Pure Mathematics 3

Level Threshold Guide

Provenance (2026-09-29). Primary source: research/veridian/WMA13-verified-facts.md, read in full for §3 (assessment structure), §4 (mark-scheme conventions), §5 (all six per-topic sections), §6 (cross-topic patterns), §7 (topics with standalone-lesson-grade evidence) and §8 (flagged/unverified claims). This document does not re-open any raw Pearson PDF itself — every figure and quotation below traces to that compiled, provenance-checked file, which is where the primary-source re-verification against question papers, mark schemes and examiner reports actually happened. Before writing anything, this pass also read forge-exam-technique-guide.md in full and grepped the rest of content/veridian/technique-hub/wma13/ (all codex-deep-dive-*.md files, codex-formula-and-method-bank.md, forge-battle-card.md, forge-forty-eight-hour-checklist.md, forge-error-pattern-tracker.md, signal-frequency-and-boundaries.md, all ten signal-topic-brief-*.md files) for every mark-scheme anchor this document considered using, specifically to check for duplication before writing — the result of that check is reported honestly in §0 below, because it materially shapes what this document could add. The full lesson files in content/veridian/curriculum/wma13/*.ts were not re-opened line-by-line for this pass; where a claim below rests on what a lesson already covers, that is taken from the relevant CODEX deep-dive's own account of having read that lesson in full, not re-checked independently here. content/veridian/question-bank/wma13.ts was checked directly (two items, both already cited in forge-exam-technique-guide.md). Mark-scheme and examiner-report wording is paraphrased throughout; direct quotations are kept under fifteen words each and attributed to their series. Not affiliated with or endorsed by Pearson Edexcel.

#WMA13 (Pure Mathematics 3) — Level/Threshold Guide

#0. What this document is, honestly, and what it isn't

This paper has no holistic level descriptors — nothing like a five-band essay rubric where an examiner judges a whole answer against a paragraph of qualitative criteria. WMA13 is marked atomically: every question is broken into M (method), A (accuracy, chained to a specific M), and B (independent) marks, occasionally deepened into dM/ddM (a method mark itself dependent on one or two earlier method marks). The real "level/threshold" question for this paper is therefore never "which band does this answer sit in" — it is always "does this specific line of working cross this specific mark's own boundary, or not." That is what this document maps.

A fact worth stating before anything else, because it shapes everything below: by the time this document was written, this course's WMA13 technique-hub already contained thirteen other files, and between them they had already mined essentially every real mark-boundary anchor sitting in the primary verified-facts source — forge-exam-technique-guide.md alone works through the Oct 2024 product/chain-rule dM1, the Oct 2024 exponential-maximum dM1, the Oct 2023 modulus-log ddM1, the Jan 2022 "reject the extra solution" A-mark, the Jan 2023 range-notation B-mark, and four more; the per-topic CODEX deep-dives independently re-derive the Jan 2022 cosec B-mark, the Jan 2025 "show D = 0" B1*, and the full Oct 2020 double-angle equation's M/A/dM ladder. Re-presenting any of those as if freshly discovered here would not be additive — it would be relabelling. So this document does two things instead, and only two:

Reorganises the paper's real mark-boundary evidence by boundary type rather than by spec topic — a cut no existing WMA13 document takes. Every other document in this folder answers "what does topic X require"; this one answers "what, mechanically, turns a 0 into a 1 at an M-mark, an A-mark, a B-mark, and a printed-answer mark," pulling one clean illustration of each from across the whole paper regardless of which spec area it happens to sit in. Where the fullest worked treatment of an anchor already lives elsewhere, this document says so and points there rather than re-deriving it.
States one comparison that the topic-organised documents structurally cannot show, because each stays inside one spec area: the same underlying slip, examined in two different series, can sit on two different sides of the M/A boundary — because the threshold is set by that specific scheme's own wording, not by some fixed severity attached to the error itself. §3 below is the one genuinely new piece of analysis in this document; everything else is real evidence, honestly re-cut.

