A forensic decomposition of DSE14 from 2012 to 2024 — the question templates, the 3D geometry traps, and the diagram generation pipeline that makes this topic different from every other.
We parsed every Paper 2 from 2012 to 2024. DSE14 (Trigonometry) is one of the heaviest topics — and the only one that requires 3D spatial reasoning.
7 sub-topics, wildly unequal weight. Two sub-topics account for 25 of 39 questions:
Every DSE14 question falls into one of 4 templates. The first covers Section A (foundation marks); the rest target Section B (the hard half).
The foundation template. Appears in every single year without exception. Usually Q18–Q25 in Section A.
The pattern: Given a figure with 2+ connected right triangles and labeled angles (\(\alpha, \beta, \theta\)), express one ratio (e.g., \(\text{AB}/\text{CD}\)) in terms of trig functions of those angles. Chain the ratios through shared sides.
Tested in: 2012, 2013, 2014, 2015 (4 of 13 years). Has not appeared since 2016.
The toolkit: Six compound-angle conversions: \(\sin(90°\pm\theta)\), \(\cos(90°\pm\theta)\), \(\sin(180°\pm\theta)\), \(\cos(180°\pm\theta)\), \(\sin(270°\pm\theta)\), \(\cos(270°\pm\theta)\). Convert everything to \(\sin\theta\) and \(\cos\theta\), then simplify.
The hardest template in all of DSE Math Paper 2. Appears in 11 of 13 years (missing only 2015). Always Q38–Q41, the last positions in the exam.
The cognitive demand: Read a 3D figure. Identify the correct right triangle(s) hidden within it. Compute side lengths via 3D Pythagoras. Apply the cosine formula to find the target angle. 4–6 steps, and each wrong triangle choice is unrecoverable.
Tested in: 2014, 2022, 2024 (3 of 13 years). Always Q39 in Section B. Low frequency but completely predictable format.
The algorithm: (1) Convert to a quadratic in \(\sin x\) or \(\cos x\). (2) Factor. (3) Solve for each root. (4) Count solutions in \([0°, 360°)\), checking all quadrants.
The 3D trigonometry questions are the most valuable to track — hardest, highest-weighted, and the ones Renee's students struggle with most. Here's every single one:
| Year | Figure Type | Core Task | Key Technique | Difficulty |
|---|---|---|---|---|
| 2012 | Regular tetrahedron | Find dihedral angle ∠AED | Cosine formula on isosceles triangle | ADV |
| 2013 | Regular tetrahedron | Find volume given height | 3D Pythagoras + base area | ADV |
| 2014 | Perpendicular to plane | Find tan∠AEB (foot of perp) | Pythagorean triple + area method | ADV |
| 2015 | No 3D question this year | |||
| 2016 | Rectangular box | Find sin∠PFQ (midpoint + edge point) | 3D Pythagoras + cosine formula, 5 steps | HARD |
| 2017 | Perpendicular planes | Find angle between line and plane | Two Pythagorean triples + projection | ADV |
| 2018 | Cuboid with projection | Find cos∠YBX (projection of X on plane) | 3D Pythagoras chain: XY→YA→YB→XB | ADV |
| 2019 | Point above ground | Find ∠RPS from elevation angles | Cosine formula twice (ground + 3D) | HARD |
| 2020 | Right triangular prism | Find area of △BDP (parametric) | Heron's formula + 3D coordinates | HARD |
| 2022 | Cube (cross-sections) | Compare dihedral angles α vs β | Coordinate geometry + dot product | HARD |
| 2023 | Right pyramid (square base) | Find cosθ (dihedral along edge BV) | Foot of perp + cosine formula | HARD |
| 2024 | Tetrahedron on ground | Find cos∠PRS from angle constraints | All lengths cancel — pure angle reasoning | HARD |
2012–2014: regular solids (symmetric, fewer steps). 2016–2024: irregular solids, parametric constraints, side ratios instead of lengths, dihedral angles. The difficulty is escalating. The cosine formula is THE dominant tool — required in 8 of 11 problems.
Every question is decomposed into structured data. Here are three entries — a 2D ratio chain, a 3D problem with diagram spec, and an identity question:
| Field | DSE17 (Dispersion) | DSE14 (Trig) |
|---|---|---|
| stem | Text only — no figures needed | Often references a figure — requires diagram |
| diagramSpec | N/A — never needed | Present on 11 questions — figure type, vertices, angles, constraints |
| skillsTested | Same schema | Same schema |
| errorTraps | Computational traps (sign errors, formula confusion) | Spatial traps (wrong triangle, wrong projection, wrong angle) |
| toAce | Same schema | Same schema, but often longer (4–6 steps vs 2–3) |
This is the capability DSE17 doesn't need and DSE14 can't do without. Every 3D trig question comes with a figure that students must read. Any question engine for this topic must generate correct, exam-style diagrams.
