Pure Mathematics 3

Forty Eight Hour Checklist

Provenance: written 2026-09-29, against research/veridian/WMA13-verified-facts.md (read directly in full, all nine numbered sections — §1 spec contents, §2 source archive, §3 assessment structure, §4 mark-scheme conventions, §5 per-topic findings, §6 cross-topic patterns, §7 lesson-readiness notes, §8 flagged issues, §9 lesson-audit log), content/veridian/technique-hub/wma13/forge-exam-technique-guide.md (read in full), content/veridian/technique-hub/wma13/codex-formula-and-method-bank.md (read in part, Part 1 memorise/booklet split), and content/veridian/technique-hub/wma13/signal-frequency-and-boundaries.md (read in full, including its independently-fetched Pearson grade-boundary table). The 11 lesson files in content/veridian/curriculum/wma13/ and content/veridian/question-bank/wma13.ts were checked by filename, header and trap-taxonomy locator rather than re-opened line by line, matching the sourcing approach the sibling WMA12/WBS paper checklists in this same folder tree already use for this document type. No new primary-source PDF was opened for this document — everything below is reframed from the four sources named above into a time-boxed countdown with explicit self-test items and an explicit "don't start this now" list; where a fact already has a home in the facts bank, FORGE, CODEX or SIGNAL, this document points at it by name and citation rather than re-stating it in full. One genuine discrepancy between two of this course's own WMA13 documents was found while writing this and is recorded in "Flagged gaps" below rather than silently resolved in either direction.

#WMA13 — 48-Hour Pre-Exam Checklist

Pearson Edexcel International A-Level, Unit P3: Pure Mathematics 3. 90 minutes, 75 marks, every question compulsory, calculator permitted throughout but with no symbolic algebra/differentiation/ integration facility and no communication or stored-formula facility (spec Appendix 6; facts bank §3). Formula booklet provided (Mathematical Formulae and Statistical Tables, "the Yellow booklet").

#What this document is, and isn't

This is not a content lesson and not a fifth copy of the exam-technique guide. It assumes you have already worked through the relevant lessons in content/veridian/curriculum/wma13/ and read (or at least skimmed) FORGE and CODEX at some point before the last 48 hours. What this document adds is a time-boxed triage: what actually has a documented payoff in the last two days, in what order, self-tested against this specific paper's own confirmed weak points — and, just as concretely, what this paper's own real record says is a poor use of your remaining hours. Every claim below is traceable to the facts bank, FORGE, CODEX or SIGNAL by name and, where a specific question is cited, by series and question number; nothing new is asserted here that a primary source doesn't establish.

#Two things to accept about this paper before the countdown starts

The paper's shape is not fully fixed, and marks are not spread evenly. Question count is confirmed at 10 for January 2023 and 9 for October 2020 (facts bank §3, §8) — say "9 or 10 questions" to yourself, not a specific number, unless you know the exact figure for the series you're sitting. Per-question tariffs in real papers checked range from 5 to 11 marks within a single series (January 2023's own breakdown: 6+6+5+7+9+8+9+5+11+9 = 75; FORGE Part 1). The 72-seconds-per-mark figure (90 minutes ÷ 75 marks) is a real, arithmetic-checked planning tool — but apply it to each question's own printed tariff as you reach it, not to a memorised question order.
A numerically correct final answer with no visible method is a real, repeated partial or total zero on this paper — not a technicality. The clearest documented case: October 2024 Q8(c), a two-exponential maximum-heart-rate question carrying the explicit "solutions relying entirely on calculator technology are not acceptable" instruction. The real mark scheme's own structure for that part is M1, A1 (differentiate), then dM1, M1, A1 (actually solve) — and the examiner report states plainly that a candidate who wrote the correct derivative, set it to zero, and jumped straight to $t=1.158$ with no algebra shown scored only the first two marks: the differentiation itself is visible, checkable work, but every mark chained to solving the equation is void without shown method, regardless of how correct the final number is (facts bank §5, "4. Differentiation"). This governs almost everything in the self-test bank below.

