IB Mathematics: Applications and Interpretation | AI SL & HL Guide

IB Mathematics: Applications and Interpretation | Current AI SL & HL, Modelling, Technology, Statistics and the 2029 Transition

IB Mathematics: Applications and Interpretation is a Diploma Programme Mathematics course built around the use of mathematics to describe, analyse and make decisions about real situations. It is not “Mathematics with less algebra”, and it is not simply “the statistics course”. AI still depends on algebra, functions, probability, calculus and exact reasoning. Its distinctive demand is that those mathematical objects are frequently used inside models, data, technology-rich investigations and interpretation.

The first control is version accuracy. The current AI course remains the live Diploma Programme route through its final assessment in November 2028. The revised AI course begins first teaching in August 2027 and has first assessment in May 2029. That creates an overlap period in which two teaching generations can exist at the same time. A student must prepare for the course guide attached to their cohort.

1. Current AI and revised AI must be separated

For current-course students, use the present Applications and Interpretation guide and current specimen/past-paper structure. For students beginning the revised course in August 2027, use the first-assessment-2029 materials as OxfordAQA-style version control is not relevant here—the IB itself is the source of truth.

The current course has its final assessment in November 2028. The revised course is first assessed in May 2029. A resource should therefore carry at least:

• AI SL or AI HL;
• current course or first-assessment-2029 course;
• school examination year;
• calculator/technology expectations;
• official IB source checked date.

2. AI is a modelling course—but modelling is not decoration

A model is a mathematical representation of a situation. Building one requires choices:

• What quantities matter?
• Which quantities are inputs and outputs?
• What assumptions are being made?
• Which relationship is plausible?
• What domain is meaningful?
• How should the parameters be interpreted?
• How will the model be checked?

The equation is only one part of the model. The assumptions and interpretation are equally important.

3. Original modelling task: taxi fare

A taxi fare is modelled by:

[ C=4.50+1.80d, ]

where (C) is cost in dollars and (d) is distance in kilometres.

The intercept 4.50 represents the fixed starting charge. The coefficient 1.80 is the rate per kilometre.

For (d=8):

[ C=4.50+1.80(8)=18.90. ]

A stronger AI question asks whether the model remains realistic for very long trips, whether waiting charges are ignored, whether distance is charged continuously or in blocks, and how the model might be refined.

4. Model evaluation is a mathematical act

If the model predicts impossible outputs, that matters. If residuals show a pattern, that matters. If the parameter values do not make sense in context, that matters.

AI students should routinely ask:

• Does the result have the correct unit?
• Is the magnitude plausible?
• Is extrapolation justified?
• What assumptions are strongest?
• What data would test the model?

5. Original modelling task: linear versus exponential growth

Model A:

[ A(t)=500+40t. ]

Model B:

[ B(t)=500(1.05)^t. ]

Model A adds a fixed amount each period. Model B grows by a fixed percentage.

Students should compare:

• initial values;
• growth mechanism;
• short-term difference;
• long-term behaviour;
• appropriate contexts.

The task is not simply “which value is bigger at (t=10)?” It is understanding why the models behave differently.

6. Technology is required, but technology does not own the Mathematics

The current AI course explicitly expects students to use technology and become proficient with graphic display calculators. Current AI external papers are technology-required. That creates a special responsibility: the learner must distinguish what the tool calculated from what the learner decided.

A good technology workflow is:

predict → enter → inspect → interpret → verify.

7. Predict before graphing

Suppose:

[ f(x)=x^2-4x-5. ]

Before graphing, factor:

[ f(x)=(x-5)(x+1). ]

So the graph should cross at (x=-1) and (x=5). A graphing calculator can confirm the shape and turning point, but algebra already protects the root count.

8. A graphing window can hide mathematics

If:

[ f(x)=(x-2)(x-50), ]

a standard window may show the root near 2 and hide the root at 50.

AI students should never interpret “not visible on the screen” as “does not exist” without checking the domain and window.

9. Numerical answers should be identified as numerical

A solver can produce:

[ xapprox1.41421356. ]

If the underlying equation is (x^2=2), the exact answer is:

[ x=pmsqrt2. ]

Technology can approximate; it does not automatically establish the exact form.

