Family: Foundation
Profile: Control
Canonical owner: eduKateSG
Research class: Frontier
Research checked: 10 August 2026

Quick Read

Being “good at mathematics” is not one thing.

A student can:

• understand a concept but calculate inaccurately;
• execute a familiar procedure but fail to choose it in a mixed problem;
• manipulate symbols accurately but misunderstand the graph they represent;
• solve a question today but fail to reconstruct the method next week;
• perform with hints but fail independently;
• obtain the correct answer without checking whether it makes mathematical sense;
• work accurately without time pressure but lose control during an examination;
• know substantial mathematics yet become unreliable when anxiety, overload or avoidance changes the operating state.

This is why MathematicsOS does not represent mathematical capability as one number.

It uses a 12-component Mathematics Capability Vector:

Concept → Procedure → Symbol → Representation → Selection → Transfer → Modelling → Verification → Fluency → Retention → Independence → Regulation

The important idea is not that these twelve components form a scientifically discovered universal scale.

They do not.

The vector is a control architecture: a disciplined way to prevent a mark, impression, dashboard or isolated correct answer from being mistaken for the whole learner.

The purpose is simple:

Measure enough of mathematical capability to locate the active bottleneck, choose the smallest useful repair, and then test whether the repaired capability survives delay, variation and independence.

---

1. The Direct Answer

Mathematical capability should be represented as a receiver-specific, topic-specific, context-specific and time-specific vector, rather than as a permanent label such as “strong at maths” or “weak at maths”.

The canonical MathematicsOS representation is:

[ K_{\text{math}}(receiver, topic, context, time) ]

with twelve components:

[ CON, PRO, SYM, REP, SEL, TRN, MOD, VER, FLU, RET, IND, REG ]

where:

• CON = conceptual structure
• PRO = procedural accuracy
• SYM = symbolic and algebraic control
• REP = representation construction and translation
• SEL = strategy and route selection
• TRN = transfer across surface forms and contexts
• MOD = modelling, assumptions and interpretation
• VER = verification, error detection and correction
• FLU = accurate fluency and timed execution
• RET = delayed retention
• IND = independent performance without external support
• REG = metacognitive, motivational and emotional regulation

The master specifically requires every component to remain attached to the topic, task family, representation, support level and time horizon being observed. A mark may provide evidence about part of the vector, but it cannot legitimately collapse the entire vector into a single learner identity.
That distinction is fundamental.

A student does not possess one permanent quantity called “mathematics ability”.

What we actually observe is performance under conditions.

From those observations, we infer capability—with uncertainty.

---

2. Why a Vector Is Necessary

Singapore’s mathematics curriculum itself is already broader than raw calculation. The official framework places mathematical problem solving at the centre and treats concepts, skills, processes, metacognition and attitudes as inter-related components.

But assessment systems inevitably compress.

A test gives a student a finite number of questions in a finite period. A score then summarises what happened.

That score matters.

It can be highly operationally useful.

But compression loses information.

Consider two students who both score 70%.

Student A may have:

• strong conceptual understanding;
• good route selection;
• weak algebraic accuracy;
• excellent checking;
• slow execution.

Student B may have:

• weak conceptual understanding;
• memorised procedures;
• fast execution;
• poor transfer;
• little verification.

The number is the same.

The control problem is entirely different.

If both students receive the same intervention because both obtained 70%, the system has thrown away precisely the information needed to improve them.

This is the reason for the vector.

---

3. The Vector Is Not Twelve New Grades

There is an important boundary here.

MathematicsOS should not turn the twelve components into twelve arbitrary percentages and then average them into a new super-score.

That would merely rebuild the original problem with more decoration.

The vector is better understood as a structured evidence map.

For each component, MathematicsOS asks:

1. What did we actually observe?
2. Under what conditions?
3. What support was available?
4. How fresh is the evidence?
5. How many task families have been sampled?
6. Has the capability survived a delay?
7. Has it survived a change in representation or context?
8. What alternative explanation could produce the same result?
9. What evidence is still missing?
10. What next probe would reduce uncertainty most efficiently?

