Pure Mathematics 2

Exam Technique Guide

Provenance: written 2026-09-28, against research/veridian/WMA12-verified-facts.md (the existing provenance-checked facts bank, read in full — §1–§9), the 11 real lesson files in content/veridian/curriculum/wma12/ (read for structure and to extract every existing trap-taxonomy item, so this guide cross-references rather than duplicates them), and primary-source spot checks run directly in this session against real PDFs under ~/Downloads/PAPERI/IAL_Unit2_Complete_2024_2025/IAL_Materials/Mathematics/ for the January 2025 and June 2025 series (pdftotext -layout, cross-read against the rendered text) — those two series were, by the facts bank's own admission (§5), only grepped for keyword matches before this pass, not read start-to-finish, so the January 2025 and June 2025 findings below are new to this course, not carried over from the facts bank.

#WMA12 Exam-Technique Guide

#What this document is, and isn't

This is the exam-craft layer for WMA12 (Pure Mathematics 2) — how to convert what you already know from the 11 real lessons in content/veridian/curriculum/wma12/ into marks under real exam conditions. It does not re-teach the mathematics: no derivations, no worked examples of the content itself. content/veridian/curriculum/wma12/paper-anatomy.ts already carries the standing, fully-cited reference on the paper's basic shape (90 minutes, 75 marks, M/A/B vocabulary) — this document builds on that rather than repeating it, and cites it directly rather than re-deriving its numbers. Where this document and that lesson file overlap (the core M/A/B definitions, the two Jan 2021 / Jan 2024 worked findings), the numbers are the same because they trace to the same primary sources; nothing here contradicts that file.

Every specific claim below is traceable to a real source: a spec paragraph, a mark-scheme line, or an examiner-report sentence, cited by series and question number wherever a question number is confirmed. Nothing is invented. Genuine gaps — places where the evidence ran out — are flagged as gaps, not smoothed over.

#Part 1 — Paper structure and a real timing budget

#The fixed facts

90 minutes, 75 marks, no choice of question, no sections. Verbatim from the specification's own P2.2 Assessment information (Issue 3, April 2019, p.17): "The assessment is 1 hour and 30 minutes... The assessment is out of 75 marks... Students must answer all questions." Every real question paper reviewed across 20 series (June 2019 – January 2026) ends with the literal line "TOTAL FOR PAPER IS 75 MARKS."

The average is 72 seconds per mark (90 min ÷ 75) — but the paper is not evenly weighted, and question count is not fixed. Confirmed question counts range from 9 to 11 across the archive: June 2021 and October 2020 both used 9 questions; June 2023 (3+6+7+7+5+9+8+5+7+10+8) and October 2024 (5+7+7+7+7+5+6+8+7+10+6) both used 11, each independently summing to 75. This is a real, checkable variable a candidate should not assume away.

#Three real, complete per-question shapes — not one template

Three full series have now been mark-counted question-by-question against their own primary PDFs (two by paper-anatomy.ts, one independently by this pass):

Series	Per-question marks	Range
January 2021 (10 Q)	6, 7, 8, 8, 4, 8, 6, 7, 10, 11	4–11 (2.75×)
January 2024 (10 Q)	3, 3, 6, 9, 8, 8, 9, 6, 14, 9	3–14 (4.7×)
June 2025 (10 Q) — independently extracted this session from `wma12-01-rms-20250814.pdf`	6, 7, 5, 9, 6, 10, 9, 8, 11, 4	4–11 (2.75×)

All three sum to exactly 75. The practical consequence: budget against the number printed next to the question, not against a memorised "the last question is always the biggest" assumption — January 2024's heaviest question (14 marks, nearly a fifth of the whole paper) was Q9, not the closer; June 2025's heaviest (11 marks) was Q9 too, but its lightest was the closer, Q10, at only 4.

#Topic order is not fixed either — a real correction worth flagging explicitly

The facts bank states (§6.1, hedged as "almost every series") that Proof anchors Question 1. June 2025 breaks that pattern outright: this session's own read of wma12-01-pef-20250814.pdf confirms Q1 is binomial expansion ("The first question of the paper tested the binomial expansion...") and Proof by exhaustion is Question 10, the paper's last question. The full real topic order for June 2025, confirmed directly against the examiner report's own per-question headings:

Binomial expansion — 2. The circle — 3. Area between a curve and a line — 4. Factor/remainder theorem — 5. Trapezium rule + definite integral (follow-through) — 6. Logarithms — 7. (not read this session) — 8. Trigonometric equations — 9. 3-D shape optimisation (surface area from a volume constraint, differentiation, second-derivative proof) — 10. Proof by exhaustion.

