eduKateSG Mathematics + Civilisation Research
Research checked: 10 August 2026
Evidence status: Strong classroom causal evidence in the studied setting; wider-scale replication still in progress.
Quick Read
Interleaving is not simply “mixing questions.”
It changes what the learner has to do.
In blocked practice, the page often tells the student which method to use:
quadratic question → use quadratic method
differentiation question → differentiate
simultaneous equations → solve simultaneous equations
Once the question type is already known, much of the selection problem has disappeared.
Interleaving restores it.
The student sees different problem families together and must decide:
What kind of problem is this?
Which mathematical structure matters?
Which method should I retrieve?
What is the first valid move?
That distinction matters because examinations, unfamiliar problems and real mathematical work rarely announce the correct method in advance.
A major classroom cluster-randomised trial involving 787 Grade 7 students found substantially better performance on an unannounced delayed test after a higher dose of interleaved practice than after mostly blocked practice: 61% versus 38%, d = 0.83. The groups received the same practice problems; their scheduling differed. (ies.ed.gov)
But this does not mean:
- teach every new topic through random mixed questions;
- eliminate blocked practice;
- mix problems before the learner has acquired the relevant methods;
- assume the exact 61% versus 38% result transfers to Singapore or Additional Mathematics;
- confuse difficulty during practice with failure to learn.
The better architecture is:
learn the method → stabilise the method → distinguish it from neighbouring methods → interleave → delay → test selection independently
That is the important result.
Direct Answer
Interleaving is valuable in mathematics because knowing a method and knowing when to use that method are different capabilities.
Blocked practice can help a learner acquire and stabilise a procedure.
Interleaved practice adds another requirement:
select the procedure from competing possibilities.
The strongest interpretation is therefore not:
interleaving is always better than blocked practice.
It is:
blocked and interleaved practice can perform different jobs at different stages of learning.
That distinction fits the wider MathematicsOS architecture.
A learner may possess several procedures yet still fail because the correct procedure is not selected when the problem changes form. MATHCIV-016 therefore owns the route-selection problem. It should not duplicate MATHCIV-014’s retrieval-practice article or MATHCIV-015’s spacing article.
Three Ways to Read This Article
For a student
If you can do a chapter exercise but become unsure when several topics appear together, the problem may no longer be the individual procedure.
The weakness may be:
recognition → discrimination → route selection.
For a parent
A student completing repetitive worksheets accurately may indeed be improving.
But those worksheets do not necessarily reveal whether the student can independently decide which mathematics to usewhen the labels disappear.
For a teacher or tutor
Interleaving is not a demand to make learning chaotic.
It is a controlled way of removing an instructional cue:
“Use the method we just taught.”
Once that cue disappears, method selection becomes observable.
1. The Hidden Help Inside a Blocked Worksheet
Consider this practice sequence:
Factorise:
- (x^2+7x+12)
- (x^2+9x+20)
- (x^2-5x+6)
- (x^2+2x-15)
The learner does useful mathematical work.
But the heading has already solved one problem:
What operation should I attempt?
The student knows every question is a factorisation question.
Now compare a mixed set:
- factorise an expression;
- solve a quadratic equation;
- determine a gradient;
- simplify an algebraic fraction;
- solve an inequality;
- find an equation of a line.
The calculation may not be harder.
The control problem is harder.
Before executing a method, the learner must identify the relevant structure.
That is a different capability.
2. Mathematics Contains at Least Two Problems
For many school tasks, the learner faces:
[
\text{Problem A: execute a method}
]
and
[
\text{Problem B: decide which method should be executed}
]
Blocked practice heavily trains Problem A.
Interleaving can make Problem B visible.
This is why a learner may appear fluent during chapter practice and then deteriorate on a mixed examination paper.
The student has not necessarily “forgotten everything.”
The system may instead be failing at:
[
\text{Question}
\rightarrow
\text{Structural recognition}
\rightarrow
\text{Method selection}
\rightarrow
\text{Execution}
]
If the second or third stage fails, perfectly adequate procedures downstream may never be activated.
That is a routing failure rather than simply a missing-knowledge failure.
3. What the Major Classroom Trial Actually Tested
The strongest school-mathematics evidence in the current MATHCIV evidence ledger comes from Rohrer, Dedrick, Hartwig and Cheung’s preregistered classroom trial.
The study involved 787 Grade 7 students in 54 classes taught by 14 teachers across five public schools in one Florida school district. Classes were randomly assigned to a higher-interleaving condition or a mostly blocked comparison condition. Students worked through the intervention for about four months. Importantly, the two conditions used the same practice problems; the principal manipulation was how those problems were scheduled. (ies.ed.gov)
Both groups later completed an interleaved review.