We evaluated diagram generation on 9 past-paper 3D questions (ground truth from official exam papers):
| Year | Figure | Programmatic | Visual | Issue |
|---|---|---|---|---|
| 2014 Q40 | Perp to plane | 11/11 | PASS | — |
| 2019 Q40 | Point above ground | 10/10 | PASS | — |
| 2024 Q40 | Tetrahedron on ground | 10/10 | PASS | — |
| 2012 Q40 | Regular tetrahedron | 9/9 | PARTIAL | Projection angle slightly off |
| 2016 Q39 | Rectangular box | 9/9 | PARTIAL | Edge proportions inaccurate |
| 2017 Q39 | Perp planes | 11/11 | PARTIAL | Plane intersection unclear |
| 2018 Q41 | Cuboid projection | 8/8 | PARTIAL | X/Y positioning off |
| 2020 Q38 | Triangular prism | 7/7 | PARTIAL | BF line rendered at wrong angle |
| 2022 Q40 | Cube cross-sections | 7/7 | FAIL | Cube shape structurally deformed |
We flipped the approach: generate questions first, then diagrams. The LLM writes exam-style questions with full solutions, then writes geometry-accurate TikZ code. node-tikzjax renders TikZ to SVG. Triple validation: programmatic checks + LLM review + human review.
Here are 3 examples from the 7 generated questions, each with its AI-rendered diagram. See all 7 →
3d-perpendicular-projection.3d-cosine-formula, 3d-figure-decomposition.3d-cosine-formula.The generation pipeline runs entirely without manual code: LLM generates the question and writes geometry-accurate TikZ, node-tikzjax renders to SVG, programmatic validators check structure. 7 figure types covered across 2 improvement passes. Full generation showcase →
These are the 14 irreducible skills we extracted. Every DSE14 question from 2012–2024 can be solved using some combination of these.
| Skill | Level | What It Means | Tested |
|---|---|---|---|
| trig-ratio-basic | FOUND. | Apply sin/cos/tan in a single right triangle. Direct application. | 8× |
| trig-ratio-chain | CORE | Chain trig ratios across 2+ connected triangles. Express composite ratio (AB/CD) through shared sides. | 8× |
| 3d-pythagoras | CORE | Apply Pythagoras across perpendicular planes to find 3D distances. | 8× |
| 3d-figure-decomposition | ADV. | Mentally extract right triangles from tetrahedra, cubes, prisms, pyramids. | 7× |
| 3d-cosine-formula | ADV. | Compute 3D side lengths via Pythagoras, then apply \(a^2 = b^2 + c^2 - 2bc\cos A\). | 5× |
| 3d-perpendicular-projection | ADV. | Find foot of perpendicular from point to plane. Angle between line and plane = angle to projection. | 5× |
| trig-identity-transform | CORE | Simplify using complementary, supplementary, and Pythagorean identities. | 4× |
| 3d-dihedral-angle | ADV. | Find angle between two planes sharing a common edge via foot-of-perpendicular construction. | 4× |
| trig-equation-solving | ADV. | Convert to quadratic in sinx or cosx, factor, count roots in all quadrants. | 3× |
| trig-graph-reading | FOUND. | Determine amplitude, frequency, phase, and vertical shift from graph of \(y = h + k\cos(nx°)\). | 3× |
| cosine-formula-2d | CORE | Apply \(a^2 = b^2 + c^2 - 2bc\cos A\) in 2D non-right triangles. | 3× |
| bearing-direction | CORE | Convert between compass bearings and angles. Use alternate angles with parallel North lines. | 2× |
| trig-identity-properties | CORE | Given constraints, determine which trig statements must hold. Test edge cases. | 1× |
| sine-formula-2d | CORE | Apply \(a/\sin A = b/\sin B\) in circle/sector problems. | 1× |
trig-ratio-basic + 3d-pythagoras, they can attempt 16 of 39 questions. Add 3d-cosine-formula + 3d-figure-decomposition and they cover 30 of 39. The remaining 9 are identity, equation, and graph — formulaic templates that drill well.
DSE14 is the proof case for diagram-dependent topic mastery. Everything we build here applies to any topic that requires spatial reasoning.
| Capability | DSE17 | DSE14 |
|---|---|---|
| Competency map | Done (47 Qs) | Done (39 Qs) |
| Knowledge graph | Done | Done + diagramSpec |
| Skill map | 20 skills | 14 skills |
| Exam intelligence | Done | Done |
| Mutation timeline | Done | Done (3D trig focus) |
| Deep dive report | Done | This page |
| Generated questions | 5 demonstrated | 7 with diagrams |
| Diagram generation | N/A | 7/7 generated, 75/75 checks |
DSE17 (Dispersion) proved the competency-map methodology works for text-only topics. DSE14 (Trigonometry) proves it for diagram-dependent topics — and that's where the moat deepens, because: