OCR A-Level Mathematics B (MEI) H640 | Mechanics, Statistics, Comprehension and 2026 Data-Set Update

OCR A-Level Mathematics B (MEI) H640 for 2027: Pure, Mechanics, Statistics, Comprehension and Large Data Set Control

OCR A-Level Mathematics B (MEI) H640 is distinctive because the qualification does not treat Pure Mathematics, Mechanics, Statistics and mathematical communication as isolated paper identities. The three assessment components repeatedly ask students to carry Pure Mathematics into applied and interpretive settings. For the 2027 examination year, there is also a specific large-data-set boundary that must be kept current: OCR states that H640/02 uses Large Data Set 9 in June 2027, while H640/02 moves to Large Data Set 10 from June 2028 onward.

This page therefore separates two layers. The mathematical layer is durable: algebra, functions, calculus, trigonometry, vectors, mechanics, statistics, modelling and comprehension. The administrative layer is dated: component structure, large data set, examination series and current OCR instructions.

1. H640 has three assessment components

OCR’s current specification-at-a-glance gives the A Level H640 structure as:

• H640/01 Pure Mathematics and Mechanics — 100 marks, 2 hours, 36.4%;
• H640/02 Pure Mathematics and Statistics — 100 marks, 2 hours, 36.4%;
• H640/03 Pure Mathematics and Comprehension — 75 marks, 2 hours, 27.3%.

The structure matters because Pure Mathematics appears throughout the qualification. A student cannot prepare Mechanics, Statistics or Comprehension as if Pure Mathematics were a separate finished block.

2. The 2027 large-data-set rule is specific

OCR’s September 2026 subject update says that H640/02 uses Large Data Set 9 for June 2027. OCR also says that H640/02 will use Large Data Set 10 from June 2028 onwards. AS Mathematics B (MEI) H630/02 moves to Large Data Set 10 from June 2027, so AS and A Level learners are not necessarily using the same examination data set in the same year.

That distinction prevents a common current-information error: seeing “Large Data Set 10 from 2027” and applying it to H640 A Level rather than to H630 AS.

3. Why the large data set exists

OCR’s H640 specification treats the pre-release large data set as a teaching context, not a memory test. Students are expected to become familiar with the data, explore it using spreadsheets or statistical software, investigate correlation, model probabilities, perform sampling and connect the data to hypothesis testing.

The examination does not simply ask candidates to reproduce the spreadsheet. Selected data or summary information may be supplied in the paper, while familiarity with the context can give a material advantage.

4. Familiarity should mean statistical understanding

A weak large-data-set preparation strategy tries to memorise unusual values. A stronger strategy asks:

• What are the variables?
• What units are used?
• Which variables are categorical or numerical?
• Which distributions are skewed?
• Which variables might be associated?
• What would a scatter plot show?
• Which summaries are robust?
• What sampling questions could be asked?
• What contextual limitations matter?

The student should know the data well enough to reason with it rather than merely recognise names.

5. Pure Mathematics remains the common language

Across H640, Pure Mathematics supports every component. The important dependencies include:

• algebraic manipulation;
• functions and graphs;
• sequences and series;
• trigonometry;
• exponentials and logarithms;
• differentiation;
• integration;
• numerical methods;
• vectors;
• proof and mathematical argument.

A Mechanics problem can fail because algebra is weak. A Statistics problem can fail because logarithms are weak. A Comprehension problem can fail because the student cannot reconstruct a mathematical argument.

6. Algebra should be treated as infrastructure

Consider:

[ rac{x^2-9}{x-3}. ]

For (x eq3), factor:

[ rac{(x-3)(x+3)}{x-3}=x+3. ]

The simplification is valid only while the domain restriction (x eq3) remains attached. Advanced Mathematics repeatedly rewards students who preserve conditions rather than cancel symbols mechanically.

7. Functions should be read across representations

A function is not only a formula. The learner should move among:

• equation;
• graph;
• table;
• rate of change;
• intercepts;
• turning points;
• domain and range;
• context.