What this document does not do, on principle. It does not state or imply a raw-mark grade boundary anywhere — that is a real, separately-sourced, Pearson-published figure and it already has its own honest home in signal-frequency-and-boundaries.md §2. It does not attach a likelihood or percentage to any pattern; every frequency claim below is a plain count of series checked, stated as such. And where the underlying research file itself flags a claim as unverified or corrected from an earlier draft, that flag is carried over rather than smoothed away.

#1. The M0 → M1 boundary: a recognisable attempt, not a recognisable answer

Pearson's own general marking guidance defines an M mark as credit for "a correct method or an attempt at a correct method" — deliberately not for a correct answer. The boundary this actually draws in practice is narrower than it sounds: the method has to be written down in a form the scheme can recognise as that method, not merely arrived at.

The cleanest real illustration, drawn from Oct 2020 Q1 (solve $2\cos2x=7\cos x$ for $0°\leqslant x<360°$ — the full ladder for this question is already built out in codex-deep-dive-compound-and-double-angle-formulae.md §3.B; only the M-boundary itself is the point here). Substituting $\cos2x\equiv2\cos^2x-1$ should turn the equation into $2(2\cos^2x-1)=7\cos x$ .

Just misses: a script that writes $2\times2\cos^2x-1=7\cos x$ — the bracket around $(2\cos^2x-1)$ missing, so the " $-1$ " is not being multiplied by 2 at all, silently changing the equation. The real mark scheme scores this M0, unless the identity was stated correctly on an earlier line or the candidate visibly recovers it.
Just clears: the same substitution written with the bracket intact, $2(2\cos^2x-1)=7\cos x$ . Nothing else about the working needs to be correct yet — this line alone earns the M1, and everything after it is free to go wrong without touching this specific mark again.

The boundary is not "did the candidate know the double-angle identity" — most who make this slip did know it. The boundary is whether the line as written is distinguishable from a different, wrong equation. This is the concrete mechanism behind the paper's single most repeated piece of technique advice — quote the formula or identity as its own line before substituting into it (forge-exam-technique-guide.md §2.3) — because a scheme reading a candidate's working from the outside cannot credit an identity it cannot see used correctly.

A second, structurally different M-boundary worth naming once: whether a method mark can be earned by implication, with no formula ever written. The general marking principles allow this — "the method mark can be gained by implication from correct working with values" — but only when the numbers themselves are unambiguous evidence of the right process; the moment there is a mistake in that working, the implied mark is the first thing lost, because there is no quoted formula-line left to fall back on. The Oct 2020 Q3 quotient-rule finding already carries this exact trade-off (forge-exam-technique-guide.md §2.3) and is not reworked here — it is named to make the general shape of this boundary explicit: an unwritten method is provisionally credited, on loan against the arithmetic staying perfect.

#2. The A0 → A1 boundary: the number is right; something else about it isn't

Because every A mark is chained to an M mark, an A-boundary can only be crossed after the M-boundary already has been. What actually separates A0 from A1, across this paper's real questions, is never "is the number correct" alone — it is one of a small number of recurring conditions attached to the number. Three real, distinct conditions, each already fully sourced elsewhere and restated here only under this heading, to show they are the same kind of boundary wearing different topic clothing:

Completeness, not correctness (Jan 2022 Q2(b), full ladder in codex-deep-dive-reciprocal-and-inverse-trigonometric-functions.ts and forge-exam-technique-guide.md §5.2a): the question asks for "the smallest positive solution." A script that correctly finds both $24.3°$ and $180°+24.3°$ and states both has done strictly more correct mathematics than a script that finds and states only $24.3°$ — and scores less, because the A-mark's own wording is " $\theta=$ awrt $24.3°$ ONLY." Just clears: one value, the right one, alone on the line. Just misses: the right value present, but not alone.
The right variable, not just the right number (Jan 2023 Q1(a), cited in forge-exam-technique-guide.md §5.2c): the range of $f$ is accepted written as $y\leqslant9$ or $f\leqslant9$ , but the scheme explicitly refuses " $x\leqslant9$ " — a numerically perfect statement, in the wrong letter. A range describes outputs; writing it in the input variable crosses back over the boundary into A0 however correct the number 9 is.
The right form, not a numerically equal one (Oct 2020 Q7(a), the harmonic-form R/α marks; full ladder in codex-deep-dive-harmonic-form.md, the exact-answer convention itself in forge-exam-technique-guide.md §2.6): $\alpha=$ awrt $1.326$ (radians) is A1; the numerically identical $\alpha=$ awrt $75.96°$ is A0, because the question fixed the required unit and the scheme enforces that literally.