#The countdown

#T-48h to T-24h — the last full review session

Run one complete self-test against the mark-scheme mechanics, not just the maths. Pick a past paper you've already attempted (or attempt one now) and re-mark your own work using the real M/A/B rules — specifically the dM1/ddM1 dependency distinction (§below) and the "quote the formula first" rule — rather than just checking final answers against the scheme. If you haven't done this yet, do it now: it tests exam mechanics reading alone doesn't surface.
Work through the cross-topic habits list and the per-topic self-test below once, in full, marking honestly which items you can answer from memory and which you can't.
If — and only if — you found a genuine, specific gap, spend the rest of this block closing that one gap, using the relevant curriculum lesson file or CODEX's matching method skeleton. Do not open a topic you weren't already shaky on going in — see "What not to start" below.
Do not attempt a second full past paper today if you already ran one. One honestly re-marked paper against the real mark-scheme mechanics is worth more here than two rushed ones marked only against final answers.

#T-24h to T-3h — the day before

Re-read the memorise-vs-booklet split once, out loud or from memory, then check yourself against the real list (facts bank §3; CODEX Part 1). The booklet's P3 section prints the quotient-rule formula, the reciprocal-integration results for $\sec^2kx$ , $\tan x$ , $\cot x$ , and the sum/difference/factor trig identities — but it does not print $\cos^2A+\sin^2A\equiv1$ , $\sin2A$ / $\cos2A$ / $\tan2A$ , the standard differentiation results ( $\sin kx\to k\cos kx$ , $a^x\to a^x\ln a$ , product/chain/sum rules), or the standard integration results ( $\cos kx\to\tfrac1k\sin kx+c$ , $\tfrac1x\to\ln|x|+c$ , the reverse-chain-rule form). These are exactly the results a tired brain reaches for the booklet on and finds nothing — check yourself against the real split, not a guess at what "feels" memorised.
Run the "confirmed repeat offenders" list below — the five patterns with the deepest cross-series confirmation in the whole archive, not a guess at what's "likely."
Confirm your calculator is on the permitted list: no symbolic algebra, no differentiation/integration facility, no stored retrievable formulae, no communication/internet facility (spec Appendix 6). Check this tonight, not tomorrow morning.
Do not start a brand-new past paper this late in the day. A rushed, partial attempt the evening before produces more anxiety than signal; use the time on the self-test bank instead.

#T-3h to T-0 — exam morning

Ten-minute skim only, not new work: the "confirmed repeat offenders" list, plus whichever of your own T-48h self-test items you marked as shaky.
Physically check: Yellow booklet (Mathematical Formulae and Statistical Tables), a permitted calculator, nothing else that could look like a stored-formula device.
Remind yourself of three cross-topic rules that cost real marks regardless of topic: quote the formula/rule/identity before you use it; keep exact surds and the angle unit the question asks for until told a decimal is acceptable; if a sign-change conclusion is needed, the word "continuous" has to actually be there.
Do not use this window to learn anything new. Everything in "What not to start" below applies with extra force this close to the exam.

#The last few minutes before you're told to begin

Read every question's own instruction line before starting it. A specific minority of questions each series carry a bold "show all stages of your working" or "solutions relying entirely on calculator technology are not acceptable" rubric (verified verbatim, printed directly above January 2023 Q5, and recurring across multiple series — facts bank §3). These are not boilerplate; they are individually policed.
Write the mark allocation next to each question as you reach it and budget from that printed number, not from memory of a practice paper's shape.

#Cross-topic habits — self-test these first, they apply to every question

Five real, mark-scheme-mechanical habits, each independently confirmed costing marks on multiple series regardless of which spec area the question is testing.

Quote the rule/formula/identity before you substitute into it. Confirmed independently across at least five series (Oct 2020 Q3, Oct 2021 Q9(a), Jan 2022 Q6(a), Jun 2023 Q10(a), Jan 2024 Q7(a)) — the general marking principles state an unquoted method mark "may be lost if there is any mistake" in the working it's inferred from. Self-test: pick any rule-based question type and write the general rule on its own line, before any numbers. If your habit is to go straight to numbers, that's the gap.
Never let a calculator's solver stand in for shown algebra on a question that flags it. Confirmed across at least four series (Jun 2022, Oct 2022, Jan 2024, Jan 2025 — facts bank §5, "3. Exponentials and logarithms"), and named as a real, specific mark loss on the exponential maximum-heart-rate question above. Self-test: on any equation-solving step, is the equation you're about to solve written down as its own line before you reach for a solver of any kind?
Keep exact surds and the angle unit the question names, until told a decimal is acceptable. Confirmed directly on the harmonic-form construction marks (Oct 2020 Q7(a): $R=\sqrt{17}$ must stay exact, and $\alpha$ in radians scores A0 if given as the numerically-equivalent degree value) and on a log-linear-graph exact answer ( $T=\tfrac1{36}$ , Jun 2023 Q2(a)(ii), not accepted as a decimal or an unexpanded power). Self-test: does your final line for an "exact value" question still contain a surd, fraction, or the named angle unit — or did it quietly become a rounded decimal somewhere in the working?
On a "show that"/proof question, write every intermediate line — the destination matching the printed answer isn't enough. Confirmed across at least four series (Oct 2020, Jan 2022, Jun 2023, Jan 2024 — FORGE's self-mark checklist). Self-test: in your last proof attempt, could a marker see which double-angle or identity form you substituted, or only infer it from the answer?
Only ever tag A or B marks "ft" — an M mark is never conditional in that way. Verbatim general marking guidance: "A and B marks may be f.t." — M marks are absent from that sentence, because an M mark rewards the method itself, independent of the value it's applied to. This course's own lesson-audit history got this wrong three separate times before catching it (differentiation-rules.ts, numerical-methods.ts, reciprocal-and-inverse-trigonometric-functions.ts — facts bank §9). Self-test: if you're re-marking your own mock, treat any "M ft" you write as a signal to re-check — it is very likely a plain M1.