This is particularly important in H640/03 comprehension questions, where a mathematical text may present an argument in a different representation from the one a student would normally choose.

8. Original function task

Let:

[ f(x)=x^3-3x^2-9x+5. ]

Then:

[ f'(x)=3x^2-6x-9=3(x-3)(x+1). ]

The stationary points occur at (x=-1) and (x=3). A complete answer should continue from derivative calculation to sign analysis and interpretation of the original graph.

9. Mechanics begins with a model

In Mechanics, equations such as (F=ma), constant-acceleration formulae and momentum relationships are valid inside assumptions. Before calculating, identify:

• the object or system;
• the direction convention;
• forces;
• whether acceleration is constant;
• whether particles/strings are being idealised;
• whether resistance is included or neglected.

Model first, formula second.

10. Original Mechanics task: force and acceleration

A particle of mass 4 kg experiences a resultant horizontal force of 18 N.

[ a=rac{F}{m}=rac{18}{4}=4.5 ext{ m/s}^2. ]

If the 18 N is only one applied force and another force opposes motion, the calculation is wrong because the resultant has not yet been formed.

11. Original Mechanics task: connected particles

Two particles of masses 2 kg and 3 kg are connected by a light inextensible string on a smooth horizontal surface. A horizontal force of 25 N pulls the 3 kg particle.

Total mass is 5 kg, so:

[ a=25/5=5 ext{ m/s}^2. ]

For the 2 kg particle:

[ T=2(5)=10 ext{ N}. ]

The system view finds acceleration; the component view finds tension.

12. Mechanics checking should include dimensions and sign

If a calculated acceleration has units of newtons, the expression is structurally wrong. If a particle expected to decelerate produces a positive acceleration under the chosen sign convention, inspect the force equation before accepting the result.

13. Statistics begins with the data-generating model

Students should not choose a distribution merely because a formula is familiar.

For a binomial model, check:

• fixed number of trials;
• two outcome categories;
• constant success probability;
• independence.

For a normal model, ask whether a continuous approximately bell-shaped model is plausible in context.

14. Original binomial task

If (Xsim B(15,0.2)), then the probability of no successes is:

[ P(X=0)=0.8^{15}. ]

The probability of at least one success is:

[ 1-0.8^{15}. ]

The complement is the efficient route.

15. Expectation is not a guaranteed observation

For (Xsim B(n,p)):

[ E(X)=np. ]

If (n=20) and (p=0.3), the expected number of successes is 6. That does not mean every sample will contain exactly six successes.

16. Hypothesis testing should return to context

A hypothesis test is incomplete when the candidate ends with “reject (H_0)” and never explains what that means for the original claim.

Use the sequence:

model → hypotheses → significance level → test probability/statistic → decision → contextual conclusion.

17. Large-data-set work should include visual analysis

OCR’s specification explicitly expects technological exploration of the large data set. Students should create and interpret:

• histograms;
• box plots;
• scatter plots;
• summary statistics;
• subgroup comparisons.

The purpose is to build statistical intuition about the actual data context.

18. Same average, different distribution

Suppose two subsets both have mean 10.

Subset A: 9, 10, 10, 10, 11.

Subset B: 0, 5, 10, 15, 20.

The same mean hides radically different spread. A large-data-set learner should automatically ask about distribution shape and variability.

19. Correlation should not become causation

A scatter plot can show association. It does not by itself establish that one variable causes the other. Possible confounding variables, selection effects and common causes remain.

H640 statistics preparation should therefore include sentences such as:

“The data show an association between X and Y, but the observational relationship alone does not establish causation.”

20. Sampling connects the data set to inference

OCR encourages students to use the LDS as a population for repeated sampling. This helps learners see:

• sample-to-sample variability;
• effect of sample size;
• difference between sample statistics and population parameters;
• why one sample can mislead.