The master therefore requires performance samples plus uncertainty, rather than intuition alone.

So the output should look more like:

Stable / Fragile / Missing / Context-bound / Support-dependent / Unknown

than:

73.4% mathematical capability.

---

4. CON — Conceptual Structure

Conceptual capability concerns the relationships that make a mathematical idea what it is.

A student may be able to perform a procedure without possessing the structure underneath it.

For example, a student may remember that:

[
(a+b)^2=a^2+2ab+b^2
]

but still fail to explain:

• why the middle term exists;
• why (a^2+b^2) is not equivalent;
• how the expression relates to area;
• what changes when the signs change;
• how expansion and factorisation reverse one another.

Useful sensors

Look for whether the learner can:

• explain a relationship without copying a definition;
• distinguish examples from non-examples;
• predict what should happen before calculating;
• connect related concepts;
• identify which conditions matter;
• explain why a procedure is valid.

Failure signatures

Possible signals include:

• correct answers with incoherent explanations;
• memorised formulas applied outside valid conditions;
• inability to recognise an equivalent form;
• inability to predict direction, sign or approximate behaviour;
• collapse when surface wording changes.

First repair route

Return to the mathematical invariant.

Use:

meaning → structure → contrasting examples → worked relation → learner reconstruction

before increasing drill volume.

A conceptual problem should not automatically be repaired with more repetition of the same procedure.

---

5. PRO — Procedural Accuracy

Procedural capability concerns executing a valid mathematical process correctly.

A student may understand the concept and still make unstable operational moves.

Examples include:

• expanding incorrectly;
• differentiating a term incorrectly;
• losing a denominator;
• applying an algorithm in the wrong sequence;
• carrying an invalid intermediate state forward.

Useful sensors

Inspect:

• full written working;
• first invalid step;
• repeated procedural errors;
• consistency across matched questions;
• self-correction;
• latency at each stage.

Failure signatures

Look for:

• correct method followed by execution breakdown;
• inconsistent performance across nearly identical items;
• recurring arithmetic or transformation errors;
• dependence on answer keys for confirmation.

First repair route

Isolate the unstable operation.

Then use:

correct model → guided completion → independent matched execution → delayed retest

The repair target is not “do more maths”.

It is the smallest unstable procedure capable of producing downstream error.

---

6. SYM — Symbolic and Algebraic Control

Symbolic capability concerns preserving mathematical relationships while transforming notation.

This becomes especially important in Additional Mathematics, where many problems require a student to carry a mathematical state through several transformations.

Examples include:

• expanding and factorising;
• rearranging equations;
• manipulating rational expressions;
• controlling indices, roots and logarithms;
• substituting parameters;
• maintaining domain restrictions;
• preserving signs and equivalence.

Useful sensors

Ask whether each written line remains mathematically equivalent to the previous line where equivalence is intended.

Also inspect:

• sign changes;
• denominator restrictions;
• brackets;
• substitution;
• equality;
• illegal cancellation;
• domain loss.

Failure signatures

A particularly revealing failure is:

the student understands what needs to be done but destroys the mathematical state while doing it.

That is not primarily a concept problem.

It is symbolic-control failure.

First repair route

Reduce the expression to short equivalence-preserving transitions.

Require:

one valid state → next valid state → verification

before rebuilding long multi-step chains.

---

7. REP — Representation Construction and Translation

Mathematics exists through multiple representations:

• symbols;
• graphs;
• diagrams;
• tables;
• numerical examples;
• coordinates;
• geometric objects;
• verbal descriptions;
• contextual models.

Capability therefore includes the ability to move between them.

A function such as

[y=(x-2)^2+1]

is simultaneously:

• an algebraic expression;
• a transformation of (y=x^2);
• a graph;
• a vertex;
• a mapping;
• a relationship between two variables.

A learner who knows only one representation possesses narrower capability than a learner who can translate among them.

