How Case-Based Reasoning Works | Retrieve, Reuse, Revise, Retain

eduKateSG Learning Node Series · 0144

Experts rarely solve every problem from first principles.

A doctor remembers a patient who looked similar. A lawyer remembers a precedent. An engineer recalls a failure mode from an earlier project. A teacher sees a misconception that appeared in another class. A student recognises that today’s unfamiliar question has the same skeleton as something solved before.

The old case does not give the new answer automatically. It gives a starting point.

Case-based reasoning works when prior experience is retrieved for a new problem, adapted to the differences, tested, revised and stored as a richer case for future use.

The 50-Second Read

Case-based reasoning uses remembered cases as resources for solving new problems.
The classic cycle is often summarised as Retrieve, Reuse, Revise and Retain.
Retrieval depends on noticing the right similarity. Surface similarity can mislead; deep structural similarity often matters more.
Reuse does not mean copying. The old solution may need adaptation.
Revision matters because even a strong precedent can fail when an important condition changes.
Retention turns the new experience into future memory.
Case libraries can help learners compare multiple examples instead of treating one worked solution as a universal template.
Failure cases can be particularly valuable because they show where an apparently reasonable solution breaks.
Expertise involves knowing not only which case resembles the current problem but which differences are consequential.
Good teaching makes the retrieve–adapt–test–reflect process visible so learners can build their own case memory rather than memorise one answer.

Canonical Owner Boundary

This Learning Node owns problem solving through retrieval, adaptation, revision and retention of prior cases. How Analogical Encoding Works owns comparison that reveals shared structure. What Is an Exemplar? owns remembered instances used for judgment. How Cognitive Flexibility Theory Works owns revisiting complex domains through multiple representations and cases. This page owns the full case-based reasoning loop: finding a prior case, deciding what transfers, revising for the differences and retaining the new experience.

1. Humans Think With Stories of Previous Problems

Rules are powerful because they compress many cases. But people often reason the other way too: they encounter a new situation and ask, consciously or not, “What does this remind me of?”

The remembered episode contains more than a rule. It contains context, decisions, errors, outcomes and sometimes the reasons a solution worked.

That richness can be useful when the new problem is messy enough that one abstract rule does not determine the answer.

2. Case-Based Reasoning Has a Classic Four-Step Cycle

Case-based reasoning research often uses the four-Re cycle associated with Aamodt and Plaza:

Retrieve: find one or more relevant prior cases.
Reuse: apply what appears transferable from the old case.
Revise: test and modify the proposed solution when the new situation differs.
Retain: store the new experience so future reasoning has a richer case base.

A detailed review in The Knowledge Engineering Review develops these four processes and the difficult design questions behind them.

3. Retrieval Is the First Intelligence Test

If the wrong case is retrieved, everything after it can be elegant and useless.

A student sees a geometry diagram that looks familiar and chooses the same theorem as last time. But the condition that made the theorem valid is absent. A doctor remembers a similar symptom cluster but misses the age difference that changes the risk profile. A teacher uses the same intervention because two students both appear “careless,” though one has a knowledge gap and the other a checking problem.

Good retrieval depends on similarity that matters.

4. Surface Similarity Is Seductive

Novices often retrieve by appearance.

Same keywords. Same picture. Same topic. Same numbers. Same chapter.

Experts are more likely to notice structural features: the relationship between quantities, the causal mechanism, the constraint pattern, the role of evidence or the type of uncertainty.

Learning improves when students are taught to ask not only “What looks similar?” but “What relationship is doing the work?”

5. Deep Similarity Enables Transfer

Two problems can involve different stories but the same structure.

A mixture problem and a finance problem may both be weighted-average problems. A literary inference and a historical source question may both require separating observation from interpretation. A physics equilibrium problem and an economics market-clearing model may both involve competing flows reaching a stable relation.

Case-based reasoning becomes educationally powerful when learners can retrieve across surface boundaries.

6. Reuse Is a Hypothesis, Not a Copy Command

The old case says, “This worked before.” It does not say, “This will work unchanged.”

Reuse should be treated as a proposed route. The learner must identify which features are shared and which differ.

Copying is case-based imitation. Adaptation is case-based reasoning.

7. The Revise Step Is Where Judgment Lives

Revision distinguishes a living reasoning system from a lookup table.

The current problem contains a changed constraint. The learner tests the borrowed solution. It partly works, fails or produces an unexpected consequence. The difference becomes information.

Revision asks: What exactly must change because this case is not the old one?

8. Retention Turns Experience Into Future Capability

A learner can solve a problem successfully and still waste the experience.

If the final answer is filed away without a record of what made the case distinctive, future retrieval may fail.

Retain the useful lesson: the problem structure, the chosen route, the condition that mattered, the failed attempt, the adaptation and the outcome.

That is far more valuable than storing the answer alone.

9. A Case Is More Than a Worked Example

A worked example usually demonstrates how to solve a target problem. A case contains broader context: what the problem looked like, what decisions were made, which constraints mattered, what went wrong and why the final route succeeded.

Case-based learning is therefore especially useful in domains where judgment depends on context rather than one algorithm.