21. Original sampling activity

Take a known dataset of 100 values. Draw repeated samples of size 10 and record the sample mean. Then repeat with samples of size 40. The larger samples should usually show less variation in sample mean.

The activity makes sampling variability visible rather than abstract.

22. H640/03 Comprehension is not an English paper

The Comprehension component assesses Pure Mathematics through a mathematical article or passage. The student must read unfamiliar mathematical exposition and then reason with it.

The challenge is not literary style. It is mathematical recovery:

• identify definitions;
• track notation;
• follow the argument;
• distinguish a result from an example;
• use the article’s ideas in a new step.

23. Build a comprehension annotation system

While reading, mark:

• D — definition;
• R — stated result;
• E — example;
• A — assumption;
• C — conclusion;
• ? — step not yet understood.

This keeps the mathematical hierarchy visible.

24. Do not memorise the article wording

A comprehension question can ask the learner to rebuild an idea in a slightly different setting. Copying a phrase without understanding the mathematical dependency will fail.

After reading a section, close the text and explain the key result in your own mathematical language.

25. Original comprehension mini-passage

Suppose a sequence is defined by:

[ u_{n+1}=0.5u_n+6. ]

If the sequence converges to a limit (L), then the limit must satisfy:

[ L=0.5L+6. ]

Therefore:

[ 0.5L=6,quad L=12. ]

A comprehension question can then ask why convergence was assumed before solving the fixed-point equation. The equation gives a candidate limit; it does not by itself prove the sequence converges.

26. Pure Mathematics and Comprehension should be trained together

Once per week, read a short mathematical article, proof or textbook section outside the immediate class notes. Extract definitions and reconstruct one argument.

The aim is to reduce dependence on familiar teacher phrasing.

27. Section A and Section B require different pacing

OCR describes H640/01 and H640/02 as having a shorter-question Section A and a longer/problem-solving Section B.

A student should not spend disproportionate time perfecting an early short question while later extended problems remain untouched.

28. Two-pass strategy

Pass 1:

• secure short questions;
• complete longer questions with clear routes;
• mark high-friction problems.

Pass 2:

• return to difficult sections;
• check extended algebra;
• verify modelling conclusions.

29. Error classification should be component-aware

30. Fresh retest after feedback

If an error was caused by using correlation as causation, retest with a different observational context. If the error was force modelling, change the physical arrangement. If the error was comprehension, use a different mathematical passage.

Fresh transfer shows whether the repair is general.

31. 2027 LDS preparation cycle

Because H640/02 uses LDS 9 in June 2027, students should:

1. obtain LDS 9 from OCR;
2. learn the variables and context;
3. explore distributions;
4. create scatter plots and summaries;
5. compare subgroups;
6. perform sampling investigations;
7. connect the data to hypothesis-testing questions;
8. practise interpreting selected summaries without having the full dataset printed in the exam.

32. Do not switch H640 to LDS 10 too early

OCR’s September 2026 update explicitly says LDS 9 remains the H640/02 dataset for June 2027. LDS 10 is the H640/02 dataset from June 2028 onwards. Keep these cohort years visible on revision resources.

33. Use old datasets for teaching, not current-year confusion

OCR says teachers can continue using previously published datasets for teaching and mock work. That is useful because the statistical methods transfer. But final current-year familiarity should return to the prescribed dataset for the live H640 series.

34. Original LDS-style question

Suppose a dataset contains household income and weekly discretionary spending. A scatter plot shows positive association.

Ask:

• Would Pearson correlation be appropriate?
• Are there outliers?
• Is the relationship linear?
• Can causation be claimed?
• What subgroup might distort the pattern?

The learner should think statistically before pressing a calculator key.

35. Mechanics should be checked with limiting cases

If a friction coefficient tends to zero, does the model approach a smooth-surface result? If an applied force tends to zero, does acceleration behave as expected?

Limiting cases are powerful independent checks.