Don't walk in expecting Proof at Q1. It usually is — but "usually" is not "always," and a candidate who freezes because "Q1 isn't proof, something's wrong" has lost time on a non-problem.

#What's memorised vs. what's printed in the formula booklet — the real split

The booklet (Mathematical Formulae and Statistical Tables, described as "(Yellow)" on every real paper cover, Issue 2, January 2021) prints: the arithmetic-series formulae, the geometric-series formulae including S∞, the combinatorial binomial expansion and, on the same page, the general-real-n series form that P2 never actually needs, the change-of-base log formula, and the trapezium rule in full. It does not print: the addition/subtraction/power laws of logs, sin²θ+cos²θ≡1, tanθ≡sinθ/cosθ, or anything for Proof, algebraic division/the factor theorem, circle geometry, or differentiation/integration technique — those eight areas are 100% memorise-and-apply, not look-up. Knowing which half of your toolkit is on the page in front of you and which half has to already be in your head is a genuine time-saver: don't spend 20 seconds hunting the booklet for something that was never going to be there.

#Calculator policy, and the trap inside it

Calculators are permitted on the entire paper — there is no non-calculator section anywhere in this qualification, confirmed on every cover page reviewed, June 2019 through January 2026. But this is not a "calculator-friendly" paper in the way that sounds: a real, bold, per-question instruction — "In this question you must show all stages of your working", sometimes paired with "Solutions relying entirely on calculator technology are not acceptable" — appears above 2–5 questions on almost every series from October 2023 onward (confirmed directly on October 2023, June 2023, October 2024, and, this session, both January 2025 and June 2025), and a related instruction, "you must show detailed reasoning," appears on at least two further questions per series. The calculator-permitted rule and the show-your-working rule are not in tension — they coexist, and the second one is not decorative. See Part 2 for exactly what "no working" costs.

#Flagged gap — no fabricated weighting table

This document deliberately does not give a topic-by-topic percentage weighting (e.g. "≈15% of marks are calculus") because no pass to date — including this one — has topic-tagged all 75 marks across all 20 archived series. What's above is real: three fully-verified per-question shapes and one fully-verified topic order. A genuine cross-series weighting analysis is future work, not something to guess at here.

#Part 2 — The M/A/B mark-coding system, as this paper's real mark schemes actually use it

#The three codes, verbatim, independently re-confirmed this session

Directly from the "General Instructions for Marking" page printed at the front of every WMA12 mark scheme — checked this session against two further series (January 2025, wma12-01-rms-20250306.pdf; June 2025, wma12-01-rms-20250814.pdf) beyond the January 2020 series the facts bank already verified, word-for-word identical in all three:

M marks: Method marks are awarded for "knowing a method and attempting to apply it", unless otherwise indicated. A marks: Accuracy marks can only be awarded if the relevant method (M) marks have been earned. B marks are unconditional accuracy marks (independent of M marks). Marks should not be subdivided.

Nowhere in any real WMA12 mark scheme checked — six series now, across two independent passes — does an AO1/AO2/AO3 label appear on an individual mark. Every mark is M, A, B, or a dependent/follow-through variant of one of the three.

#Dependent marks: M1 vs dM1 vs ddM1 — the distinction that actually changes what you can salvage

A plain M1 is independent: it can be earned even if an earlier line was wrong, provided the method is correctly applied to whatever the candidate actually has on the page. A dM1 ("dependent M1") can only be earned if a specific earlier mark was already earned. This is not a technicality — it determines how much of a question is recoverable after an early slip.

Real worked case, re-verified a third time this session against WMA12_01_pef_20210304.pdf: January 2021, Question 2 (7 marks — stationary points and their nature). Part (a) is M1(differentiate) A1(correct derivative) M1(set = 0, solve the quadratic — independent, scorable even on a wrong derivative) A1. Part (b) is M1(differentiate again) dM1(substitute one of the two already-found x-values in, draw a consistent conclusion — dependent) A1(correct values and conclusions at both points). A wrong derivative in part (a) still leaves the "solve the quadratic" method mark available; a correct second derivative in part (b) that never has a number substituted into it earns nothing further, because the dM1 it's gating on has not been satisfied. The real examiner report for this exact question: "there were many who did not then substitute the values for x found in part (a) into their second derivative," and "only 50% scored" the mark this dependency gates.

