracing estimates how likely a learner is to know something based on sequences of attempts.
But the model first needs to know what each attempt is evidence about.
That is where knowledge-component mapping enters. If an item requires components A and B, success or failure updates beliefs about A and B—depending on the tracing model and assumptions.
A poor component map can therefore make a sophisticated tracing algorithm confidently wrong.
22. The Model–Data Feedback Loop
Suppose two items are mapped to the same component but learner performance on them behaves very differently.
Maybe one item contains an additional representation demand. Maybe wording creates language load. Maybe the component should be split.
Data can therefore improve the cognitive model.
The relationship is reciprocal: theory proposes components; learner data tests whether that decomposition explains performance.
23. Knowledge Components and Practice Design
A component map helps answer a practical question: what should the next practice item exercise?
If the learner is secure on equation solving but weak on translating verbal relationships, another page of naked equations is low-value practice.
Practice should target the component with the highest expected learning return while still preserving integration with the full task.
24. Isolate, Repair, Reintegrate
Component-level teaching can become fragmented if students never return to the whole performance.
A useful loop is:
Whole task → diagnose component → isolate and repair → reconnect to whole task → test transfer.
The component is a repair target, not the final educational product.
25. Components Can Become Automatic Subroutines
As expertise grows, smaller components become fluent and consume less conscious attention.
A fluent reader does not consciously decode every familiar word. An experienced algebra student does not deliberate over each integer operation. A skilled writer can manage sentence structure while thinking about argument.
Automation frees cognitive capacity for higher-level coordination.
26. But Expertise Also Reorganises Knowledge
Experts do not simply have more isolated components.
They chunk, coordinate and reorganise knowledge into larger structures. What was once a sequence of separate steps becomes one meaningful pattern.
A component model must therefore allow the possibility that the useful unit changes with expertise.
27. The Same Component Can Be Representation-Specific
A learner may understand a relationship in an equation but not on a graph.
Should that count as one component or two?
The answer depends on evidence. If performance transfers reliably between representations, one broader component may work. If it does not, the model may need representation-specific components plus a translation component.
This illustrates why decomposition is empirical, not purely philosophical.
28. Language Can Be a Hidden Component
A student may understand the mathematics but misread “at least,” “no more than,” “respectively” or “in terms of.”
If language demand is not represented in the model, the system may diagnose a mathematical weakness that is not there.
Knowledge-component maps should include relevant language and representation requirements when they genuinely affect performance.
29. Motivation Is Not a Knowledge Component
A learner can fail an item despite knowing the required components because of inattention, fatigue, anxiety or low effort.
This reminds us that performance is not a pure readout of knowledge.
Knowledge-component models explain part of the system. They should not swallow every cause of performance into “knowledge.”
30. Components and the First Weak Link
The first weak link is often a component that fails early enough to contaminate everything downstream.
If a student cannot identify the correct base quantity in percentage change, later arithmetic correction will not solve the repeated problem. If a reader cannot resolve pronoun reference, later inference may be built on the wrong entity.
Component mapping helps locate the earliest consequential break.
31. Component Maps Can Improve Feedback
“You got five questions wrong” is weak feedback.
“Four of the five errors occurred when the problem required identifying the base quantity before calculating percentage change” is stronger.
The second message points to a reusable component and therefore a transferable repair.
32. Component Maps Can Improve Assessment Sampling
An assessment may contain twenty questions but accidentally sample the same component repeatedly.
A component matrix can reveal coverage gaps. Perhaps the test heavily samples calculation but barely samples representation, method selection or explanation.
Assessment quality improves when the task set reflects the intended capability map rather than question count alone.
33. Component Maps Can Improve Curriculum Coherence
If a component appears in Primary 6, Secondary 1 and Secondary 3 under different topic names, a curriculum map can make the continuity visible.
Teachers can then deliberately revisit, extend and transfer the same underlying capability rather than teaching each chapter as a disconnected island.
This is especially valuable for cross-cutting reasoning skills.
34. Cross-Domain Comparison: Engineering Fault Trees
When a complex engineering system fails, investigators do not stop at “the machine failed.” They decompose the system into components and pathways until the failure mechanism becomes actionable.
Learning diagnosis uses similar logic. The final wrong answer is the system-level failure. Knowledge components help trace possible internal causes.