10. Education Has Used Case-Based Reasoning Ideas Directly

Janet Kolodner, Michael Cox and Pedro González-Calero reviewed case-based reasoning-inspired approaches to education, noting links with constructivist designs such as Goal-Based Scenarios and Learning by Design. They describe case libraries as resources for learning and as frameworks for articulating understanding.

The core educational idea is not that students should memorise a library of stories. It is that previous experiences can become indexed resources for future reasoning.

11. Case Libraries Externalise Experience

Novices have a small internal case base.

A well-designed case library lets them borrow experience. They can inspect how multiple people handled similar but non-identical problems.

The library becomes more useful when cases are indexed by meaningful features rather than only topic labels.

12. One Case Can Teach the Wrong Rule

If a student sees one successful example, they may generalise the wrong feature.

They may think a persuasive essay always begins with a rhetorical question because the model did. They may think every quadratic equation should be factorised because the example factorised neatly. They may think a science investigation always needs the same apparatus because one lab used it.

Multiple cases reveal what varies and what stays invariant.

13. Contrast Improves Case Retrieval

Place two similar cases side by side and ask what changed.

This connects directly with How Contrasting Cases Work and How Analogical Encoding Works.

Comparison helps learners build better indexes for later retrieval because they notice which features are structural and which are incidental.

14. Failure Cases Are Not Inferior Cases

A successful case shows one route that worked. A failure case can reveal the boundary where a tempting route stops working.

In complex decision domains, that can be more valuable.

Ask: Why did the reasonable solution fail? Which feature was misread? What warning signal was present? What would have changed the decision earlier?

15. Near Misses Build Better Judgment

A near miss resembles the target case strongly but differs in one consequential way.

These cases train learners to resist superficial matching. The question becomes, “Which difference changes the rule?”

This is exactly the kind of judgment experts use when deciding whether precedent transfers.

16. Case-Based Reasoning in Mathematics

A student meets an unfamiliar rate problem.

Instead of searching memory for a formula by keyword, they retrieve a previous case with the same relational structure: two quantities changing at different rates under a shared constraint.

Then they ask what differs. Are the units different? Is the rate constant? Is the relationship additive or multiplicative? Does the previous method still apply?

The old case becomes a scaffold for modelling rather than a template for copying.

17. Case-Based Reasoning in English

Writers build case memory too.

A student remembers how a previous introduction established stakes quickly. The new task has a different audience and purpose, so the phrasing cannot be copied. But the function can be adapted.

Strong writing education therefore exposes students to multiple cases and asks what rhetorical job each move performs.

18. Case-Based Reasoning in Comprehension

A reader encounters an unreliable narrator.

Prior reading cases may help: contradiction between narration and action, selective reporting, exaggerated self-justification, information other characters know but the narrator ignores.

The learner does not ask, “Have I seen these exact words?” They ask, “Have I seen this pattern of evidence?”

19. Case-Based Reasoning in Science

Science students can use cases to reason about experimental design.

A previous experiment suffered from a confounding variable. A new investigation looks different but contains the same design problem: two factors changed at once.

The earlier case becomes a warning pattern.

20. Case-Based Reasoning in Medicine

Clinical expertise is deeply case-rich, but medical reasoning also shows the danger of case retrieval.

A vivid previous patient can bias diagnosis if similarity is superficial. Good clinicians use cases alongside base rates, mechanisms, tests and explicit differential diagnosis.

The lesson transfers to education: remembered cases are evidence, not authority.

21. Case-Based Reasoning in Law

Legal reasoning makes the structure visible because precedent must be compared with the present facts.

Which features are materially similar? Which differences are legally consequential? Does the principle transfer? Should the case be distinguished?

That is case-based reasoning under formal argumentative discipline.

22. Case-Based Reasoning in Engineering

Engineering organisations accumulate failure cases: fatigue cracks, interface mismatches, water ingress, tolerance problems, cascading control errors.

The value lies not only in remembering that the failure happened but in indexing the conditions that made it possible.

A new design can then retrieve the warning before the failure repeats.

23. Case-Based Reasoning in Teaching

Teachers build enormous case libraries over a career.

“This student is making an error that resembles the denominator problem I saw last year.” “This class discussion is stalling in the same way as the previous one.” “This intervention worked for one learner but only because the prerequisite knowledge was already strong.”

Professional growth improves when teachers articulate these cases rather than leaving them as intuition only.

24. Reflection Improves Retention

After solving a difficult problem, ask:

What type of case was this?
Which feature mattered most?
What did I initially retrieve?
Where did the analogy fail?
What adaptation solved the mismatch?
What warning sign should trigger this memory next time?

This creates a better index for future retrieval.

25. Retrieval Practice Can Include Cases

Students usually retrieve facts and formulas. They can retrieve cases too.

“Name a previous problem where the same structural decision appeared.” “Which earlier essay had the same audience problem?” “Which experiment taught us why that control variable matters?”

Case retrieval becomes part of studying rather than something that happens accidentally during exams.

26. Case Notes Should Store the Reason, Not Only the Result

A useful case note has at least five elements:

the problem context;
the important features;
the chosen solution;
why it worked or failed;
the lesson that transfers.