A ddM1 (doubly-dependent) exists too — confirmed, e.g., on the facts bank's Jan 2021 Q3(ii) log-equation anchor (M1 dM1 A1 ddM1 A1) — a chain where the final method mark depends on two prior stages both having gone right, not one.

The practical rule this produces: identify, before you start writing, whether the next step you're about to take is going to be independent (worth attempting even if you're not confident in what came before) or dependent (only scores if a specific earlier number survives intact). Sixteen out of the paper's real dM1/ddM1 instances checked this session (28 across the June 2025 mark scheme alone, 19 across January 2025's) confirm this chain structure is the norm, not the exception, on any question with more than one part.

#The abbreviation glossary — with real, checked examples, not just the definitions

Verbatim from the glossary printed in every mark scheme: bod – benefit of doubt; ft – follow through; cao – correct answer only; cso – correct solution only ("there must be no errors in this part of the question to obtain this mark"); isw – ignore subsequent working; awrt – answers which round to; SC – special case; oe – or equivalent; dep – dependent; indep – independent; ag – the answer is printed on the paper (answer given).

Real, checked instances, from two additional series independently pulled this session:

isw, June 2025 Q3 mark scheme: " $(x-1)^2+(y-10)^2=34$ but not $(x-1)^2+(y-10)^2=(\sqrt{34})^2$ isw after a correct answer" — once the correct answer is on the page, further (even slightly clumsy) manipulation after it doesn't cost the mark.
awrt, both series repeatedly — e.g. January 2025 Q3: "A1: x = awrt 3.76 or x = awrt 2.99"; June 2025 Q8: "A1: awrt 50.3, awrt 309.7 and no others in the range. Must be in degrees." — a rounded decimal is accepted specifically because the question doesn't demand an exact form; contrast this directly with §"exact answers" below, where a rounded decimal is not accepted.
oe, January 2025 Q5(a): "B1: Identifies h = 1.5 or may be implied by e.g. 0.75 oe outside of the bracket."
A1ft, both series, e.g. January 2025 Q5(b)(ii): "M1, A1ft" on an integral whose limits follow through from a candidate's own (possibly wrong) part (a) answer — a wrong earlier number does not automatically zero out a later part, provided the method on the later part is sound and applied consistently to that earlier number.
SC (special case): listed in the standing glossary of every series checked, but no live instance was found in either of the two real mark schemes spot-checked this session, nor in the facts bank's own quoted examples — flagged honestly as an unconfirmed-in-practice code, not claimed as something you'll definitely see.

#The rule that decides whether a correct answer scores anything at all

Verbatim, General Principles for Pure Mathematics Marking, printed in every series: "Where a method involves using a formula that has been learnt... the formula should be quoted first... Where the formula is not quoted, the method mark can be gained by implication from correct working with values, but may be lost if there is any mistake in the working."

Two independent, real, and directly quotable enforcements of this rule, one already in the facts bank and one found fresh this session:

January 2024, Q5(i) (facts bank): "Many candidates correctly used the sum to infinity formula, however a number resorted to using the summation button on their calculator which, even if [the value] was found, no marks were scored."
January 2025, Q5(a) — new this session, wma12-01-rms-20250306.pdf — the mark scheme states outright, in its own marking notes: "Note that awrt 18.8 with no working is 0 marks", and again on the very next line: "Condone invisible brackets to be implied by further work or awrt 18.8 but do not allow just awrt 18.8 with no working seen to score this mark."

A mark scheme built entirely from M/A/B lines has nothing to read on the M line if there is no visible method — the correct number, sitting alone, is invisible to it.

#Exact answers — when a correct decimal is worth zero

Verbatim, the general rule: "where, for example, an exact answer is asked for, or working with surds is clearly required, marks will normally be lost if the candidate resorts to using rounded decimals."