10. Statistics is central because AI asks what evidence means

Statistics in AI should not be taught as a sequence of calculator menus. Students need to understand:

• sampling;
• bias;
• centre and spread;
• correlation;
• regression;
• probability distributions;
• hypothesis testing at the appropriate level;
• interpretation in context.

11. Original statistics task: same mean, different spread

Dataset A:

[ 9,10,10,10,11. ]

Dataset B:

[ 0,5,10,15,20. ]

Both have mean 10. Dataset B has much greater spread.

A student who reports only “both averages are 10” has lost important information.

12. Median is robust, not universally superior

Consider:

[ 12,13,13,14,48. ]

Mean = 20. Median = 13.

The median better represents the central cluster. But if the 48-minute delay matters operationally, the mean captures part of the total burden.

AI students should choose statistics according to the reader job.

13. Correlation is not causation

If a scatter plot shows that students with more sleep tend to have higher test scores, the association does not prove that increasing sleep alone caused the higher scores.

Possible confounders include:

• health;
• study habits;
• stress;
• age;
• school schedule.

A strong AI answer states what the data support and what they do not.

14. Regression should be bounded by evidence

A regression line can be useful inside the data range. Extrapolation outside the observed range can become unreliable.

Ask:

• What range was observed?
• Is the relationship plausibly linear?
• Are there influential outliers?
• Does the context support extrapolation?

15. Original regression interpretation task

Suppose a regression model is:

[ y=12+3.4x. ]

If (x) is weekly study hours and (y) is a score, the gradient 3.4 means the model predicts a 3.4-point score increase per additional study hour.

It does not prove that one extra hour causes exactly 3.4 more marks for every student.

16. Probability starts with the model

Before selecting a distribution, state the assumptions.

For a binomial model:

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

17. Original binomial task

If:

[ Xsim B(15,0.2), ]

then:

[ P(X=3)=inom{15}{3}(0.2)^3(0.8)^{12}. ]

A calculator can evaluate the expression. The student must justify the distribution.

18. “At least one” often suggests a complement

If ten independent trials each succeed with probability 0.3:

[ P( ext{at least one success}) = 1-P( ext{zero successes}) ]

[ =1-(0.7)^{10}. ]

The complement is shorter than summing ten separate cases.

19. Conditional probability changes the sample space

If:

[ P(A)=0.4,quad P(B)=0.5,quad P(Acap B)=0.2, ]

then:

[ P(A|B)=rac{0.2}{0.5}=0.4. ]

The condition (B) changes the relevant denominator.

20. Functions remain a mathematical spine

AI modelling depends heavily on functions. Students should understand:

• linear;
• quadratic;
• exponential;
• logarithmic where applicable to the current course;
• trigonometric;
• piecewise;
• regression models;
• transformations and parameters.

21. Original function task: break-even model

Company A charges:

[ A(x)=40+2x. ]

Company B charges:

[ B(x)=10+5x. ]

Break-even:

[ 40+2x=10+5x. ]

[ 30=3x,quad x=10. ]

At 10 units, costs are equal. The student should then identify which model is cheaper below and above 10.

22. Piecewise models are useful for real systems

A parking fee may charge:

[ C(t)=5 ]

for the first hour and then:

[ C(t)=5+3(t-1) ]

for later hours.

Piecewise structure can be more realistic than forcing one equation across every condition.

23. Calculus in AI should remain interpretive

Calculus is used to understand change, accumulation and optimisation. A derivative is not only a symbolic operation.

For:

[ f(x)=x^2-6x+10, ]

[ f'(x)=2x-6. ]

Stationary point:

[ x=3. ]

Completing the square:

[ f(x)=(x-3)^2+1 ]

confirms that the point is a minimum.

24. Original optimisation task

A rectangle has perimeter 40. Let one side be (x), so the other is (20-x).

Area:

[ A(x)=x(20-x)=20x-x^2. ]

[ A'(x)=20-2x. ]

Maximum when (x=10). The maximum-area rectangle is a square.

25. Rates of change must return to units

If distance is measured in kilometres and time in hours, derivative of distance with respect to time has units km/h.

AI interpretation is incomplete if the derivative is given as a number without a contextual unit.

26. Integration should be read as accumulation

If:

[ r(t)=4t+2 ]

is a rate, accumulated quantity from (t=0) to (t=3) is:

[ int_0^3(4t+2),dt = [2t^2+2t]_0^3 =24. ]

The integral has meaning because the integrand has rate units.

27. Geometry and trigonometry remain relevant

AI students may use:

• measurement;
• trigonometry;
• bearings;
• 3D geometry;
• vectors at HL/current syllabus where relevant;
• geometric models in real contexts.

Technology does not remove the need to understand the diagram.

28. Original trigonometric modelling task

A building is observed from 50 m away at an angle of elevation (38^circ).

Ignoring observer height:

[ h=50 an38^circ. ]

A stronger question asks how the answer changes if the measuring instrument is 1.6 m above the ground.

29. Units can break an otherwise correct model

If speed is in km/h and time is in minutes, convert before multiplying.

For 72 km/h over 25 minutes:

[ 25 ext{ min}=rac{25}{60} ext{ h}. ]

Distance:

[ 72cdotrac{25}{60}=30 ext{ km}. ]

30. AI SL and AI HL should not be chosen by reputation

Course choice should consider:

• student strengths;
• intended university subjects;
• institution-specific prerequisites;
• ability to sustain HL algebra and analysis;
• comfort with statistics and modelling;
• technology fluency.

Admissions requirements can differ by university and programme. Check them directly.

31. AI is not universally accepted for every quantitative degree

Some highly mathematical university routes specify AA HL or another particular Mathematics preparation. Other programmes accept AI HL, and some accept AI SL depending on the programme.

Do not choose AI based only on internal school advice without checking target admissions requirements.

32. The current AI assessment model

Under the current course, external assessment carries 80% and the mathematical exploration carries 20%.

At current SL, Paper 1 and Paper 2 are technology-required papers weighted equally within the external assessment. At current HL, Paper 1 and Paper 2 are technology-required, and Paper 3 adds two compulsory extended problem-solving questions; the exploration remains 20%.

Students should verify exact current paper durations and marks against the current IB guide for their cohort.

33. Technology required does not mean technology always useful

A graphing calculator can solve:

[ 3x+5=20. ]

But mental algebra is faster:

[ x=5. ]

Use technology when it increases insight or efficiency, not because it is available.

34. Paper 1 preparation should include short-response efficiency

Current AI Paper 1 includes shorter-response work. Students should:

• identify the mathematical object quickly;
• enter technology accurately;
• show enough method;
• interpret output;
• move on.

35. Paper 2 preparation should include sustained response

Extended-response questions can combine:

• model construction;
• technology;
• statistics;
• calculus;
• interpretation;
• evaluation.

The learner must carry assumptions and variables across several parts.

36. HL Paper 3 should be trained as problem solving

Paper 3 is not simply “harder Paper 2”. It uses extended problem-solving questions. The student needs to:

• explore;
• make intermediate deductions;
• use technology intelligently;
• adapt when the first route fails;
• communicate a sustained argument.

37. Paper 3 preparation should include dead-end recovery

When the first method stalls:

1. record what has been established;
2. change representation;
3. inspect data/graph numerically;
4. look for a pattern;
5. return to the algebra.

Do not erase useful partial structure.

38. The Mathematical Exploration is not an extended calculator report

A good exploration has:

• a clear mathematical question;
• appropriate assumptions;
• selected tools and techniques;
• mathematical development;
• interpretation;
• reflection and refinement.

The purpose is inquiry, not decoration.

39. Exploration question design

Weak:

“Basketball statistics.”

Stronger:

“How well does a logistic model describe the relationship between shot distance and scoring probability in a selected dataset?”

The stronger version defines a relationship and gives Mathematics something to do.

40. Exploration assumptions should be explicit

If a model assumes independence, constant rate, linearity, normality or representative sampling, state it.

An assumption hidden in software output is still an assumption.

41. Exploration technology should serve inquiry

Useful tools can include:

• spreadsheet;
• graphing software;
• GDC;
• statistical package;
• dynamic geometry.

The student should explain why the tool helps answer the mathematical question.

42. Exploration interpretation should return to the problem

A regression coefficient or (R^2) value is not the conclusion. Explain what it means in context, where the model works and where it may fail.