Without the “why,” the memory is much harder to adapt safely.

27. Too Many Cases Can Create Retrieval Noise

A larger case library is not automatically better.

If cases are poorly indexed, the learner remembers many examples but cannot find the relevant one. If cases are redundant, important distinctions disappear in repetition.

Curate cases to maximise coverage of meaningful variation.

28. Experts Sometimes Need to Ignore the Most Similar Case

The nearest case can be wrong because one small difference changes the mechanism.

Expertise includes knowing when similarity should be overridden by a rule, constraint, base rate or causal model.

Case-based reasoning should therefore cooperate with rule-based and model-based reasoning, not replace them.

29. AI Makes Case Retrieval Cheap

Modern retrieval systems can search enormous case libraries and place relevant examples beside a new problem.

Recent work combining retrieval-augmented generation with case-based reasoning shows renewed interest in this architecture for AI systems.

The educational opportunity is obvious: students can retrieve analogous examples quickly. The risk is equally obvious: if the system performs retrieval and adaptation invisibly, the learner may receive a good answer without developing the similarity judgment themselves.

30. Ask AI for Cases, Then Make the Learner Compare

A better workflow is:

retrieve two or three relevant prior cases;
ask the learner to rank their similarity;
identify one feature that transfers and one that does not;
adapt the chosen solution;
test the adaptation;
write the new case note.

The tool expands the case base while the learner keeps the reasoning.

31. Cross-Domain Comparison: Maintenance Engineering

A technician sees an unusual vibration pattern and remembers a previous machine with a similar signature.

The earlier repair is not copied blindly because the machine configuration differs. The case narrows the search, suggests checks and offers a possible route. Testing decides whether the analogy survives.

This is case-based reasoning in practical form.

32. Cross-Domain Comparison: Emergency Response

Emergency organisations study prior incidents because real crises contain uncertainty that cannot be fully scripted.

But the purpose of after-action review is not to reproduce yesterday’s response. It is to build a library of patterns, warnings and adaptations for tomorrow’s different incident.

33. Cross-Domain Comparison: Chess

Strong chess players remember enormous numbers of structured positions. They do not calculate every possibility from nothing.

But a familiar pattern remains conditional on the exact position. One changed piece can make the remembered plan invalid.

Pattern memory accelerates reasoning; position-specific verification protects against false transfer.

34. A Practical Case-Based Learning Protocol

Represent the new case: state the problem and important features clearly.
Retrieve: find prior cases with potentially relevant structure.
Compare: identify similarities and differences before choosing a route.
Reuse carefully: borrow the principle or solution components that still fit.
Adapt: change the route to reflect consequential differences.
Test: look for evidence that the adapted solution actually works.
Revise: repair the route when the new case resists the old one.
Reflect: identify what this case taught that the previous case did not.
Retain: store the new case with useful retrieval cues.
Generalise cautiously: extract rules only when multiple cases support them.

35. Failure Mode: Copying the Previous Answer

The learner recognises the topic and reproduces the old method without checking conditions.

Repair: require a difference scan before reuse. “What changed, and which change could invalidate the previous route?”

36. Failure Mode: Retrieving by Keyword

Two problems share vocabulary but not structure.

Repair: index cases by relationships, constraints and mechanisms as well as topic words.

37. Failure Mode: Storing Only Successful Cases

The learner’s memory contains polished solutions but no map of where tempting approaches break.

Repair: retain selected failure and near-miss cases with explicit lessons.

38. Failure Mode: Case Hoarding

Students collect hundreds of examples and rarely compare them.

Repair: curate for variation. A smaller set of well-contrasted cases may build better indexes than a giant archive.

39. Failure Mode: No Retain Step

The learner solves a novel problem, celebrates and moves on.

Repair: create a short post-solution record of what made the case distinctive and what should trigger retrieval later.

40. The Missing-Node Scan

If students can solve familiar exercises but freeze when the story changes, if they copy methods based on keywords, if worked examples are remembered as scripts rather than cases, if past mistakes keep repeating because nobody stores the lesson, or if learners have seen many examples but cannot identify which previous one is relevant now, the missing node may be case-based reasoning.

Look for the gap between experience and reusable experience.

Education already produces thousands of cases. The question is whether learners index, compare, adapt and retain them well enough to make the next problem easier.

41. The Return Path

A learner meets a problem they have never seen before.

At first it looks new.

Then a prior case appears in memory—not because the wording matches, but because the underlying relationship does.

The learner retrieves it, borrows the useful part, notices the difference, changes the route, tests the result and remembers what happened.

The new problem has become part of the learner’s future intelligence.

Case-based reasoning works when experience becomes a living library rather than a museum: retrieve what resembles the present, adapt what still applies, revise what does not, and retain the new case so tomorrow begins with more usable memory than today.

Research and Further Reading

eduKateSG Learning Node Series · 0144 · Previous: 0143 — How Epistemic Emotions Work.

How Case-Based Reasoning Works | Solve New Problems by Retrieving, Comparing and Revising Old Cases