Real, checked instance: January 2021 Q2's own A1 line for the stationary points' x-coordinates reads, in full (verified against the rendered mark-scheme page, not just extracted text, because the source carries an invisible recurring-decimal dot): "x = 8/3, −2 or equivalent single fraction (do not accept 2.67 but 2.6[recurring] can be accepted.)" — a decimal is accepted only if it is written in a form that unambiguously communicates the exact value (a recurring-decimal notation), never a truncated or rounded one. The same rule recurs on circle radii (January 2023: "a common error was giving the answer r = √20" as a decimal) and on binomial/log final forms (§6.6, "single-log-form-left-uncombined" in the logs lesson).

#The misreading rule and the multiple-attempts rule

Misreading: "For misreading which does not alter the character of a question or materially simplify it, deduct two from any A or B marks gained, in that part of the question affected." Misreading a question is not automatically fatal — but it is not free either, and it's specifically an A/B penalty, never an M one.

Multiple attempts: "If all but one attempt is crossed out, mark the attempt which is NOT crossed out. If either all attempts are crossed out or none are crossed out, mark all the attempts and score the highest single attempt." Crossing out a wrong start and writing a clean second attempt costs nothing — leaving both un-crossed-out risks the examiner marking whichever scores lower if the two conflict, not the higher one automatically.

#Part 3 — Self-mark checklist

Run the universal checklist on every question, then the topic-specific triggers where they apply. This is written against the real mark-scheme mechanics in Part 2, not generic exam advice.

#Universal, every question

☐ Read the printed mark allocation before writing anything. At 72 seconds/mark on average, a 4-mark question and an 11-mark question are not the same commitment — and the real per-question spreads in Part 1 mean you cannot assume the shape from memory.
☐ If the question carries "show all stages of your working" or bans calculator-only solutions, write the formula or method line before touching a single number. A correct final value with no method line is a real, repeatedly confirmed zero (January 2024 Q5(i); January 2025 Q5(a), both cited above) — not a harsh technicality, the literal mechanism of an M/A/B scheme.
☐ If an exact form, a surd, or a specific "form $x = \log_a k$ "-type answer is demanded, do not round until you are certain the question allows it — and check whether it does at all. Default to keeping fractions and surds exact throughout; only round at the very end, and only if "awrt" or "3 s.f." appears in the question.
☐ Before moving to the next line, ask: is what I'm about to write dependent on a specific number from the line above surviving into it? If yes (a dM1/ddM1 situation — see Part 2), that number has to actually appear, substituted, not just be sitting correctly derived two lines up.
☐ If a part says "using your answer to (a)" or "hence," do not recompute the earlier part from scratch. A real, quantified cost: June 2025 Q5(b), where candidates who "re-did the trapezium rule in part (b)... gained no marks as the question had insisted on using their answer to part (a)," and the report adds explicitly, "this would also have been detrimental to their timing." Zero marks and lost time, for the same habit.
☐ State your final answer to a multi-term working on its own, isolated line — boxed, circled, or underlined. Confirmed cost of not doing this: January 2024 Q2's final mark was "commonly not awarded to candidates who failed to isolate the required term from a list."
☐ Check the units the question actually asked for — degrees vs. radians (recurring across at least two 2025 trig questions, see Part 4), exact vs. decimal, coefficient vs. coefficient-with-the-variable-attached.
☐ Sanity-check the physical plausibility of your final answer before moving on, especially on optimisation and geometry questions: a negative length, a negative radius, or a root outside the question's own stated domain is a signal to check the line above, not just to write it down. Confirmed real cost of skipping this: January 2025 Q3(iii)-type slips ("a few achieved a negative value but failed to realise that this was an impossible answer" — June 2025 Q9) and circle questions where a wrong root is carried through unchecked.

#Topic-specific triggers

Circle questions: sketch first. Real, direct examiner advice: "A sketch is very helpful to the candidate in seeing what is required to solve the problem and it is advisable for candidates to use a sketch" (January 2021 Q9).
Binomial/log/series questions with a choice of formula: the P2-appropriate form is (almost) always the simpler-looking one sitting next to a more general form on the same booklet page — see Part 4's "formula-choice" entries.
Proof questions: check which proof method the question actually names (exhaustion vs. counter-example vs. neither) before starting — an unrequested algebraic proof is real, confirmed wasted time even when eventually correct.
Any question with a "(b) using your answer to (a)" or a follow-through structure: your part (a) answer's form matters as much as its value — an answer left in the wrong form can silently break a later part's follow-through eligibility (confirmed, June 2025 Q9(b): "candidates tried to follow through an answer to part (a) which was not in the correct form, meaning their derivative was not in the correct format either and leading to no marks being awarded").