43. Exploration refinement is stronger than pretending perfection

If the model is weak:

• change the model;
• segment the data;
• remove/justify an outlier;
• change variables;
• compare alternatives.

Reflection on limitations is mathematical maturity.

44. Current-to-2029 transition

The revised AI course begins first teaching in August 2027 and first assessment in May 2029. The IB says the update is a refinement of the existing course, with a continued focus on modelling, statistical analysis, technology, problem solving and inquiry.

Current-course students whose final assessment is before May 2029 should not silently adopt revised-course mark allocations or topic deletions.

45. Revised-course content changes

The IB’s April 2026 update says the revised course removes some content rather than adding new material. The page identifies reductions at SL and HL and a refined internal-assessment model. Teachers should map those changes against the official first-assessment-2029 guide rather than rewriting the entire course.

46. Revised assessment model should be version-labelled

The IB states that the external assessment model is broadly unchanged while some mark totals are reduced for the new course. Therefore:

• current-course mock papers should remain current-course;
• first-assessment-2029 mock papers should use the revised guide;
• question banks should be labelled by generation.

47. AI vs AA should not become a status contest

AA emphasises analytical development and symbolic Mathematics more strongly. AI emphasises applications, modelling, statistics, interpretation and technology more strongly.

Both can be demanding. The right course depends on the learner and intended destination.

48. AI learner failure mode: calculator dependence

Symptom: the learner can produce regression output but cannot explain slope, residuals or domain.

Repair:

1. predict qualitative behaviour;
2. run the technology;
3. explain parameters in words;
4. check one point manually;
5. state a limitation.

49. Failure mode: weak algebra under the model

Symptom: the learner understands the context but cannot rearrange or solve the model.

Repair the algebraic object outside the full AI context, then return.

50. Failure mode: interpretation without evidence

Symptom: the student writes a plausible story not supported by the calculated result.

Repair: every interpretation sentence must point back to a number, graph, parameter, interval or statistical result.

51. Failure mode: correlation becomes cause

Repair: require the sentence:

The observed association is consistent with…, but does not by itself establish that X causes Y.

52. Failure mode: model used beyond its domain

Repair: write the observed domain and prediction domain before extrapolating.

53. Failure mode: IA topic too broad

Repair: convert a subject area into a mathematical question with variables, a relationship and an intended outcome.

54. A six-week AI SL revision cycle

Week 1: diagnostic

Algebra, functions, statistics, probability, technology and interpretation.

Week 2: functions/models

Linear, exponential, regression and graphical reasoning.

Week 3: statistics/probability

Distribution choice, correlation, inference and interpretation.

Week 4: calculus/geometry

Rates, accumulation, optimisation and applied geometry.

Week 5: timed papers

Mixed short and extended responses.

Week 6: fresh transfer

Repair recurring errors and retest in changed contexts.

55. A six-week AI HL revision cycle

Week 1: SL foundation audit.

Week 2: HL functions/calculus.

Week 3: HL probability/statistics.

Week 4: technology-rich problem solving.

Week 5: Paper 3-style sustained tasks.

Week 6: full-course integration and exploration maintenance.

56. AI revision should use current official papers first

Use IB sample/specimen and school-authorised past examination materials to calibrate style. Third-party questions are useful after the learner understands the official demand.

57. Error ledger

58. One-week repair loop

Day 1: identify first failed decision.

Day 2: repair the mathematical object.

Day 3: practise with technology.

Day 4: change context.

Day 5: delayed retrieval.

Day 6: timed fresh problem.

Day 7: decide stable / fragile / blocked.

59. University-pathway check

Before final course selection or university application:

• check exact programme requirement;
• check SL/HL requirement;
• check AI/AA acceptance;
• check any score prerequisite;
• use the university’s current admissions page.

60. Parent check

Ask:

• AI SL or HL?
• current course or revised 2029 course?
• what kind of error repeats?
• is the learner interpreting calculator output?
• is the exploration mathematically focused?

61. Tutor check

Do not teach AI as “statistics plus calculator”. Map every weakness to the underlying object: algebra, functions, probability, calculus, modelling or interpretation.