Useful sensors

Ask the learner to move:

equation → graph

graph → equation

words → mathematical representation

table → pattern

diagram → relation

Failure signatures

Watch for:

• correct algebra but incorrect graph;
• correct graph reading but inability to construct the expression;
• treating visual and symbolic information as unrelated;
• adding multiple representations without integrating them.

Research supports the usefulness of multiple external representations, but also warns against assuming that simply adding more representations is automatically better. A 2024 systematic review and meta-analysis found positive but heterogeneous effects; the design, familiarity, informational redundancy and support around the representations matter.

First repair route

Do not add representations indiscriminately.

Select the second representation that exposes the missing invariant.

Then explicitly teach the translation.

---

8. SEL — Strategy and Route Selection

Some students can perform several mathematical methods correctly but do not know when to use which one.

This is a different capability.

Consider a student who can:

• factorise;
• use the quadratic formula;
• complete the square;

but chooses inefficiently or incorrectly when presented with an unfamiliar quadratic problem.

The procedures exist.

The routing system is weak.

Useful sensors

Use mixed problem sets in which the method is not announced.

Ask before calculation:

What kind of problem is this?

What is the first valid move?

Why this route rather than another?

Failure signatures

Look for:

• searching randomly among memorised methods;
• solving based on superficial keywords;
• correct work only when worksheets are blocked by topic;
• choosing expensive methods when simpler routes exist;
• beginning calculation before identifying structure.

First repair route

Contrast near-neighbour problem families.

Train:

recognise → discriminate → select → execute → compare

This is one reason interleaving can become valuable after initial acquisition: it forces learners to practise identifying which method is appropriate instead of merely repeating the method signalled by the worksheet heading.

---

9. TRN — Transfer

Transfer asks whether capability survives when something changes.

The change might involve:

• wording;
• representation;
• numerical values;
• problem direction;
• surface context;
• combination with another topic;
• unfamiliar structure.

Transfer is especially important because familiar performance can give an exaggerated impression of capability.

A major 2025 Nature study demonstrated strikingly asymmetric transfer between applied market arithmetic and academic school arithmetic: children could perform strongly in one context yet struggle with mathematically related tasks presented through another.

The result does not mean school mathematics is useless or informal mathematics is superior.

It means capability can be strongly context-indexed.

Useful sensors

After stabilisation, change one dimension at a time:

• same structure, different wording;
• same relation, different representation;
• same concept, reversed question;
• same method embedded in a mixed problem;
• same mathematics in a contextual setting.

Failure signatures

A common pattern is:

“I knew how to do it on the worksheet, but I didn’t know this was the same thing.”

That is a transfer problem.

First repair route

Expose the invariant across variation.

Ask repeatedly:

What changed?

What stayed mathematically the same?

Recent 2026 experimental evidence adds an important refinement. Retrieval practice and worked examples do not have one universal ranking: their relative benefit for generalisation changes with initial instruction and task demands. When learners already had explicit initial instruction, retrieval practice produced stronger generalisation in the reported experiments; without that initial instruction, worked examples could be superior under repeated-item conditions.

That strengthens a central MathematicsOS rule:

intervention depends on receiver state.

---

10. MOD — Modelling, Assumptions and Interpretation

Modelling is the capability to construct, use and critique a mathematical representation of some target situation.

A modelling answer is not complete merely because the calculation is correct.

The learner must also understand:

• what the variables represent;
• which assumptions were made;
• what domain is meaningful;
• whether the result is physically or contextually possible;
• which information the model excludes;
• how strongly the conclusion follows from the model.

Useful sensors

Ask:

• What does this variable mean?
• Why is this relationship appropriate?
• What assumptions are being made?
• What range is meaningful?
• What would make this model fail?
• Does the answer make sense in the original situation?

Failure signatures

Watch for:

• mathematically valid but contextually impossible results;
• unexplained parameters;
• treating a model as reality;
• ignoring units;
• extrapolating far outside the observed range.