#Confirmed repeat offenders — traps documented on three or more independent series

Domain of the inverse function omitted, even with correct algebra. The single most consistently tested sub-topic in the whole archive, appearing as Q1 or Q2 in nearly every series reviewed. Confirmed directly: Jun 2022 Q2 ("the majority of candidates still failed to state the domain... and did not achieve the B mark"), Jan 2022 Q6 ("a majority... were not aware that the domain was required"), Oct 2021 Q1(b). Self-test: every time you find $f^{-1}(x)$ , is stating its domain — in $y$ / $f(x)$ notation, not $x$ -notation — a reflex, not an afterthought?
Modulus equation/inequality: only one branch solved, or the wrong combined region selected at the end. Confirmed across Jan 2022 Q7, Oct 2020 Q4(c), Jun 2023 Q6(c). Self-test: after finding both critical values, do you actually go back and select the region the original inequality needs, rather than stopping once both values are found?
Modelling answer missing its unit or a contextual word ("loss", "decreasing", the actual unit). Confirmed across at least four series with near-identical wording each time: Oct 2021 Q3(b), Jan 2022 Q4(c), Oct 2022 Q5(a), Jan 2025 Q2(a). Self-test: does every modelled numeric answer you write end with a unit and, where relevant, a direction word?
Sign-change conclusion missing the word "continuous." The single most repeated specific error in the entire archive — a real mark scheme requires "both values correct... reason, which must mention continuity... conclusion" as one bundled, dependent accuracy mark, and examiner reports name the missing continuity clause specifically across essentially every series with this question type (Oct 2022, Jan 2022 Q5(a), Jan 2024, improving only from Jan 2025 onward). Self-test: does your root-location conclusion literally contain the word "continuous," not just "sign change"?
Iterates rounded prematurely, or reported to the wrong number of decimal places partway through a chain. Confirmed across at least three series: Oct 2020 Q6(c) ("did not give the value of their iterate to the required number of decimal places"), Jan 2024 Q2(c)(ii) ("an error in rounding or premature rounding of intermediate steps"), and the general Oct 2022 cross-series summary. Self-test: in a chain of three or more iterations, are you carrying full calculator precision between steps and rounding only the value you're asked to report?

#Per-topic self-test — one pass through every spec area

Short, specific self-checks, each tied to a real citation already established in the facts bank, FORGE or CODEX. If any item below is a genuine "no," go to the named lesson file to fix it — not this document.

