Past-year question, shown for revision (SEAB / Cambridge). Hover or tap a chip to change it.
Every mark of 27 yearly papers (1997–2025), sorted into 22 topics, re-checked against the real papers, and read for what it says about the paper coming in 2026.
Averages use the 2023–25 papers (the current syllabus) unless stated; marks are out of 180.
Together they averaged 36 marks (20%) in 2023–25. Algebra alone has more than doubled: 9 marks in 1997–2006 → 18 now.
Algebra, equations, trig/Pythagoras, mensuration, graphs, statistics and the real-world problem appeared every single year of 2016–25 and took 58% of the marks in 2023–25.
Since 2016 the last Paper 2 question is one integrated real-world problem (9–11 marks): money, graphs, a model, then a justified decision.
Sets (3 marks) and matrices (4 marks) have appeared in all ten recent papers. Short, predictable, and easy marks.
Speed & rates fell from 7.6 to 2.7 marks; construction from 5.0 to 1.0; circle properties and similarity are drifting down.
Sequences scored 15, 13, 2, 0 in 2019, 2023, 2024, 2025 (about 6, 10, 2, 0 under an independent re-coding: still wild). The data show no rotation: one year’s mark tells you almost nothing about the next.
Darker = more marks. Read across for a topic’s history; read down for what a whole paper looked like. The right-hand column compares the last ten years with 2009–15.
more marks than before
less than they used to be
same weight, year after year
swing a lot: never bank on them
Bars, left to right:1997–062009–152016–222023–25
Marks for sequences, 2016–25. After two quiet years it is tempting to expect a big sequences question in 2026. The numbers do not support that.
The real value sits inside the chance range: there is no sign of rotation. A topic that was light last year was not more likely to be heavy this year, and vice versa. Over the longer run (2009–25) topics are slightly sticky, because real shifts such as algebra’s climb persist.
So: treat gaps as noise. Prepare for sequences because it is on the syllabus and can score big, not because it is owed.
Paper 1 only: financial arithmetic (about 5 marks) and constructions. They live in the short questions.
Paper 2 only: the real-world problem (10), and coordinate geometry is almost entirely Paper 2 (about 7 of 8 marks).
Paper 1 heavy: algebra (14 of 18), trigonometry, statistics and vectors.
Paper 2 heavy: graphs (10 of 14), equations and modelling, probability and mensuration.
Hover over a chip (or tap it) to see a real past-year example of that question type.
Counted from the concept label of every coded question part, so it shows how the question is asked, not just which topic it belongs to. “Explain / justify / comment” appears every year: reasoning is marks, not decoration.
Eight papers (2009, 2012, 2016, 2019, 2022, 2023, 2024, 2025: the 4016, 4048 and 4052 syllabuses) were re-coded from scratch by independent reviewers who never saw the analyst’s ledger, reading every printed mark from the page images, then compared. The 1997–2006 papers were not re-checked. On average 94% of marks landed in the same topic. Every paper reconciled to its printed total.
Gaps are the re-coded marks minus the ledger’s, for the three biggest topic disagreements of 3+ marks.
1. Boundary calls, not errors. Nearly every disagreement is a part that sits between two topics: an algebraic proof that ends in Pythagoras, a speed problem that is really a quadratic, a number-grid proof (sequence or algebra?). The ledger’s rule (“the skill driving the work”) is sensible and applied consistently.
2. Sequences are less wild than they look. The ledger gives 2019 a 15-mark sequences load; the reviewers code 6 (the number-grid parts go to equations and algebra). The real swing is more like 0–10 than 0–15. The ‘0 in 2025’ and ‘2 in 2024’ are confirmed.
3. In the 2023–25 papers, algebra looks even heavier. The re-coding gives algebra + equations 40 marks (ledger 36) and trigonometry 13 (ledger 16), because proofs that use Pythagoras count as algebra there. The conclusion is the same, only louder: algebra and equations are the single biggest block.
Also: travel-graph (speed) marks come out lower in the re-coding for 2012 and 2016 (by 5 each), so that topic’s fade is, if anything, steeper than shown.
Method: recency-weighted average of 2016–25 (each year counts 80% as much as the next one), rescaled to 180, with the real-world problem fixed near 10. The bar shows the middle 80% of what each topic scored in those ten years. It is a planning guide, not a probability: three papers of the 4052 syllabus are too few to fit anything stricter. It uses the ledger’s coding; on the stricter re-coding expect algebra 2–4 marks higher and trigonometry 2–3 lower. Note on the practice pack: the ChatGPT specimen papers carry more graphs (20 marks vs 13 forecast) and travel graphs (8 vs 3), and fewer ratio, finance, percentage and similarity marks than this forecast, so use them with that tilt in mind.