First repair route

Use:

Reality → quantities → assumptions → relation → calculation → interpretation → limitation

This component is one of the bridges between school mathematics and the larger CivilisationOS thesis.

Mathematics becomes socially valuable partly because models allow complex relationships to become inspectable.

But the model must never be mistaken for Reality.

---

11. VER — Verification, Error Detection and Correction

Verification is one of the most important—and frequently neglected—mathematical capabilities.

A learner should eventually be able to ask:

How could I discover that my own answer is wrong?

Possible verification channels include:

• substitution;
• inverse operations;
• estimation;
• dimensional or unit checks;
• graph behaviour;
• sign analysis;
• domain restrictions;
• alternative methods;
• differentiation of an integral;
• integration checks;
• boundary cases.

Useful sensors

Do not merely ask whether the answer is correct.

Ask:

How do you know?

Then remove the answer key.

Failure signatures

Look for:

• acceptance of impossible answers;
• checking by rereading the same calculation;
• dependence on the tutor’s facial expression;
• inability to devise an independent check;
• high-confidence errors.

First repair route

Attach at least one verification operation to each major problem family.

The objective is eventually:

solve → inspect → challenge → correct

A mathematically capable learner should not need an external authority to validate every line.

---

12. FLU — Accurate Fluency and Timed Execution

Fluency is not merely speed.

Speed without stable accuracy can be fast failure.

MathematicsOS therefore treats fluency as accurate, efficient execution in which:

• the route is appropriate;
• working remains readable;
• errors remain recoverable;
• performance survives realistic cognitive load.

Useful sensors

Measure jointly:

• accuracy;
• completion time;
• route quality;
• unnecessary steps;
• recovery after error;
• working clarity.

Failure signatures

There are several distinct cases:

slow but correct
This may be a fluency issue.

fast but inaccurate
This may be premature speed.

fast on blocked worksheets, slow on mixed problems
This may actually be selection weakness.

accurate untimed, unstable under time pressure
This may involve fluency, regulation, retrieval or cognitive-load constraints.

First repair route

Never optimise latency before the method is valid.

Use:

accuracy → stability → route efficiency → controlled timing → whole-task execution

---

13. RET — Delayed Retention

Immediate performance is not the same as retained capability.

A student may solve five examples correctly immediately after instruction because the representation remains active.

The more difficult question is:

Can the student reconstruct the mathematics after the support and immediate memory have faded?

The MathematicsOS measurement contract therefore requires delayed independent reconstruction before claiming that a learner retains a capability.

Useful sensors

Re-test after a meaningful interval:

• without warning;
• without reviewing the worked example first;
• with a structurally matched but fresh question.

Failure signatures

The classic signature is:

“I understood it last lesson.”

That statement may be true.

Receipt may have occurred.

Retention may not.

First repair route

Use:

reconstruction → spacing → retrieval → feedback → later reconstruction

rather than repeated immediate exposure alone.

---

14. IND — Independent Performance

Independence asks whether the learner owns the operation.

Support can include:

• tutor prompting;
• peer help;
• worked solutions;
• hints;
• calculators where they replace the target operation;
• answer keys;
• AI.

A supported answer can be useful evidence of learning in progress.

But it is not automatically evidence of independent capability.

Useful sensors

Record support explicitly.

Then test the same structural capability under reduced support.

The master requires independence to mean performance without tutor, answer key, peer or AI when independence is the target.

Failure signatures

Look for:

• correct work immediately after hints;
• repeated requests for confirmation;
• inability to initiate without prompting;
• AI-produced explanation followed by weak unaided reconstruction;
• correct copying without generative control.

First repair route

Fade assistance systematically.

Use:

supported → partially supported → prompt-only → independent → delayed independent

Help should progressively disappear when the objective is independent mathematical operation.

---

15. REG — Regulation

Mathematical performance also depends on how the learner monitors and controls their own state.