#Part 4 — Common-error digest

#How to use this section

The 11 real lesson files already carry 66 examiner-confirmed trap-taxonomy items, each with its own real citation, worked wrong-path, and (where applicable) a distractor-matched MCQ. This section does not reproduce them — it locates them by topic so you know where to drill, and then adds the errors found genuinely new to this course during this pass's own primary-source checks (mostly from January 2025 and June 2025, the two series the facts bank itself flagged as under-read).

#Locator — existing trap-taxonomy coverage by lesson

Proof (proof.ts): algebraic-proof-attempted-when-exhaustion-required · numerical-examples-mistaken-for-general-proof · multiple-of-k-assertion-without-quotient-shown · product-misread-as-sum · case-defining-relation-applied-backwards · column-values-transposed-between-derived-variables · case-list-not-exactly-complete · conclusion-omitted · counter-example-search-continued-past-the-first-valid-one
Factor/remainder theorems (factor-and-remainder-theorems.ts): root-sign-taken-straight-from-the-bracket · remainder-equation-set-to-zero · long-division-chosen-when-substitution-was-asked-for · factorisation-stopped-one-step-short · hence-factorise-reached-without-using-the-given-factor · non-integer-factor-form-given · answer-left-as-a-list-not-a-product · show-that-steps-omitted
The circle (the-circle.ts): centre-sign-not-flipped · general-form-read-as-centre-radius-form · solved-parameter-not-substituted-into-earlier-general-centre · radius-given-as-decimal-not-exact-surd · implicit-differentiation-chosen-over-the-property · origin-and-circle-centre-both-labelled-o · substitution-direction-not-checked-for-fractions · general-area-formula-chosen-over-axis-aligned-base-height
Sequences/arithmetic series (sequences-and-arithmetic-series.ts): sum-proof-last-term-off-by-one · sum-proof-reversal-skipped · sum-proof-insufficient-terms-shown · induction-attempted-for-standard-proof · sum-proof-assumes-the-result · sum-proof-pairing-fails-for-odd-n · arithmetic-and-geometric-formulas-confused
Geometric series (geometric-series.ts): arithmetic-and-geometric-formulas-confused · overlap-term-double-counted · geometric-series-restarted-from-the-original-first-term · sum-to-infinity-first-term-misidentified · calculator-summation-shown-without-formula · standard-proofs-not-known · geometric-proof-subtraction-direction-reversed · convergence-condition-read-as-r-less-than-one
Binomial expansion (binomial-expansion.ts): term-index-mismatched-to-power · general-series-form-chosen-over-combinatorial-form · bx-term-not-raised-as-a-whole · sign-lost-on-odd-powers-of-a-negative-b · pascal-triangle-row-miscounted · required-term-not-isolated-from-a-list
Logarithms/exponentials (logarithms-and-exponentials.ts): log-subtraction-rewritten-as-division · log-of-a-sum-expanded-as-sum-of-logs · log-squared-confused-with-log-of-the-square · exponent-rearrangement-left-unfinished · single-log-form-left-uncombined · logs-taken-before-the-equation-is-rearranged
Trigonometric equations (trigonometric-equations.ts): divides-out-a-common-trig-factor · wrong-identity-substituted-for-cos-squared · coefficient-dropped-when-distributing-after-substitution · incomplete-multiply-through-after-tan-substitution · compound-angle-not-isolated-before-inverse-function · coefficient-distributed-across-the-whole-compound-angle · compound-angle-bracket-range-scaled-by-coefficient · decimal-given-instead-of-exact-kπ-radian-form · squared-compound-angle-sign-or-period-copy-dropped · premature-rounding-before-final-angle
Stationary points/curve sketching (stationary-points-and-curve-sketching.ts): second-derivative-found-but-never-substituted · y-coordinates-found-when-only-x-was-asked · second-derivative-set-to-zero-and-solved · y-values-eyeballed-instead-of-gradient-tested · critical-value-inequality-wrong-region-chosen · second-derivative-test-inconclusive-mishandled · no-turning-points-asserted-without-a-discriminant-shown · stationary-point-x-coordinate-rounded-instead-of-exact
Definite integrals/areas (definite-integrals-and-areas.ts): answer-produced-with-no-method-shown · curve-and-line-subtracted-in-the-wrong-order · line-boundary-ignored-entirely · sign-change-inside-the-interval-not-split · slowest-valid-method-chosen-under-time-pressure · y-values-used-as-x-limits
Trapezium rule (the-trapezium-rule.ts): strip-width-from-point-count-not-gap-count · endpoints-and-interior-ordinates-swapped · bracket-closed-before-scaling · conclusion-restated-instead-of-reasoned · trapezium-rule-assumed-tested-in-isolation