36. Calculus should be checked through representation

If differentiation gives a positive derivative across an interval but the graph drawn is decreasing, one representation is wrong. Use graph, algebra and numerical samples as independent evidence.

37. Numerical methods should retain the target equation

An iteration can produce numbers that appear to converge. The learner should still know which equation the fixed point solves and whether the iteration is valid in the chosen starting region.

38. Original numerical-method task

Use:

[ x_{n+1}=rac12left(x_n+rac{10}{x_n} ight) ]

to approximate (sqrt{10}). Starting with (x_0=3):

[ x_1=rac12(3+10/3)approx3.1667. ]

The sequence is useful because its fixed point satisfies (x=10/x), hence (x^2=10).

39. Proof should distinguish evidence from argument

Checking ten values is evidence for a pattern. It is not proof of a universal statement.

For example, (n^2+n) is even for every integer (n) because:

[ n^2+n=n(n+1), ]

a product of consecutive integers, one of which is even.

40. Six-week H640 cycle

Week 1: Pure diagnostic

Test algebra, functions, trigonometry, exponentials/logarithms and calculus dependencies.

Week 2: Mechanics and Statistics models

Repair model assumptions, force diagrams and probability-distribution choice.

Week 3: LDS 9 familiarity

Explore the current 2027 dataset using technology.

Week 4: Comprehension

Read and reconstruct unfamiliar mathematical arguments.

Week 5: timed component sections

Practise H640/01, /02 and /03 styles under time.

Week 6: mixed transfer

Retest recurring failure categories using unseen problems.

41. Student release checklist

• I know the three H640 components.
• I know that LDS 9 is the H640/02 dataset for June 2027.
• I know LDS 10 applies to H640/02 from June 2028.
• I can work across Pure, Mechanics and Statistics.
• I state models before formulas.
• I can interpret statistical output.
• I can read unfamiliar mathematical prose.
• I can identify definitions, assumptions and conclusions.
• I can check answers using an independent representation.
• I can transfer feedback to a fresh problem.

42. Official-source control

Use OCR’s current H640 qualification page, specification and subject updates for live administrative facts. The September 2026 OCR update is the current source for the Large Data Set transition: H640/02 uses LDS 9 in June 2027 and LDS 10 from June 2028 onwards.

Official routes: OCR Mathematics B (MEI) H630/H640 and OCR Large Data Set pre-release arrangements, September 2026.

43. The durable H640 model

The complete preparation loop is:

Pure foundation → applied model → current LDS context → mathematical comprehension → timed component practice → first-error diagnosis → fresh transfer → current OCR rule check.

That model keeps the 2027 dataset change accurate without turning the qualification into a date-maintenance page.

H640 Mastery Clinic: Three Papers, One Mathematical Operating System

The easiest way to under-prepare for OCR Mathematics B (MEI) H640 is to study each paper as if it were a separate subject. The stronger model is to treat H640 as one mathematical operating system expressed through three different assessment environments.

Paper 01 asks whether Pure Mathematics can survive contact with Mechanics. Paper 02 asks whether Pure Mathematics can survive contact with Statistics and the current large data set. Paper 03 asks whether Pure Mathematics can survive contact with unfamiliar mathematical exposition. The shared bottleneck is therefore not “which paper am I weak at?” but “which mathematical capability fails when the representation changes?”

Paper 01: Pure Mathematics and Mechanics

The Mechanics component becomes much easier when every question is decomposed into four decisions:

1. What is the system?
2. What assumptions are being made?
3. Which direction or sign convention is being used?
4. Which mathematical relationship follows from that model?

Students often reverse the order: they search memory for a formula and then force the diagram to fit. That is fragile.

Original Mechanics clinic: inclined plane

A 5 kg particle rests on a rough plane inclined at 20 degrees to the horizontal. Suppose a force acts up the plane and the particle is in limiting equilibrium.

The first job is not calculation. The first job is to resolve forces parallel and perpendicular to the plane, identify normal reaction, friction direction and any applied force. If coefficient of friction is (mu), limiting friction is (mu R), but only when the system is at limiting equilibrium.

The phrase “limiting equilibrium” is therefore a modelling condition, not decoration.

Connected-particle reasoning

For two connected particles, decide whether to treat the pair as one system or as separate bodies.

Whole-system view can eliminate internal tension and reveal acceleration quickly.

Single-particle view can then recover tension.

This two-level representation is more robust than writing simultaneous force equations immediately without deciding what each equation represents.

Moments and equilibrium

When taking moments, choose a pivot strategically. A good pivot removes unknown forces whose lines of action pass through it.

That is method selection, not a memorised rule.

A student should be able to explain why a chosen pivot simplifies the system.

Paper 02: Pure Mathematics and Statistics

H640/02 is where current-year information matters most because the pre-release large data set is part of the paper’s statistical context. For June 2027, OCR specifies Large Data Set 9. For H640 from June 2028, OCR moves to Large Data Set 10.

Students should label every LDS revision resource with the examination year. This prevents an otherwise excellent statistics lesson from becoming a wrong-current-year preparation pack.

Build a data dictionary

For LDS 9, create a one-page data dictionary containing:

• variable name;
• meaning;
• unit;
• data type;
• possible missing-value issues;
• obvious contextual relationships;
• likely statistical representations.

This is more useful than memorising isolated values.

Original LDS-style investigation

Suppose a dataset contains age, annual income, region and weekly travel distance.

Ask:

• Which variable is categorical?
• Which are continuous or discrete numerical variables?
• What plot is suitable for income distribution?
• What graph is suitable for income versus travel distance?
• Which summary is resistant to extreme income values?
• Would a positive correlation prove income causes longer travel?

These are the habits OCR wants large-data-set familiarity to support.

Residual thinking

When fitting a linear model, do not stop at correlation coefficient or regression equation. Inspect residuals.

A curved residual pattern can show that a linear model is missing structure even when the correlation looks strong.

Hypothesis-test language

Students should practise saying:

“There is sufficient evidence at the stated significance level to reject the null hypothesis in favour of the alternative.”

Then translate that into the actual context.

Do not write “the alternative is proved”. Statistical evidence is not logical proof.

Paper 03: Pure Mathematics and Comprehension

The comprehension component rewards students who can learn Mathematics while reading. This is a distinct skill from recalling classroom notes.

Train it by taking a short mathematical article and asking:

• What is newly defined?
• What result is assumed?
• What result is proved?
• What example illustrates rather than proves?
• What notation is local to this article?
• What step would I have to reconstruct if the text were removed?

Original comprehension passage: iteration

Suppose a function (g) has a fixed point (L), meaning (g(L)=L). An iteration is defined by:

[ x_{n+1}=g(x_n). ]

If the sequence converges to (L) and (g) behaves continuously near the limit, then taking limits suggests:

[ L=g(L). ]

A comprehension question might ask why solving (L=g(L)) does not by itself prove convergence. The answer is that the fixed-point equation identifies possible limits; convergence requires additional conditions.

Comprehension note-taking should be temporary

Do not cover the article in permanent annotations before understanding the argument. Read one paragraph, write the mathematical job in the margin, then test whether you can reconstruct the dependency.

Cross-paper transfer task 1: function + mechanics

A particle’s displacement is:

[ s(t)=t^3-6t^2+9t. ]

Velocity:

[ v(t)=3t^2-12t+9=3(t-1)(t-3). ]

The same derivative machinery used in Pure Mathematics now carries a physical interpretation. The zeros of velocity identify moments of instantaneous rest.

The learner should then inspect sign to determine direction of motion.

Cross-paper transfer task 2: exponential + statistics

Suppose a population is modelled by:

[ P(t)=P_0e^{kt}. ]

Taking logarithms gives:

[ ln P=ln P_0+kt. ]

This can transform an exponential relationship into a linear one for modelling. Pure logarithms, graph interpretation and statistical modelling meet in one task.