35. Cross-Domain Comparison: Software Dependencies
A software application may fail because one dependency is missing or incompatible. Reinstalling the entire system is inefficient if the broken library can be identified.
Education often does the equivalent of reinstalling: reteach the whole chapter because one component is unstable.
Component diagnosis makes repair more surgical.
36. Cross-Domain Comparison: Sports Technique
A tennis serve is one fluid performance, but coaching can isolate grip, toss, shoulder rotation, contact point and follow-through.
The athlete does not compete by performing five separate drills. The decomposition exists to improve the integrated action.
Knowledge components should serve the same relationship between diagnosis and whole performance.
37. A Practical Knowledge-Component Mapping Protocol
- Start with a real performance: choose the task learners must eventually do.
- List recurring decisions and knowledge: what must the learner know or recognise?
- Separate components that produce different error patterns: do not hide useful distinctions.
- Merge distinctions that do not change instruction: avoid taxonomy overload.
- Map tasks to components: identify which items provide evidence about which capabilities.
- Collect learner data: look for patterns across attempts.
- Challenge the model: do the predicted component patterns actually appear?
- Repair targeted components: choose focused instruction or practice.
- Reintegrate: return to the whole performance.
- Test transfer: verify that the repaired component works in new contexts.
- Revise the map: treat the model as provisional.
38. Failure Mode: Topic Labels Pretend to Be Components
The map says “algebra,” “geometry,” “comprehension” and “writing.”
Repair: decompose until the labels correspond to capabilities that can explain error patterns and guide instruction.
39. Failure Mode: Every Step Becomes a Component
The map grows into hundreds of micro-labels that no teacher can use.
Repair: merge components that do not lead to meaningfully different instructional decisions.
40. Failure Mode: The Component Map Is Never Validated
Experts design a beautiful taxonomy and assume it matches learner cognition.
Repair: compare the model against actual performance patterns. Use errors, response times, transfer and learning curves as evidence.
41. Failure Mode: Isolation Without Reintegration
Students become good at component drills but still fail the real task because coordination is missing.
Repair: return components to complete problems, essays, experiments and performances as early as the learner can manage.
42. Failure Mode: A Knowledge Problem Is Diagnosed When the Cause Is Elsewhere
Fatigue, anxiety, motivation, language or interface problems produce failure that the model interprets as missing knowledge.
Repair: use multiple evidence sources and maintain categories outside the knowledge model for state and context factors.
43. Failure Mode: The Map Becomes the Curriculum
Because components can be measured, the school teaches only what fits the component model.
Repair: preserve complex outcomes such as creativity, judgment, synthesis and whole-task fluency even when they resist neat decomposition.
44. The Missing-Node Scan
If learners repeatedly fail different-looking questions for the same hidden reason, if teachers reteach whole chapters because the precise weak link is unclear, if adaptive systems know which item was wrong but not what cognitive capability it represents, or if practice is organised by textbook page rather than underlying skill, the missing node may be a knowledge-component model.
Look for these signals: broad labels such as “careless,” “weak at algebra” or “poor comprehension”; repeated errors crossing chapter boundaries; students fluent at procedures but unable to choose them; assessments with many items that all measure the same narrow skill; and interventions that improve the practised worksheet but not related tasks.
The hidden question is always the same: what smaller capability would explain this pattern of performance?
45. The Return Path
Return to the student who got the simultaneous-equations problem wrong.
One system writes “wrong” and moves to the next question.
A better system asks what the question required, which knowledge components were involved, which evidence from other attempts supports each hypothesis, and where the earliest consequential break occurred.
Then it repairs that component.
Then it returns the learner to the whole problem.
The decomposition has done its job when it disappears back into fluent performance.
Knowledge components work when they make learning small enough to diagnose without making education so small that the whole performance disappears.
Research and Further Reading
- Koedinger, Corbett & Perfetti (2012) — The Knowledge–Learning–Instruction Framework
- CID: A Framework for the Cognitive Analysis of Composite Instructional Designs
- How Knowledge Tracing Works
- How Task Analysis Works
- How Learning Dependencies Work
eduKateSG Learning Node Series · 0116 · Previous: 0115 — How Pre-Assessment Works.