#New for this pass — genuinely uncaptured patterns, each independently source-checked this session

1. The "double reciprocal" normal-gradient error (circle geometry). Distinct from any of the-circle.ts's own eight traps. Real, verbatim, January 2023 Q6: after correctly finding a normal's gradient from the negative reciprocal of the tangent's, "some went on to find the negative reciprocal and use this as their normal gradient" — applying the perpendicular-gradient rule a second time to an already-correct value. This is a distinct failure mode from the already-covered implicit-differentiation-chosen-over-the-property and substitution-direction-not-checked-for-fractions traps — it happens after the gradient is right, not instead of the right method.

2. Diameter mistaken for radius, and "circle" questions answered as if they asked for a tangent or radius equation. New, from June 2025 Q2 (5 marks — find the equation of a circle from the coordinates of a diameter's endpoints), independently read this session from wma12-01-pef-20250814.pdf: "The majority of candidates also realised they had to use the distance formula with most choosing to apply it to the end points of the diameter. This caused issues for some as they did not recognise that they had found the diameter (or diameter squared) and treated it as though it was the radius (or radius²). This misunderstanding meant they lost at least three of the five marks for this part." A second, separate failure on the same question: "Some candidates did not understand what the question was asking and instead found the equation of the radius, or the tangent at one of the given points" — answering a plausible-sounding but different question entirely. Neither pattern currently has a trap-taxonomy entry in the-circle.ts.

3. Degrees given where an exact radian (kπ) form was required — a recurring, cross-series pattern not yet named as its own trap. trigonometric-equations.ts already has decimal-given-instead-of-exact-kπ-radian-form, which covers rounding a correct radian value to a decimal — a related but distinct error is working the whole question in degrees when the question is set (and must be answered) in radians. Confirmed twice, independently: October 2022 Q5 ("Work was sometimes seen in degrees") and June 2025 Q8(i) ("There were also candidates who worked in degrees" despite a radian-form answer being required). Worth a dedicated pre-question check: confirm the interval given (e.g. $0 \le x < 2\pi$ vs. $0 \le x < 360°$ ) before picking up the calculator's angle mode.

4. Cube root taken as square root when solving a stationary-point equation of the form $x^3 = k$ . New, from January 2025 Q3(b) (a differentiation optimisation question, 4 marks for finding a critical x-value from $dA/dx=0$ reducing to $x^3=k$ ), independently read this session from wma12-01-pef-20250306.pdf: "Taking a square root rather than a cube root when solving $x^3=\frac{160}{3}$ " — one of three named ways marks were dropped on this question, alongside "slips when writing down area formulae" and "failing to write correct notation, $\frac{d^2A}{dx^2}$ , or a correct reason, $\frac{d^2A}{dx^2}>0$ , when attempting to prove part (c)" (the second-derivative-test conclusion needing both the correct notation and an explicit stated inequality, not just a number). This is a distinct, more specific failure than stationary-points-and-curve-sketching.ts's existing second-derivative-test-inconclusive-mishandled trap — it's a root-extraction slip one step earlier in the chain, on questions where the stationary-point equation isn't a quadratic.

5. A newly-documented question type: 3-D shape optimisation (volume constraint → surface-area formula → differentiation → second-derivative proof), combined with real, confirmed error patterns. June 2025 Q9 (11 marks) is not covered anywhere in the facts bank's §6 topic breakdown. Independently read this session from wma12-01-pef-20250814.pdf: "It was common for candidates to treat the prism as closed despite the diagram and statement that it was open" (adding an unwanted area to the surface-area formula); "One common error was to not divide by 2 when finding the triangular cross section in the volume calculation"; and, on solving for the critical x-value, the same cube-root-vs-square-root slip as finding #4 above, independently confirmed on a second, different series: "Square rooting instead of cube rooting and inaccurate rearrangement... A few achieved a negative value but failed to realise that this was an impossible answer." This looks like a genuine, recurring family of errors specific to real-world-context optimisation questions (misreading the physical constraint of the diagram; the cube-root slip; not rejecting a physically-impossible root) rather than a one-off. It is not yet built into any lesson's trap-taxonomy and is flagged here as a lead for a future content pass, not claimed as fixed.