1.1 Algebraic division (rational-expressions-and-algebraic-division.ts): Have you checked a division is actually finished — remainder's degree strictly less than the divisor's, not a leftover $x$ term after dividing by a quadratic (Oct 2024 Q9(c): "terminate the division early")? On a "show [constant] = ..." question, do you substitute/verify explicitly rather than just stating the value (Jan 2025 Q4(a): candidates who found A, B, C correctly still lost the mark for not showing D=0)?
1.2 Functions (functions-composition-and-inverses.ts): Domain-of-inverse habit above. On a question defining two functions, have you checked which letter your inverse actually belongs to (Jan 2023 Q1(c): a correctly-derived $g^{-1}(x)$ labelled $f^{-1}$ scores A0)? If a composed argument is algebraic rather than numeric, have you used any stated sign/positivity constraint to collapse it before setting up an equation (Jun 2026 Q2(d))?
1.3 The modulus function (modulus-and-graph-transformations.ts): Both-branches/region habit above. For $y=f(|x|)$ , are you mirroring the positive- $x$ branch rather than only reflecting in the $y$ -axis (Jun 2023 Q6(d))?
1.4 Transformations (modulus-and-graph-transformations.ts): On a multi-part question, are you transforming the point the question stem actually names, not a point calculated in an earlier part (Jun 2026 Q3(c) — zero-marked even with flawless transformation arithmetic, because the wrong point was transformed)?
2.1–2.2 Reciprocal/inverse trig (reciprocal-and-inverse-trigonometric-functions.ts): Can you state $\csc\theta=1/\sin\theta$ without pairing it with cosine? If a coefficient sits with the trig function, does it stay outside the reciprocal ( $3\csc\theta\neq\frac1{3\sin\theta}$ ; Jan 2022 Q2(a) scores this B0)?
2.3 Double angle / harmonic form (compound-and-double-angle-formulae.ts, harmonic-form.ts): Exact-value habit above, for R and α specifically. When recovering $\theta$ from a solved $2\theta$ , do you remember to halve (Jan 2022 Q2(b): "did not divide by 2, leaving their answer as 48.59…°" instead of 24.3°)? When a question asks for "the smallest" or one specific solution, do you actively reject every other genuine solution you found, not just report all of them (same question — a candidate who found both solutions but kept both lost the mark anyway)?
3.1–3.3 Exponentials/logs (exponentials-and-logarithmic-graphs.ts): Units/context habit above. On a log-linear graph, is your log base consistent throughout the question, not switched to base 10 or $e$ /ln mid-way (Jun 2023 Q2, confirmed error)? If asked to interpret a constant $b$ in words, have you stated all three of direction, the percentage, and "per year" — the real reject list explicitly refuses answers missing just one of the three (Jun 2026 Q5(b))?
4.1–4.3 Differentiation (differentiation-rules.ts): Quote-the-rule habit above. On a $\frac{dy}{dx}=1/(\frac{dx}{dy})$ turning-point question, do you set $\frac{dx}{dy}=0$ directly, rather than "solving" the already-inverted reciprocal equalling zero — the real mark scheme scores the latter zero even when it happens to land on the right value (verified against the primary October 2024 mark scheme and report)?
4.4 Exponential growth/decay (exponential-growth-and-decay.ts): For a rate of change, do you differentiate rather than substitute a value into the original model (Jun 2023 Q7(b) — both wrong approaches scored zero)? If the exponent is negative, does the sign survive the chain rule (Jan 2022 Q8(c), Jan 2024 Q5(c))? For $a^x$ , do you reach for $a^x\ln a$ , never the power rule, even under time pressure (Oct 2024 Q4(c): a candidate treated $2.24^t$ as if $t$ were a fixed power and scored nothing)? For a two-exponential maximum, do you isolate one exponential term per side and combine the two rate constants before taking logs (Oct 2024 Q8(c); "others took logs at too early a stage")?
5.1–5.2 Integration (integration-by-recognition.ts): Do you write "+c" on every indefinite integral (Oct 2021 Q5(ii), Jan 2022 Q3(i), both confirmed real losses)? Do you keep the modulus sign on any $\ln|f(x)|$ result rather than evaluating ln of a negative (Oct 2020 Q9(c): "left the values as $5\ln(-2)$ and $5\ln(-4)$ not realising these are undefined")? Do you check the booklet before re-deriving a standard result by substitution (Jun 2023 Q9(c), $\int\csc x\,dx$ )?
6.1–6.2 Numerical methods (numerical-methods.ts): Continuity habit above — the single most repeated error in the whole archive. For each new iterate, do you write the value being substituted, not just the calculator's running ANS (Jun 2022 Q8(c): "an answer not implying a correct method... had been shown")? Are you giving each iterate to the exact number of decimal places asked, all the way through the chain (rounding habit above)?

#What NOT to start learning fresh in the last 48 hours

Each item below is something this course's own research has explicitly flagged as rare, low-value at this stage, or a genuinely single-instance finding — not a guess at what's unimportant.