One month afterwards, students received an unannounced test.
The reported averages were:
[
61% \quad \text{interleaved}
]
versus
[
38% \quad \text{mostly blocked}
]
with a reported effect size of:
[
d=0.83
]
The study was subsequently recorded by the What Works Clearinghouse as meeting its evidence standards without reservations. (ERIC)
That is strong causal evidence for this intervention, population, curriculum context and outcome.
It is not a universal coefficient for mathematics education.
4. The Most Important Detail Is Not 61 Versus 38
The percentage difference attracts attention.
The design is more important.
Students were not simply given “harder worksheets.”
The practice schedule repeatedly created situations in which the strategy used for one question could not simply be copied into the next.
That means a learner repeatedly encountered something close to:
[
\text{Inspect}
\rightarrow
\text{Classify}
\rightarrow
\text{Select}
\rightarrow
\text{Execute}
]
instead of:
[
\text{Repeat previous procedure}
\rightarrow
\text{Execute}
]
The educational significance is therefore broader than rearranging worksheets.
Interleaving changes what information the learner must generate internally.
5. But Interleaving Is Not One Pure Mechanism
This is an important upgrade.
The large RCT does not isolate a single psychological mechanism.
The authors discuss several plausible contributors.
1. Discrimination and method selection
When unlike problems occur close together, learners repeatedly have to distinguish between problem types and decide which strategy applies.
2. Spacing
Interleaving naturally moves repeated encounters with one problem family farther apart.
So some of the benefit may arise because the learner revisits a method after a delay rather than repeating it immediately.
3. Retrieval
Because the relevant method is no longer continuously active from the previous question, the learner may need to retrieve it again.
The original paper explicitly discusses all three possibilities. (gwern.net)
Therefore MATHCIV-016 should not claim:
“The RCT proves discrimination alone caused the gain.”
The safe conclusion is:
Interleaved practice changes the learning environment in several useful ways, one of which is repeated strategy selection.
Spacing belongs primarily to MATHCIV-015.
Retrieval belongs primarily to MATHCIV-014.
Selection and discrimination belong here.
6. The Critical Boundary: Do Not Delete Blocked Practice
This is where the 2026 upgrade matters most.
It is easy to turn a useful finding into a bad rule:
Interleaving worked, therefore blocking is bad.
The original research does not justify that conclusion.
On the paper’s caveats page, the authors explicitly note that interleaving may be ineffective or excessively difficult if students have not first received at least some blocked practice when encountering a new concept. They also note that the interleaved students in their trial probably received some blocked practice during ordinary instruction before the experimental worksheets.
Their conclusion is unusually useful:
the evidence does not suggest students should entirely avoid blocked practice.
That gives MathematicsOS a clearer sequence.
Acquisition
Learn what the method is.
Stabilisation
Execute it correctly enough that elementary mechanics do not consume the whole task.
Discrimination
Compare it with nearby alternatives.
Interleaving
Remove the chapter cue and require method selection.
Transfer
Alter representation, wording, parameters or context.
Verification
Test after delay without tutor, answer key or AI.
Interleaving therefore belongs inside a learning corridor, not above every other instructional design.
7. Productive Difficulty Is Not the Same as Useful Difficulty
Interleaved practice can feel worse.
That should not surprise us.
The original trial’s teachers generally reported that students found interleaved assignments somewhat harder and more time-consuming than blocked ones.
A 2025 study then examined students’ own perceptions directly.
In one survey, 174 Grade 7 mathematics students evaluated spacing and interleaving. A second study surveyed 233 Grade 7 students about interleaved practice. Students generally rated interleaving as less effective than blocked practice and also as less preferable, harder and more time-consuming. (ResearchGate)
This creates a useful distinction:
[
\text{felt fluency}
\neq
\text{durable capability}
]
Blocked practice often makes the next move easier to predict.
That ease may be genuinely useful during acquisition.
But it can also make performance feel more secure than later independent performance warrants.
Interleaving removes some of that predictability.
The resulting difficulty can therefore be diagnostic.
But difficulty by itself proves nothing.
A badly designed worksheet can also be difficult.
The question is whether the added difficulty trains a capability we actually need.
8. The eduKateSG Upgrade: Interleave Confusable Decisions
The research supports interleaving.
MathematicsOS adds a more precise design question:
What should be interleaved with what?
Randomly mixing unrelated questions is a weak interpretation.
A stronger design is to construct contrast sets around decisions students actually confuse.
For example:
Algebra
- expand;
- factorise;
- simplify;
- solve.
These operations may involve similar symbols while requiring different goals.