62. Current official-source control

Use the IB Diploma Programme Mathematics page and current Applications and Interpretation guide for the live course. The IB confirms that the present courses have final assessment in November 2028 and the revised course is first taught in August 2027 with first assessment in May 2029.

Official routes: IB DP Mathematics and IB Mathematics: applications and interpretation updates.

63. What not to claim

• AI is universally easier than AA.
• AI SL is accepted for every university programme.
• calculator output is mathematical reasoning.
• correlation proves causation.
• one regression works outside its evidence range automatically.
• current-course students should use revised 2029 mark totals early.

64. Final AI release test

Use one unseen modelling problem, one statistics problem, one function/calculus problem and one technology-rich investigation question. The learner should:

• define variables;
• state assumptions;
• choose mathematics;
• use technology correctly;
• interpret the result;
• evaluate limitations;
• check reasonableness.

65. The durable AI model

The complete learning loop is:

context → mathematical question → assumptions → model → computation/technology → interpretation → validation → refinement → communication.

That is the centre of Applications and Interpretation. It remains useful through the 2029 curriculum revision because it is a mathematical way of thinking, not merely an examination format.

IB AI Floor Closure: Paper Strategy, Exploration Quality, Fresh Transfer and University Readiness

The final AI preparation layer should connect examination performance, mathematical modelling, technology, the exploration and post-IB progression. A student is not fully prepared because they have completed a large question bank. They are prepared when they can identify the structure of a new problem, select a model, use technology intelligently, interpret the output and explain what the result does and does not mean.

1. Build one mixed AI session each week

A strong weekly session can contain four problems:

1. a modelling/function question;
2. a statistics/probability question;
3. a calculus or geometry question;
4. a technology-rich interpretation problem.

Do not label the topic before the learner begins.

2. Original mixed modelling problem

A gym charges a fixed monthly fee of $30 plus $4 per visit.

[ C(v)=30+4v. ]

Another gym charges $70 flat.

Break-even:

[ 30+4v=70. ]

[ v=10. ]

For fewer than 10 visits, the first plan is cheaper. For more than 10 visits, the flat plan is cheaper.

Then ask: what assumptions are hidden? Membership terms, whole visits, additional fees and time period can matter.

3. Original nonlinear modelling problem

A medicine concentration is modelled by:

[ C(t)=80e^{-0.25t}. ]

Find the half-life:

[ 40=80e^{-0.25t}. ]

[ 0.5=e^{-0.25t}. ]

[ t=rac{ln2}{0.25}. ]

The student should explain that exponential decay assumes a constant proportional rate of change.

4. Sensitivity matters

If a model depends strongly on one parameter, small estimation errors can produce large prediction differences.

For:

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

a small change in (k) compounds over time.

AI students should test whether conclusions remain stable under plausible parameter variation.

5. Residuals should be interpreted

A regression line may look good overall while residuals reveal curvature.

If residuals form a systematic pattern, a linear model may be inappropriate even if the correlation coefficient is large.

6. Original residual task

Suppose predicted values are:

[ 10, 15, 20, 25 ]

and observed values are:

[ 11, 16, 18, 22. ]

Residuals:

[ 1, 1, -2, -3. ]

The increasingly negative residuals suggest the model is beginning to overpredict.

7. Sampling should be treated as a design choice

A large sample can still be biased. A small random sample can still be noisy. Students should distinguish sample size from sampling quality.

Useful questions:

• Who is in the population?
• How was the sample selected?
• Who is missing?
• What non-response bias may exist?

8. Original sampling critique

A school asks students in the library whether they enjoy reading and concludes that 85% of the whole school enjoys reading.

The sample may be biased because library users are not representative of all students.

9. Technology should be tested against a hand-check

If a regression gives slope 2.4, choose two nearby data points and estimate the slope manually. If the rough result is dramatically different, inspect the data entry or model.

10. AI communication should identify evidence level

Compare:

Too strong: “Exercise causes higher grades.”

Better: “In this sample, exercise frequency is positively associated with grades; the data do not by themselves establish causation.”

11. Paper strategy should use a stop-loss rule

In a technology-rich exam, it is easy to spend too long troubleshooting a graph or solver. If the method has become unclear:

1. record what is known;
2. save values if useful;
3. move on;
4. return with fresh attention.