The pattern under all three: an A-mark is never just "this number." It is "this number, satisfying the one extra condition the question's own wording or the scheme's own convention attaches to it" — and that condition is different every time, which is exactly why it is the mark most reliably lost by an otherwise-strong script.

#3. The B0 → B1 boundary: independent of method, not independent of precision

A B mark needs no method line at all — but "no method required" is not the same as "any plausible-sounding statement suffices." Two real boundaries, of different shapes:

A bare identity statement is enough, until a coefficient is folded into it wrongly. Jan 2022 Q2(a) (full extraction in research/veridian/WMA13-verified-facts.md §5, cross-referenced in codex-deep-dive-reciprocal-and-inverse-trigonometric-functions.ts) awards its opening B1 for simply writing or using $\text{cosec}\,\theta=1/\sin\theta$ — the examiner report notes this mark is earned "even by otherwise weak scripts." But the same question's real mark scheme scores $3\,\text{cosec}\,\theta=1/(3\sin\theta)$ as B0 unless the plain identity is separately recovered. Just clears: the coefficient stays outside the reciprocation, $3\times(1/\sin\theta)$ . Just misses: the coefficient is folded inside the sine, changing what is actually being reciprocated. This is a different failure from confusing cosec with sec — the identity chosen is the right one; it is applied to the wrong quantity.

A single B-mark can bundle several required components, none individually sufficient. The numerical-methods sign-change mark — the single most repeated specific loss in the whole archive, per research/veridian/WMA13-verified-facts.md §5.6 and already the anchor of the "say continuous out loud" habit in forge-exam-technique-guide.md §6 and forge-forty-eight-hour-checklist.md — is not one fact but three, all required together: both function values evaluated correctly, a reason that names continuity and indicates a sign change, and a stated conclusion such as "hence a root." A script with the correct values and "sign change, so a root exists" — continuity never named — still crosses back to zero on this mark, series after series, because "sign change" alone is not what the scheme's own wording requires; it is one of three named components, not a paraphrase of all three.

A worked, if under-quoted, third case: near-miss wording is not partial credit. Jun 2026 Q5(b) (the one located anchor for interpreting a fitted growth constant $b$ in words; the full reject list already lives, verbatim, inside the exponential-growth lesson's own trap taxonomy and is summarised in codex-deep-dive-exponentials-and-logarithmic-graphs.md §5.5) requires three things stated together in one sentence — a direction word, the percentage to 2 s.f., and a per-year time unit. The scheme's own reject list shows how close "close" can still be zero: a statement with the right number and the right time unit but no direction word is not credited, and neither is one that states "the annual increase is 119%" — mistaking the new total as a percentage of the old ( $b\times100$ ) for the increase itself ( $(b-1)\times100$ ), a one-symbol arithmetic slip that changes which side of the boundary the whole sentence sits on. This B-mark's boundary is not "roughly right" — it is "all three named parts present, and the one number that looks similar to the right one is not interchangeable with it."

#4. The printed-answer boundary: every line has to be right, not just the destination

A handful of marks in this paper — flagged in mark schemes with an asterisk, B1* or A1* — attach to a question that states its own answer up front ("show that D = 0," "show that $\sin3x\equiv3\sin x-4\sin^3x$ ") and asks the candidate to justify a target already printed on the page. This is the strictest boundary type on the paper, because the destination is never in doubt — only the path is being graded, so there is nothing left to credit except a genuinely complete derivation.