Quadratics
- factorise a quadratic expression;
- solve a quadratic equation;
- find roots from a graph;
- complete the square;
- determine a discriminant.
The visible (x^2) is not enough to determine the route.
Trigonometry
- identity manipulation;
- equation solving;
- exact-value reasoning;
- triangle calculation.
The presence of (\sin) or (\cos) does not tell the student what the problem is asking.
Calculus
- differentiate;
- integrate;
- find a stationary point;
- reconstruct a function;
- calculate an area;
- interpret a rate.
The learner must classify the relationship before operating on it.
This gives us:
[
\text{Interleaving value}
\approx
\text{quality of discriminations being trained}
]
That expression is architecture, not an empirically fitted law.
Its purpose is to remind us that “mixed” is not enough.
The mixture must make a useful decision visible.
9. A New Diagnostic: The First-Move Test
A tutor does not always need a full solution to detect route-selection weakness.
Give the student ten mixed questions.
Before calculating anything, ask only:
- What is this question testing?
- What information matters?
- What is your first move?
- Why that move rather than the nearest alternative?
Then stop.
This separates:
[
\text{selection failure}
]
from
[
\text{execution failure}
]
A learner who chooses the correct method but later makes an algebraic error has a different bottleneck from one who cannot decide how to begin.
A single mark can hide that distinction.
The first move exposes it.
10. MathematicsOS: The Interleaving Protocol
A controlled implementation can use six stages.
Stage 1 — Acquire
Teach the new relation or procedure clearly.
Use correct models, worked examples and guided completion where appropriate.
Stage 2 — Stabilise
Give enough same-family work to determine whether the method itself is usable.
Do not infer stable capability from one successful example.
Stage 3 — Contrast
Introduce one or two nearby alternatives.
Ask:
What tells us these questions require different methods?
Stage 4 — Interleave
Mix sufficiently established problem families.
Remove headings that disclose the procedure.
Do not make the mix random for its own sake.
Stage 5 — Delay
Return later.
A delayed mixed probe is more informative than immediate repetition.
Stage 6 — Verify
Require:
- correct method selection;
- accurate execution;
- explanation of the decision;
- altered surface form;
- independent use;
- recovery after an error.
That produces a stronger capability claim than “completed mixed worksheet.”
11. What Should We Measure?
If the purpose is method selection, accuracy alone is insufficient.
A useful interleaving record should include:
Selection accuracy
Did the learner choose an appropriate method?
First-move latency
How long before a valid route was initiated?
Route explanation
Could the learner state why that route applied?
Execution accuracy
Once selected, could the method actually be carried through?
Confusion pair
Which alternative method repeatedly captured the learner?
Delayed performance
Did the route remain available later?
Changed-form performance
Did selection survive different wording or representation?
Independence
Could this be done without tutor cues, chapter labels, answer keys or AI?
Now the intervention is measurable.
12. Feedback May Be Part of the Effect
Another frequently omitted caveat appears in the original paper.
Students in the trial were shown solutions and asked to correct errors. The researchers note that informative and timely corrective feedback may be a necessary ingredient of successful interleaving.
Therefore:
[
\text{mixing questions}
\neq
\text{complete interleaving intervention}
]
A learner repeatedly selecting the wrong route without correction may simply rehearse the error.
The better loop is:
[
\text{Select}
\rightarrow
\text{Execute}
\rightarrow
\text{Check}
\rightarrow
\text{Correct}
\rightarrow
\text{Explain}
\rightarrow
\text{Select again later}
]
That is a control loop rather than a worksheet format.
13. What the Latest Replication Programme Changes
The 2020 trial is unusually compelling because it was randomised and conducted in normal classrooms.
But science becomes stronger when promising results are deliberately tested again.
The US Institute of Education Sciences is currently funding a systematic replication of interleaved mathematics practice. The project began in 2022 and its award period runs to July 2027. It is designed around a substantially larger and more heterogeneous sample: approximately 70 teachers, 189 Grade 7 classrooms and 3,780 students across 13 eastern US states. (ies.ed.gov)
The planned study adds features that matter:
- proximal and more distal outcomes;
- Grade 8 Mathematics Readiness;
- state-standardised tests;
- longer-term follow-up;
- implementation measurement;
- cost analysis;
- different proficiency groups. (ies.ed.gov)
This is exactly what a serious evidence programme should do.
It also gives us a publication rule:
Do not freeze the 2020 effect size into a universal educational constant while a deliberate large-scale replication is still underway.
MATHCIV-016 can confidently say the original classroom evidence is strong.
It should remain open to revision when the replication programme reports its main outcomes.
14. What This Means for Additional Mathematics
Additional Mathematics intensifies the method-selection problem because many questions sit at the intersection of several mathematical systems.