Cross-paper transfer task 3: vectors + comprehension

An article introduces a vector recurrence. The learner may need to follow unfamiliar notation while using familiar matrix/vector principles.

Train by rewriting the recurrence in words before calculating.

Exact versus approximate answers

Students should know when H640 expects:

• exact fraction;
• surd;
• multiple of (pi);
• decimal approximation;
• statistical estimate.

A calculator display should not automatically replace an exact expression.

Original exactness task

Evaluate:

[ int_0^1 (3x^2+2x),dx. ]

[ =[x^3+x^2]_0^1=2. ]

An exact integer result needs no decimal approximation.

Numerical-method error control

If a numerical method produces a root approximation, students should know:

• the original equation;
• the interval containing the root;
• the stopping accuracy;
• whether multiple roots might exist.

Original sign-change task

For:

[ f(x)=x^3-x-2, ]

[ f(1)=-2,qquad f(2)=4. ]

A continuous function therefore has at least one root between 1 and 2.

The sign change establishes existence, not uniqueness.

Proof clinic

A proof answer should match the quantifier.

Claim: for every integer (n), (n^2-n) is even.

[ n^2-n=n(n-1). ]

Two consecutive integers include an even one, so the product is even.

Testing several integers is not the proof.

Parameter clinic

Consider:

[ f_a(x)=x^2+2ax+5. ]

The discriminant is:

[ Delta=4a^2-20. ]

Two distinct real roots occur when:

[ 4a^2-20>0, ]

so:

[ |a|>sqrt5. ]

Parameter problems are valuable because they force students to reason about families rather than one numerical instance.

Statistics clinic: conditional probability

If:

[ P(A)=0.6,quad P(B)=0.5,quad P(Acap B)=0.3, ]

then:

[ P(A|B)=rac{0.3}{0.5}=0.6. ]

Here (P(A|B)=P(A)), which is consistent with independence.

The interpretation matters as much as the arithmetic.

Mechanics clinic: dimensional check

Suppose a student derives:

[ v^2=u^2+2as. ]

Both sides have units of length squared over time squared.

Dimensional consistency does not prove the equation is correctly applied, but inconsistency can expose an error immediately.

LDS clinic: outlier sensitivity

Given:

[ 12,13,13,14,58, ]

the mean is 22 while the median is 13.

A student should say that one extreme observation has a strong effect on the mean. This habit transfers directly into large-data-set interpretation.

Comprehension clinic: theorem versus example

If an article gives one numerical case after stating a theorem, the example illustrates the theorem. It does not establish the theorem.

Students should annotate these two roles differently.

Timed Paper 01 strategy

First secure questions with clear routes. For longer Mechanics problems, leave enough space to return. Preserve diagrams and intermediate equations because they may remain useful even if the route changes.

Timed Paper 02 strategy

When an LDS context appears, do not spend time trying to recall every raw value. Recover the statistical relationship from the supplied information and your contextual familiarity.

Timed Paper 03 strategy

Read the comprehension article for structure before chasing details. Definitions and named results are high-value anchors. Then use question wording to return to the relevant part of the text.

Three-paper weekly rotation

A strong weekly schedule can include:

• one Pure/Mechanics session;
• one Pure/Statistics + LDS session;
• one Pure/Comprehension session;
• one mixed retrieval session;
• one timed mini-paper;
• one correction and transfer session.

Do not let LDS work displace Statistics fundamentals

Familiarity with the dataset is useful only when the learner already understands:

• sampling;
• correlation;
• distributions;
• hypothesis tests;
• summary statistics;
• modelling.

The dataset is a context for Statistics, not a replacement for it.

Do not let comprehension work displace Pure Mathematics

Reading skill cannot rescue weak algebra. The article may explain a method, but the student must still manipulate expressions accurately.

Do not let Mechanics formula recall displace modelling

A formula such as:

[ v=u+at ]

is valid under constant acceleration. Always check the condition.