6. Redoing earlier working instead of using a stated follow-through, at direct cost to both marks and time. New, June 2025 Q5(b): "On the rare occasion some re-did the trapezium rule in part (b) which gained no marks as the question had insisted on using their answer to part (a). This would also have been detrimental to their timing." This is as much a self-mark-checklist item (Part 3) as a trap-taxonomy one — flagged in both places because it costs marks and time simultaneously, a combination worth explicit attention.

#Part 5 — Exam-day protocol

Kept deliberately narrow: this is procedural protocol traceable to real rubric wording and real mark-scheme behaviour, not general exam-wellness advice, which this pass found no WMA12-specific grounding for.

Bring an "any calculator permitted by Pearson regulations" — no CAS, no symbolic algebra/differentiation/integration, no stored retrievable formulae. Confirmed verbatim on every cover page reviewed, January 2020 through January 2026.
Know the Yellow booklet's actual contents before you open it in the exam, not just that it exists. Part 1 above lists exactly what is and isn't printed — the ten seconds you'd spend hunting for a log law that was never going to be there is ten seconds you don't have, at 72 seconds/mark average.
Read every question's own instruction line before starting it. A specific minority of questions each series carry a bold, individual instruction — "show all stages of your working," "solutions relying entirely on calculator technology are not acceptable," or "you must show detailed reasoning" — confirmed on real papers spanning January 2020 through June 2025. These are not boilerplate; Part 2 shows exactly what ignoring them costs (a confirmed real zero, twice, on two different series' worth of otherwise-correct final answers).
Do not assume the paper's shape from a single practice paper. Question count (9–11), the mark spread per question (a 4–11 range in one real series, a 3–14 range in another), and even which topic opens the paper (Proof is common at Q1 but was Q10 in June 2025) are all genuinely variable. Budget from the numbers printed on the paper in front of you.
Default to the exact form until the question itself tells you otherwise. If a question doesn't say "awrt" or a specified number of significant figures, don't round early — Part 2's exact-answer rule costs real marks on an otherwise-correct method.
When a part says "using your answer to (a)" or "hence," use it — don't recompute. Confirmed, twice over (Part 3, Part 4 #6): this specific habit costs marks and time in the same move.
In the last few minutes, prioritise isolating and labelling final answers over starting new working. The confirmed mark-loss pattern for "correct value, buried in a list, never pointed to" (Part 4) is exactly the kind of thing a rushed final few minutes makes worse, not better — circle or underline what you've already got before reaching for a new method.

#Flagged gaps — what this document does not claim

No cross-series topic-weighting percentage. See Part 1's own flagged note — this would require topic-tagging all 75 marks across all 20 archived series, which no pass to date, including this one, has done.
January 2025's full per-question mark breakdown could not be independently confirmed this session. A direct extraction attempt from wma12-01-rms-20250306.pdf this session found 9 distinct per-question mark totals summing to 71, not 75 — almost certainly one question's total was missed by the extraction (pdftotext -layout mangles some bracketed/stacked numerals, as the facts bank's own §5 methodology note already warns), but rather than present a wrong table, this document only cites the single question from that series (Q3, Q5) it could verify directly against the rendered mark-scheme text. The June 2025 breakdown in Part 1, by contrast, summed correctly to 75 and is reported.
Only one series (June 2025) has been topic-mapped question-by-question, by this pass. The "topic order is not fixed" finding in Part 1 is real and directly sourced, but it is one data point against the facts bank's own already-hedged claim, not a full replacement statistic.
The five new error patterns in Part 4 (double-reciprocal normal gradient; diameter-as-radius; degrees-for-radians; cube-root-as-square-root; the 3-D optimisation question family) are each confirmed against a single real series (January 2023, June 2025, or both January 2025 and June 2025 independently for the degrees/cube-root patterns). Whether each recurs across further series has not been checked and would need a dedicated sweep, the same way the facts bank's own §9 lesson-audit log describes for other topics.
October 2025 and January 2026 examiner reports were not found in any archive (per the facts bank, §8) — this document's real findings are bounded by series through June 2025, the most recent with a published report at the time of writing.