A student may know:
- algebra;
- functions;
- trigonometry;
- differentiation;
- integration;
- coordinate geometry.
Yet a mixed question still demands:
[
\text{Which structure controls the next move?}
]
This explains why Additional Mathematics preparation cannot end with chapter-by-chapter proficiency.
Chapter mastery is necessary.
Then headings must progressively disappear.
The final learner should not need the worksheet to say:
Differentiation Exercise 4
before recognising that differentiation is required.
The knowledge must become dispatchable.
15. The CivilisationOS Connection
There is a larger mathematical lesson here, but it needs a boundary.
Interleaving does not prove a theory of civilisation.
The useful connection is architectural.
Complex systems often require not merely possession of capabilities but selection among capabilities under changing conditions.
A city may possess multiple tools.
An institution may possess multiple procedures.
A learner may possess multiple mathematical methods.
In each case:
[
\text{Capability inventory}
\neq
\text{correct capability dispatch}
]
Mathematics gives us a particularly clean educational environment in which to observe this distinction.
The learner has a toolbox.
Interleaving asks whether the learner can select the right tool when the label has disappeared.
That is the defensible connection.
16. What Interleaving Does Not Prove
The evidence does not establish that:
- every mathematics lesson should be interleaved;
- blocked practice is ineffective;
- novices should discover every method through mixed problems;
- difficult practice is automatically better;
- the Florida Grade 7 effect size predicts Singapore students;
- Additional Mathematics would produce the identical treatment effect;
- discrimination is the sole causal mechanism;
- students should be denied examples or feedback;
- all mathematical topics benefit equally;
- interleaving by itself creates transfer.
The correct research posture is narrower.
Interleaving is a well-supported way of changing practice from:
perform the already-cued method
towards:
identify, select and perform an appropriate method.
That is important enough without exaggerating it.
17. A Parent’s Practical Test
There is a simple way to tell whether a student’s practice has progressed beyond chapter fluency.
Take several already-learned topics.
Remove the headings.
Mix the questions.
Do not tell the student which method each problem requires.
Then watch the first 20–30 seconds of each attempt.
Do not ask only:
Did they get the answer?
Ask:
Could they see what kind of mathematical object was in front of them?
Could they choose a defensible first move?
Did they switch methods when evidence contradicted the first route?
Could they explain the choice?
Could they still do it later?
That gives a much richer picture of capability.
18. A Student’s Practical Rule
Do not mix everything from day one.
Instead:
Learn it.
Practise it.
Close the notes.
Retrieve it.
Put it beside something similar.
Learn the difference.
Mix the questions.
Choose the method yourself.
Check the answer.
Return later.
The point of interleaving is not to make mathematics confusing.
The point is to make your decision part of the practice.
Conclusion
The most important thing interleaving teaches may happen before the calculation begins.
It asks:
What mathematics is this?
Which relationship governs it?
Which method belongs here?
Why this route rather than another one?
A learner can know several methods and still fail to dispatch them correctly.
That is why:
[
\boxed{\text{Knowing methods} \neq \text{knowing when to use them}}
]
Blocked practice remains valuable for acquisition and stabilisation.
Interleaving becomes valuable when the learner has enough capability that the next challenge is selection among alternatives.
The strongest current classroom trial provides substantial causal evidence for that broader practice architecture, while its own caveats tell us not to abolish blocked practice and not to ignore corrective feedback.
The newer 2025 evidence adds another warning: effective learning can feel more difficult than familiar repetition, so learner preference and immediate fluency are useful signals but not sufficient measures of durable learning. (ResearchGate)
And the ongoing IES systematic replication provides exactly the scientific correction we want:
keep the strong finding, test it harder, widen the population, use more distant outcomes, and remain willing to update the conclusion. (ies.ed.gov)
That is the MathematicsOS interpretation of interleaving:
[
\boxed{
\text{Acquire}
\rightarrow
\text{Stabilise}
\rightarrow
\text{Discriminate}
\rightarrow
\text{Select}
\rightarrow
\text{Execute}
\rightarrow
\text{Transfer}
\rightarrow
\text{Verify}
}
]
Not random mixing.
Training the choice of method, not just the method.
MATHCIV-016 status: upgraded and complete. The main upgrade over the original master is the stronger distinction between acquisition capacity and dispatch capacity, plus the new 2025 student-perception evidence and the still-running large IES replication. This also keeps MATHCIV-016 cleanly separated from MATHCIV-014 Retrieval and MATHCIV-015 Spacing, which should reduce cannibalisation across the science cluster.
The next registry article is MATHCIV-017 — “Productive Failure: When Problem Solving Before Instruction Works.”