REG includes several related but distinct processes:

• metacognitive monitoring;
• confidence calibration;
• strategy adjustment;
• persistence;
• attention;
• motivation;
• emotional state;
• anxiety or avoidance signals.

This does not mean poor mathematics performance should be psychologised.

Nor does it mean positive thinking substitutes for mathematical repair.

It means the operating state of the receiver can materially change observable performance.

Research supports treating metacognition as relevant but not magical. A 2024 meta-analysis of 35 problem-solving studies found a small overall positive effect of monitoring interventions, with effectiveness depending on intervention type; whole-task approaches, metacognitive knowledge and external standards performed differently from some other prompting approaches.

A separate 2024 meta-analysis reported a positive association between metacognition and mathematics achievement, but association is not proof of one-way causation.

Useful sensors

Observe or ask:

• How confident are you?
• Did confidence match accuracy?
• When did you realise the route was failing?
• What did you do next?
• Do difficult questions trigger avoidance?
• Does performance change sharply under timing?
• Can the learner choose a recovery strategy?

Failure signatures

Possible signals include:

• high confidence with repeated error;
• very low confidence despite stable accuracy;
• abandoning viable routes prematurely;
• continuing a clearly invalid route;
• freezing under examination load;
• avoidance of exposing working.

First repair route

Repair both mathematics and regulation where both are binding.

Do not replace mathematical instruction with reassurance.

Do not replace humane emotional support with more worksheets.

The receiver may need both.

---

16. One Student Can Have Twelve Different States

Suppose a Secondary student is learning differentiation.

A possible vector might look like this:

CON — stable
Understands derivative as rate of change and gradient.

PRO — stable
Can differentiate standard forms accurately.

SYM — fragile
Frequently loses negative signs and mishandles fractional indices.

REP — fragile
Does not reliably connect derivative sign to increasing/decreasing graph behaviour.

SEL — stable on blocked work, unknown on mixed questions

TRN — weak
Fails when the same concept is embedded in unfamiliar wording.

MOD — emerging
Can calculate rates but struggles to interpret them contextually.

VER — weak
Rarely checks whether an answer is plausible.

FLU — adequate

RET — unknown
No delayed test has occurred.

IND — support-dependent
Frequently asks the tutor to confirm the first step.

REG — generally stable, but confidence poorly calibrated

What is the student’s mathematics level?

That question is now visibly too crude.

The better question is:

Which component currently constrains the next useful capability?

That is the beginning of control.

---

17. The Moving Bottleneck Principle

The weakest component is not a permanent learner characteristic.

It can change by:

• topic;
• task;
• phase of learning;
• representation;
• time;
• support state.

A student may have excellent symbolic control in quadratics and weak symbolic control in logarithms.

A student may transfer well between algebraic forms but poorly between algebra and geometry.

A student may be independently capable when untimed and support-dependent during examination stress.

Therefore MathematicsOS uses a moving bottleneck set, not a fixed label.

The master explicitly prohibits turning dashboard states into learner identities. Its control panel should display bottlenecks, competing hypotheses, data freshness, missing probes and uncertainty—not a single “brain score” or unsupported future-grade prediction.

---

18. Sensors Are Not Reality

Every measurement has limits.

A correct question gives evidence.

It does not reveal the entire underlying mechanism.

For example:

Observation

The learner answered three factorisation questions incorrectly.

Possible inference A

Factorisation procedure is unstable.

Possible inference B

Negative-sign control is unstable.

Possible inference C

The learner misunderstood the question.

Possible inference D

The learner was rushing.

Possible inference E

A prerequisite involving multiplication or algebraic structure is missing.

The work sample cannot automatically choose among them.

That is why MathematicsOS keeps:

Observation ≠ Inference ≠ Model

The next useful action is often not another explanation.

It is a better diagnostic probe.

---

19. The Minimum Evidence Architecture

A Mathematics Capability Vector becomes useful only if evidence is collected consistently.

The master specifies a minimum measurement structure containing:

Inputs

• prior state;
• topic/task mix;
• practice attempts;
• support mode;
• hint depth;
• instructional exposure.