12. Paper 1 efficiency drill

Use short sets where the learner must decide quickly whether a question needs:

• manual algebra;
• graphing;
• solver;
• statistical function;
• numerical integration;
• no technology at all.

13. Paper 2 sustained reasoning drill

Use one extended problem with several parts. Require the learner to keep:

• variables;
• assumptions;
• units;
• model meaning

consistent across all parts.

14. Paper 3 resilience drill for HL

Give an unfamiliar problem where no obvious formula is visible. The learner should:

• explore numerically;
• look for pattern;
• form a conjecture;
• test it;
• return to symbolic Mathematics;
• write a coherent conclusion.

15. Exploration topic selection should pass four gates

A proposed exploration should have:

1. a mathematical question;
2. enough mathematical depth;
3. available data or structure;
4. scope that fits the task.

Interesting context without mathematical depth is not enough.

16. Exploration example: transport model

Possible question:

How well can a piecewise linear model describe travel time on a selected route across different times of day?

Possible Mathematics:

• piecewise functions;
• regression;
• residuals;
• comparison of means/variation;
• model evaluation.

17. Exploration example: sports probability

Possible question:

Can a binomial or logistic model describe scoring success under selected conditions?

Students should justify independence assumptions and explain where they fail.

18. Exploration example: growth model

Compare linear, exponential and logistic models for a bounded growth process.

The strongest section may be the comparison and evaluation, not the most complicated formula.

19. The exploration should show ownership

The student should be able to explain:

• why the question matters;
• why the mathematics was selected;
• what the parameters mean;
• what the result means;
• what should be improved.

20. AI technology fluency should be platform-independent

The learner should know the mathematical job even if the button sequence changes.

Jobs include:

• plotting;
• regression;
• solving;
• probability calculation;
• summary statistics;
• numerical derivatives/integrals where appropriate.

21. Do not memorise menu paths only

If a learner forgets the exact GDC menu, they should still know:

I need a binomial cumulative probability with these parameters.

That knowledge can be transferred across devices.

22. Algebra should be maintained weekly

AI students sometimes lose algebraic fluency because technology handles too much symbolic work.

Keep weekly practice in:

• rearrangement;
• fractions;
• indices;
• equations;
• function notation.

23. Original algebra-under-model task

If:

[ P=rac{A}{1+e^{-k(t-c)}}, ]

ask the learner to explain the roles of (A,k,c) before any numerical fitting.

Parameter interpretation is part of model understanding.

24. Calculus should be linked to shape

A positive derivative means increasing behaviour locally. A zero derivative is a candidate stationary point. A second derivative can help classify curvature or stationary points.

Use graphs to verify symbolic reasoning.

25. Probability should be linked to decision making

A probability of 0.08 may be small, but whether it matters depends on consequences.

Risk combines probability with impact.

AI students should distinguish a mathematical probability from a policy decision.

26. Financial Mathematics should stay model-based

Compound interest:

[ A=P(1+r)^n. ]

The formula assumes fixed periodic rate and compounding structure.

Real financial products may include fees, variable rates or different compounding intervals.

27. Original finance task

[ P=5000,quad r=0.04,quad n=6. ]

[ A=5000(1.04)^6. ]

Then ask how the answer changes if the rate varies each year.

28. Time-series thinking should preserve order

Two datasets can have identical means but opposite trends.

A:

[ 2,4,6,8,10. ]

B:

[ 10,8,6,4,2. ]

Same mean, opposite direction.

29. Simulation can reveal probability behaviour

Technology can simulate repeated random trials.

Students should compare simulated relative frequency with theoretical probability and explain sampling variation.

30. Simulation is not proof

10,000 simulated successes do not prove a theoretical identity.

Simulation is evidence about behaviour under a model.

31. AI should use units consistently

Models fail when:

• hours and minutes are mixed;
• metres and kilometres are mixed;
• percentages are entered as whole numbers;
• currency conversions are omitted.

32. Dimensional sense is a powerful check

If a calculated “speed” has units of km²/h, inspect the formula.

Units can expose model errors before numerical checking.

33. Current-course paper practice should remain current-course

Students assessed before the revised first assessment in May 2029 should use current-course paper structures and current-course mark totals.

The revised subject brief can support transition planning, not current-paper simulation.