Independent checking routes

Use:

• substitution;
• graph;
• units;
• limiting case;
• alternative method;
• estimate;
• contextual boundary.

The best check targets a different failure mode from the original method.

Original model-validation task

A linear regression predicts a negative waiting time when extrapolated far outside the observed data range. The calculation may be correct. The model is being used outside a sensible domain.

H640 students should distinguish solving the model from validating the model.

Original Mechanics validation task

A calculation gives a coefficient of friction (-0.2). Since a standard coefficient of friction in this model should not be negative, the result signals a modelling or sign error.

Original Comprehension validation task

An article defines a quantity only for (x>0). A later candidate formula produces (x=-2). The formula must be checked against the article’s original domain before it can be accepted.

Revision evidence should be local

Do not infer readiness from how many pages of notes have been completed. Use:

• timed marks;
• error categories;
• fresh transfer;
• ability to explain method choice;
• stability across all three components.

H640 release test

Before the final revision phase, the learner should complete:

1. one unfamiliar Pure/Mechanics problem;
2. one LDS-based statistical interpretation;
3. one short mathematical comprehension passage;
4. one proof/parameter problem;
5. one independent check of a calculator or model output.

If the student succeeds only when the topic is named, the learning is still fragile.

Final current-year control

For June 2027 H640/02, keep LDS 9 on the front of the statistics folder. For June 2028 onward, switch H640 to LDS 10 only when preparing that later cohort. OCR’s September 2026 update is the current control point for this transition.

Final H640 standard

A release-ready H640 student can move from Pure Mathematics into Mechanics, Statistics or unfamiliar exposition without losing the mathematical object.

The durable loop is:

structure → model/representation → execution → interpretation → independent check → fresh transfer.

H640 Release Check: One Pure Idea, One Model, One Reading Task

The final H640 rehearsal should prove that the three components now behave like one Mathematics system. Give the learner one Pure problem, one applied modelling problem and one short mathematical-comprehension passage in the same session.

For Pure Mathematics, use a function that requires more than a derivative. For example:

[ f(x)=x^3-6x^2+9x. ]

[ f'(x)=3(x-1)(x-3). ]

The learner should identify the stationary points and explain the sign of (f’) across the intervals, not stop at differentiation.

For Mechanics or Statistics, require the model before the formula. A Mechanics answer should name forces, direction and assumptions. A Statistics answer should state the distribution or hypothesis structure before calculator work.

For Comprehension, provide a short unfamiliar mathematical argument and ask the learner to identify one definition, one assumption and one conclusion, then apply the idea to a changed case. This checks whether mathematical reading has become independent of familiar textbook wording.

For June 2027, the statistical context must remain aligned with OCR’s current rule: H640/02 uses Large Data Set 9. Large Data Set 10 applies to H640/02 from June 2028 onwards. Do not merge the AS and A Level transition dates. citeturn573627search0turn573627search1

Release standard: secure Pure foundations, model-first applied Mathematics, current LDS ownership, recoverable comprehension reasoning, timed control and successful fresh transfer.

H640 Final Floor: Current Dataset, Independent Mathematics, Fresh Transfer

The final H640 check should use the current cohort label on every Statistics resource. For June 2027, H640/02 uses Large Data Set 9; from June 2028, OCR moves H640/02 to Large Data Set 10. That year boundary should stay visible on revision files so current administrative information never becomes detached from the mathematics.

Then require one fresh three-part rehearsal: a Pure-to-Mechanics problem in which the learner states the model before equations; a Statistics problem in which the learner interprets a distribution or hypothesis result in context; and a short mathematical passage in which definitions, assumptions and conclusions must be recovered before calculation.

The learner should also perform an independent check using a different representation—substitution, graph, units, limiting case or estimate. A correct line of algebra is not enough if the original model or domain has been lost.

Release standard: current OCR cohort control, one mathematical operating system across all three H640 components, visible reasoning, independent checking and successful fresh transfer.