Process

• learner-generated working;
• route selection;
• error type;
• self-correction;
• feedback;
• affect or avoidance observations.

Immediate output

• accuracy;
• latency;
• representation quality;
• verification behaviour.

Delayed output

• retention after a declared interval.

Transfer

• changed surface form;
• changed representation;
• changed context;
• multi-topic bridge.

Independence

• performance without external support.

These distinctions are part of the master measurement contract.

This allows a tutor or system to say something much more precise than:

“The student improved.”

It can instead say:

“Procedural execution became stable on matched questions; symbolic sign errors declined; near-transfer now passes independently; delayed retention has not yet been tested.”

That statement contains far more control information.

---

20. Unknown Is a Legitimate State

One of the most important MathematicsOS rules is:

Unknown does not mean weak.

If transfer has not been tested, TRN is unknown.

If there has been no delayed probe, RET is unknown.

If the learner has always worked with prompts, IND is unknown.

If no mixed questions have been attempted, SEL may be unknown.

This sounds obvious.

In practice, educational systems frequently fill missing information with assumptions.

MathematicsOS refuses to do that.

Unknown generates a measurement request.

It does not generate a learner judgement.

---

21. Repair Should Match the Component

The vector prevents a common failure in mathematics education:

every problem receives the same treatment—more practice.

But different capability failures require different repairs.

A useful first routing map is:

CON weak
→ rebuild meaning and relations.

PRO weak
→ isolate and stabilise the procedure.

SYM weak
→ repair equivalence-preserving transformations.

REP weak
→ explicitly translate between representations.

SEL weak
→ contrast problem families and practise route choice.

TRN weak
→ vary surface form while preserving the invariant.

MOD weak
→ reconnect quantities, assumptions and interpretation.

VER weak
→ install independent checking procedures.

FLU weak
→ increase efficiency only after accuracy stabilises.

RET weak
→ use delayed retrieval and spacing.

IND weak
→ fade help and retest unsupported.

REG weak
→ repair monitoring, calibration, strategy or emotional constraint alongside mathematical capability.

This is why diagnosis matters.

The intervention should correspond to the failure mechanism we actually have evidence for.

---

22. From Vector to Control Loop

The Mathematics Capability Vector is not intended to sit inside a report.

It belongs inside a runtime:

[
Sense
\rightarrow
Index
\rightarrow
Map
\rightarrow
Locate
\rightarrow
Protect
\rightarrow
Select
\rightarrow
Teach
\rightarrow
Practise
\rightarrow
Verify
\rightarrow
Writeback
\rightarrow
Meta\text{-}Control
]

The essential cycle is:

1. Sense

Collect actual learner work.

2. Map

Update the relevant vector components with uncertainty.

3. Locate

Identify the strongest current bottleneck hypothesis.

4. Probe

Choose the question that best distinguishes competing explanations.

5. Repair

Apply the smallest sufficient intervention.

6. Verify

Test the repaired capability.

7. Delay

Test again after immediate support has faded.

8. Transfer

Change the surface or representation.

9. Remove support

Test independence.

10. Write back

Update the vector only with qualifying evidence.

Then repeat.

---

23. Why This Matters for Small-Group Teaching

The vector also changes what a small mathematics class is for.

The value of a small group is not merely that fewer students occupy the room.

Its potential value lies in increased sensing bandwidth:

• more learner working can be inspected;
• questioning can become more diagnostic;
• feedback latency can fall;
• different learners can receive different probes;
• support dependence can become visible;
• the tutor can detect when the bottleneck moves.

Recent randomised evidence from Denmark provides an example of why tailored diagnosis can matter. A 2025 study tested structured small-group mathematics interventions for low-achieving Grade 2 and Grade 8 pupils and found positive immediate effects under the study’s intensive, tailored design, although medium-term persistence differed by age group and the findings do not establish a universal optimal group size.