The triple-angle derivation ( $\sin3x\equiv3\sin x-4\sin^3x$ ) as a from-scratch skill. Confirmed on exactly one question in the entire 13-series archive (Oct 2020 Q5(a)). If it isn't already familiar, don't build the general $\sin(2x+x)$ -expansion technique from zero now — spend the time on plain double-angle and harmonic-form work instead, which is confirmed on five series each.
A deep dive into "interpret constant $b$ in words" beyond the reject list above. Confirmed on exactly one series (Jun 2026 Q5(b)) — genuinely new content that no WMA13 lesson had built in at the time the facts bank recorded it. The reject list above is the whole teachable content; don't build a new revision block around a single-series pattern this late.
Every possible combined-transformation shape under spec 1.4. Confirmed in only 2 of 13 series (Jan 2024, Jun 2026), and named as "not usually a standalone question" in the frequency data. The two real, documented traps (partial application; transforming the wrong point) are what's actually been tested — don't spend fresh hours mastering every $af(x+b)$ -style combination beyond those two.
A "deeper" justification for why the $\frac{dx}{dy}=0$ shortcut works, or why the invalid reciprocal-equals-zero route sometimes lands on the same number. The correct, mark-scheme-safe method is a one-line habit already in the per-topic list above; understanding the underlying calculus of why the invalid shortcut coincidentally agrees numerically for some curves is not examined and isn't worth the time.
A brand-new full past paper less than roughly a day out, if you've already run one this week. A rushed, partially-completed paper the evening before produces less signal than a focused pass through the self-test bank, because there's no longer time to properly re-mark it against the real M/A/B mechanics that matter more than the raw score.
Memorising an exact "printed wording" for the formula booklet's $\int\csc x\,dx$ entry. The facts bank itself flags this as unverified as to precise printed form (its presence on the sheet is corroborated by an examiner-report reference, not by a direct re-transcription of the booklet PDF). Know it's recoverable from the booklet rather than needing derivation from scratch — don't spend revision time trying to memorise a "precise" wording this course's own research hasn't confirmed.

#Calibrate against grade boundaries, not against a memorised target

Real, Pearson-published raw-mark boundaries exist for this paper (SIGNAL, independently fetched from Pearson's own grade-boundary PDFs, one per session) — unlike some sibling papers in this course, where no verified figure was found. Two honest limits on how much weight to put on them:

The A* boundary alone has ranged from 58 out of 75 (January 2023, the most leniently graded series checked) to 68 out of 75 (June 2023, January 2024, June 2024) — a 10-mark swing on a 75-mark paper. Don't calibrate your own performance against a specific remembered boundary from one past paper; Pearson recalibrates each series after seeing how it was actually sat.
Grade-boundary documents from June 2025 onward print a second WMA13 row, "Unit 3A," with boundaries a few marks lower than the "Unit 3" row (e.g. June 2026: Unit 3A A*=61 vs Unit 3's 63). What distinguishes the two rows was not established by the underlying research (SIGNAL flags this directly). If you check your own upcoming series' boundary document, don't assume which row applies without checking which unit code matches your own entry.

#Flagged gaps in this checklist

A genuine discrepancy exists between two of this course's own WMA13 documents on October 2021's question count, found while writing this checklist: the facts bank's §3 and §8 both independently state October 2021 had 10 questions, while FORGE's Part 1 timing table states October 2020, October 2021, and October 2024 each had 9. This document deliberately does not resolve that disagreement in either direction — it is why the countdown above says "9 or 10 questions" rather than naming a count for any series not independently re-checked here. Whoever next audits the facts bank or FORGE should re-pull the real October 2021 question paper directly and correct whichever of the two is wrong.
No topic-by-topic time allocation is given (e.g. "spend 20 minutes on numerical methods, 15 on functions") because no pass of this course's WMA13 research has topic-tagged all 75 marks of enough series to support a real weighting — SIGNAL's own frequency table gives typical mark ranges per topic when tested, not a full-paper allocation, and this document inherits that limit rather than guessing a split.
The "confirmed repeat offenders" list is bounded by what has been independently checked so far. A trap appearing on only one or two series in the current archive may recur just as often — the per-topic self-test above still covers those — this section only names the five with the deepest current cross-series confirmation.
Per-question topic-to-number mapping is only confirmed for a handful of individual questions, not for a complete series end-to-end beyond January 2023's mark-total breakdown. Numerical methods and functions tend to open the paper, but "tends to" is not "always" — budget from the tariffs printed on the actual paper you sit.
This document does not address exam-day wellness or logistics beyond equipment checks (venue, arrival timing, general test anxiety) — no document in this course's WMA13 research addresses that territory, and nothing here should be read as covering it.