34. Revised-course learners need a specification-diff map

Schools beginning AI in August 2027 should compare:

• content removed;
• assessment marks changed;
• IA criteria changed;
• unchanged mathematical ideas;
• unchanged modelling habits.

35. Course-generation labels should be visible

Use:

• AI current course — final assessment Nov 2028;
• AI revised course — first assessment May 2029.

This avoids silent cross-generation mixing.

36. Admissions checks should be dated

University policies change. Record:

• institution;
• programme;
• AI/AA requirement;
• SL/HL requirement;
• minimum grade;
• date checked.

37. AI HL is not automatically sufficient for every STEM route

Some programmes require AA HL or another specific preparation. Other programmes accept AI HL.

The student should check the exact destination rather than rely on general reputation.

38. AI SL can be appropriate for many routes

But appropriateness depends on destination. The decision should combine learner fit with admissions requirements.

39. One final four-problem release set

Problem 1: modelling and assumptions.

Problem 2: statistics and interpretation.

Problem 3: calculus/function and technology.

Problem 4: unfamiliar sustained problem solving.

40. Release scoring should include process

For each problem, assess:

• problem specification;
• mathematical choice;
• computation;
• technology use;
• interpretation;
• checking;
• communication.

41. Final current/revised audit

• correct course generation;
• correct SL/HL;
• correct paper format;
• correct IA criteria;
• official IB guide used;
• university requirements checked separately.

42. Release standard

An AI learner is release-ready when they can turn a context into Mathematics, use technology without surrendering reasoning, interpret data critically, evaluate a model, communicate limitations and transfer the method to an unfamiliar situation.

Final rule: model deliberately, compute intelligently, interpret honestly, refine when evidence demands it.

IB AI Final Evidence Transfer | From Calculator Output to Defensible Mathematical Judgment

The last gap in AI preparation is evidence calibration. The student must be able to distinguish a result produced by a tool from a conclusion supported by Mathematics. That distinction applies to regression, probability, optimisation, numerical solutions and the exploration.

1. A numerical result needs a mathematical sentence

If technology gives a regression line

[ y=18.4+2.7x, ]

the learner should not stop at the equation. They should state what 18.4 and 2.7 mean in the context, identify the observed data range, inspect residual behaviour and avoid treating the line as causal evidence.

2. A probability output needs model ownership

If a GDC returns

[ P(Xle4)=0.781, ]

the answer is incomplete unless the learner can state the distribution, its parameters, why the distribution is appropriate and what “(Xle4)” means in the real situation.

3. A numerical solver needs an original-equation check

When a solver reports a root, substitute it back into the original model where practical. This is especially important if the equation was transformed, squared, approximated or restricted to a particular domain.

4. A model should survive a changed condition

After solving a linear-growth problem, change the rate. After fitting a regression, remove one influential observation and compare. After choosing a binomial model, change the problem so probabilities are no longer constant. The learner should recognise when the old model still works and when it breaks.

5. Original release task

A town records population (P) over time and fits:

[ P(t)=25000(1.018)^t. ]

Ask the learner to:

1. interpret 25000 and 1.018;
2. predict population after 8 years;
3. solve when the model reaches 30000;
4. state one assumption;
5. state one reason long-term extrapolation may fail;
6. explain how new data could be used to validate or revise the model.

This single task joins functions, logarithms, technology, interpretation and model evaluation.

6. Final IA transfer check

Before submission, the student should be able to close the document and answer verbally:

• What was my mathematical question?
• Why did I choose this Mathematics?
• What assumption matters most?
• What result is strongest?
• What limitation changes how the result should be used?
• What would I refine if I continued?

If these answers are unclear, the exploration may be technically busy but mathematically weak.

7. Current-course release control

Students assessed through November 2028 remain on the current AI course. Students beginning the revised course from August 2027 and assessed from May 2029 should use the revised course materials. Keep those generations visible on every mock, IA guide and topic plan.

8. Release standard

The AI learner is ready when technology, modelling, statistics and interpretation work as one system: the learner chooses the Mathematics, the tool performs appropriate computation, the result is checked, and the final statement stays no stronger than the evidence.

Final release: model clearly, compute correctly, interpret cautiously, validate independently, transfer to a new context.