The lesson is not:

small groups automatically work.

It is:

better sensing and tailored intervention can matter when implemented as an actual control system.

---

24. The Vector and Examinations

The capability vector does not diminish the importance of examination marks.

It explains them more precisely.

An examination asks the student to convert underlying capability into performance under specific conditions.

A strong examination result therefore requires several components to cooperate:

[
Concept
+
Procedure
+
Symbol
+
Representation
+
Selection
+
Transfer
+
Verification
+
Fluency
+
Retention
+
Independence
+
Regulation
]

Different questions sample different combinations.

This explains why simply “knowing the chapter” can still fail to produce marks.

The breakdown may occur in:

• question recognition;
• route choice;
• symbolic execution;
• timing;
• checking;
• transfer;
• regulation.

Marks remain important.

But now they become a sensor inside a larger diagnostic system, rather than the entire model of the learner.

---

25. The CivilisationOS Connection

At first glance, this article is about an individual mathematics learner.

But it connects directly to the larger Boosting Cities and CivilisationOS programme.

A civilisation also becomes difficult to understand when complex capability is collapsed into one number.

GDP is not civilisation.

A test score is not education.

Infrastructure quantity is not receiver access.

Delivered policy is not realised outcome.

Likewise:

a mathematics mark is not the whole mathematical capability system.

The common design principle is:

[
State\ signal
\neq
Capability
]

We need distributions, components, bottlenecks, uncertainty, transfer and receiver-realised outcomes.

The Mathematics Capability Vector therefore becomes a small, concrete laboratory for a much larger idea:

Complex adaptive capability should be measured in enough dimensions to preserve the mechanisms required for control.

But the scale mapping must remain disciplined.

A learner and a civilisation are not the same adaptive unit.

The vector is a MathematicsOS domain instrument.

It is not a universal civilisation equation.

---

26. What the Research Does—and Does Not—Prove

Research supports many ingredients behind the vector:

• transfer can be context-sensitive;
• representation design matters;
• monitoring accuracy can be improved under some interventions;
• metacognition is associated with mathematics achievement;
• instructional strategy effectiveness depends on learner preparation and task demands;
• tailored mathematics intervention can improve outcomes in bounded populations.

But these studies did not discover the exact twelve-component MathematicsOS vector.

The vector is an architectural synthesis.

Its scientific legitimacy therefore depends on what we do next.

We must test whether:

• its components can be observed reliably;
• tutors distinguish them consistently;
• the vector predicts different repair needs better than simpler approaches;
• uncertainty states prevent false diagnosis;
• component-matched repair improves delayed and transferable capability;
• the system remains practical enough to operate;
• the measurement burden does not exceed its educational value.

That is how an architecture becomes a testable system.

---

27. The Most Important Rule

The Mathematics Capability Vector should never become another way to label a child.

Its purpose is the opposite.

Instead of saying:

“This student is weak at mathematics.”

we can ask:

Which mathematical operation is currently unstable?

Under what conditions?

What evidence supports that conclusion?

What remains unknown?

What is the smallest useful repair?

How will we know the repair survived?

That shift matters.

Because once capability is decomposed correctly, a large vague problem often becomes a small solvable one.

---

Final Takeaway

Mathematical capability is not a scalar.

It is a changing configuration of conceptual, procedural, symbolic, representational, strategic, transferable, modelling, verification, fluency, retention, independence and regulatory capabilities.

The Mathematics Capability Vector gives MathematicsOS a way to observe that configuration without pretending the observation is the learner.

Its purpose is not more measurement for its own sake.

Its purpose is better control:

[
\boxed{
Sense
\rightarrow
Locate
\rightarrow
Repair
\rightarrow
Verify
\rightarrow
Delay
\rightarrow
Transfer
\rightarrow
Independent\ Use
}
]

And the deepest rule is simple:

Measure only what helps us understand what the learner can actually do, what is blocking the next capability, and what evidence would show that the repair has become real.

That is the difference between a dashboard and a learning system.