Just misses, concretely (Jan 2025 Q4(a), sourced in full in codex-deep-dive-rational-expressions-and-algebraic-division.ts): dividing $\dfrac{4x^3+2x^2+3x+8}{x^2+4}$ to the form $Ax+B+\dfrac{Cx+D}{x^2+4}$ and asked to show $D=0$ , a script that correctly finds $A$ , $B$ and $C$ and then simply writes " $D=0$ " — true, and consistent with everything found so far — earns nothing on this specific mark. The real mark scheme's own wording is that it must be "shown that D = 0 from clear and correct work," which in practice means setting up the coefficient-matching identity (or completing the division) and substituting to derive the zero, not asserting it as the leftover once the other three constants are known. Just clears: the same three constants, plus the equation that forces $D=0$ written out and solved, however briefly.

A second, distinct printed-answer boundary — the "correct maths, wrong notation" case (Oct 2020 Q5(a), the triple-angle proof; the ddM1 dependency structure inside this same question is already the subject of forge-exam-technique-guide.md §2.2 and is not reworked here — this is a different mark on the same question). After the double-angle substitutions are done correctly, the final A1* on a "show that" line is a cso mark — "correct solution only," meaning no error anywhere in that part is tolerated, including cosmetic ones. The verified-facts file is careful to flag that only one fragment of this note survives extraction cleanly, but that fragment is unambiguous: writing "sin" where the line requires "sin x" is enough, on its own, to lose this mark, even when every algebraic step is otherwise flawless. Just clears: identical algebra, with the variable written out on every line. Just misses: the same algebra, with the variable dropped once. Nothing about the mathematics separates the two scripts — only the printed-answer mark's own zero tolerance for an incomplete line does.

The general rule these two anchors share, stated once: on any mark tagged with an asterisk, the target number or identity was never the graded content — it was given. The boundary sits entirely inside the path, and a path with one derivation step skipped or one notational lapse is, mechanically, a different path from a complete one, regardless of how obviously "the same" the two look to the candidate who wrote them.

#5. Dependent-chain boundaries (dM1, ddM1): counted per question, not fixed by topic

This boundary type is the one place this document adds least, because it is already the best-documented mechanism in the whole technique-hub — forge-exam-technique-guide.md §2.2 works through a singly-dependent case (Oct 2024 product-and-chain-rule) and a doubly-dependent case (Oct 2024's exponential maximum, cross-checked against Oct 2023's modulus-of-a-log question and Oct 2020's triple-angle proof) in full, and codex-deep-dive-modulus-and-graph-transformations.md restates the counting rule again. Stated once more, only because it is a genuine boundary type and this document would be incomplete without naming it: a step is dM1 if exactly one independent method mark genuinely precedes it in that specific question, and ddM1 if exactly two do — the label is set by counting the question's own preceding method marks, never by which spec topic the question belongs to. The one thing worth adding under this heading, rather than repeating the worked examples themselves, is the failure mode this boundary produces when misjudged: grafting a real ddM1 quotation from one question onto a simpler worked example that only has one method mark preceding its own equivalent step overstates what that example's own scheme would actually require — a documented, corrected error in this course's own material (research/veridian/WMA13-verified-facts.md §4), and the reason the counting has to be redone per question rather than recalled as a fact about a topic.

#6. The one comparison a topic-by-topic document cannot show: the same slip, two different thresholds

Every section above restates a real boundary using an anchor that (mostly) already exists somewhere in this technique-hub. This section is the one place this document is not simply re-cutting existing evidence.

Two series, two different models, one identical underlying error: differentiating an exponential term with a negative exponent and losing the minus sign the chain rule attaches to it.

Jan 2022 Q8(c), model $x=400\times1.4^{-t}$ : the mark scheme's own structure treats a dropped sign leniently. A candidate who drops only the sign but otherwise differentiates and substitutes correctly still earns the method mark for the process itself — the loss lands entirely on the final accuracy mark, a B1 M1 A0 outcome. (A separate, more severe error on the same question — treating the exponent as though it were a fixed power and reducing it by one, rather than using $a^x\to a^x\ln a$ at all — scores nothing at all; that is a different mistake from the one being compared here, not a stricter version of it.)
Jan 2024 Q5(c), model $T=10+8e^{-Bt}$ : the identical sign-drop, on the identical family of model, is folded directly into the method mark's own definition. The scheme states plainly that losing the minus sign in the exponent "scores M0" — not the final answer mark, the method mark underneath it, taking every accuracy mark chained to it down with it.

Why this matters more than either fact alone. A candidate (or a lesson) that has internalised "dropping the sign on a negative exponent costs one mark" from the Jan 2022 case would walk into Jan 2024's version of the same slip and lose the entire step. The threshold is not a property of the error — "how bad is this mistake" is not a fixed quantity on this paper. It is a property of how that specific scheme chose to write that specific M-mark: whether the sign is folded into the method's own definition, or left to be checked only once, later, at the accuracy mark. Jan 2025 Q5(c) supplies a third, weaker data point in the same family — the verified-facts file is explicit this is a single, less-attested instance, not a third confirmed threshold — where the identical sign is lost during rearrangement rather than differentiation, and simply produces a negative final answer with no note on how severely the scheme penalises it.

The practical consequence for revision is the opposite of "memorise how many marks each mistake costs." It is: the general shape of a trap (here, a negative exponent's sign) is stable and worth knowing on sight; the exact cost of falling into it on any specific real question is not transferable from one series to another, and has to be checked against that question's own scheme rather than assumed. No other document in this folder states that limitation explicitly, because every other document works one scheme at a time and so never has occasion to compare two.

#7. What is genuinely left open

This document's own boundary-type taxonomy (M0/M1, A0/A1, B0/B1, printed-answer, dependent-chain) is a reading of the real evidence gathered so far, not a claim that these are Pearson's own named categories — Pearson's general marking guidance names M, A, B, dM/ddM, ft, bod, cao, cso, isw and awrt; the five-type grouping above is this document's own organisation of that real vocabulary, offered because it is useful, not because it is official.
§6's comparison rests on two series plus one weaker third instance — a real, sourced pattern, not a large sample. A future pass that locates a third or fourth negative-exponent differentiation question with a fully transcribed mark scheme would meaningfully strengthen or complicate this section, and should be added there rather than treated as confirmed at the current count.
Every worked anchor in §§1–5 was chosen because it already had a complete, independently-checked mark ladder somewhere in this course's material — this document did not go back to a raw mark-scheme PDF to locate a boundary case that isn't already documented, because (per §0) the primary source had already been mined close to exhaustively by the time this pass started. If a genuinely new anchor surfaces in a future series, it belongs first in the relevant topic's own CODEX deep-dive, and could then be added here under whichever boundary type it illustrates.

#Sources

research/veridian/WMA13-verified-facts.md — §3 (assessment structure), §4 (mark-scheme conventions, verbatim general marking guidance), §5 (Oct 2020 Q1, Q5(a), Q7; Jan 2022 Q2, Q8(c); Jan 2024 Q5(c); Jan 2025 Q4(a), Q5(c); Jan 2023 Q1(a); Jun 2026 Q5(b)), §6 (cross-topic patterns), §8 (flagged/corrected claims, the ddM1 misapplication caution).
content/veridian/technique-hub/wma13/forge-exam-technique-guide.md — the M/A/B system, dM1/ddM1 worked examples, the "quote the formula" and exact-answer conventions this document cross-references rather than re-derives.
content/veridian/technique-hub/wma13/codex-deep-dive-compound-and-double-angle-formulae.md, codex-deep-dive-harmonic-form.md, codex-deep-dive-reciprocal-and-inverse-trigonometric-functions.md, codex-deep-dive-rational-expressions-and-algebraic-division.md, codex-deep-dive-exponentials-and-logarithmic-graphs.md — full worked ladders for the anchors named above, pointed at rather than reproduced.
content/veridian/question-bank/wma13.ts — checked directly; contains no anchor beyond the two already cited in forge-exam-technique-guide.md.