Technique Masterclasses

Pure Mathematics 3

Examiner Mindset Guide

Provenance: written 2026-09-29, grounded entirely in research/veridian/WMA13-verified-facts.md (§3, §4, §5, §6 — the primary, provenance-checked source; every quote below was re-located in that file, which itself states each was matched against the actual pdftotext-converted PDF of the named mark scheme or examiner report before being recorded there). Read first against content/veridian/curriculum/wma13/ (the pedagogical lessons — nothing here duplicates their worked content) and against everything already in content/veridian/technique-hub/wma13/, in particular forge-exam-technique-guide.md §2 (the M/A/B mark-coding mechanics, dM1/ddM1, "quote the formula first," ft, misreads, abbreviations — all already built out there in technique-for-students form) and signal-examiner-intelligence.md (the rhetorical register of examiner reports across series). This document deliberately does not restate either — see "What this document is, and isn't" below for exactly where the line is drawn. All quotations are kept under ~15 words with attribution, per this course's standing rule against large verbatim reproduction of copyrighted exam-board text; nothing below is a large lift from any primary PDF. No likelihood percentage, grade boundary, or "confirmed in every series without exception" claim is asserted anywhere in this document beyond what a cited source states for that one named series.

WMA13 (Pure Mathematics 3) — Examiner Mindset Guide

What this document is, and isn't

Two documents already in this folder cover adjacent ground, and this one is written to sit beside them without repeating either:

What is left for this document is the question the brief actually asks: is WMA13 marked holistically or by discrete point-tallying (checked against the primary sources, not assumed), what does a marker check first, and what does examiner-report commentary reveal about that process actually running on real scripts. That is the whole of what follows.

1. The real marking architecture: discrete point-tallying, not holistic best-fit — checked, not assumed

Two broad philosophies exist for marking extended written work in general: a holistic, best-fit approach, where a whole response is weighed against level descriptors and placed in a band; and a discrete, analytic approach, where specific, named steps each carry their own independent mark, and the final score is the arithmetic sum of which of those named steps were actually present. These are genuinely different processes — under best-fit marking, a script's overall quality and coherence can move it between bands even where no single step is separately ticked; under discrete marking, quality and coherence do nothing on their own unless they happen to coincide with a named, mark-carrying step.

WMA13 is discrete, not holistic — verified directly, not inferred from general reputation. Three independent lines of evidence, each traceable to a named primary source:

  1. The paper's own total is built from itemised, fixed-tariff questions that must sum exactly. The spec states the paper is "out of 75 marks" (verbatim, spec p.21) but does not fix how many questions deliver that total — verified directly against real question papers, January 2023 split its 75 marks across 10 questions (6+6+5+7+9+8+9+5+11+9), checked by hand against the printed per-question tariffs; October 2020 used 9 questions instead. Either way, the total is not an examiner's overall impression of the paper — it is a fixed arithmetic sum of independently-tariffed questions, each of which is in turn built from independently-tariffed marks.
  2. Every mark within a question is typed and gated by explicit rule, not weighed against a descriptor. The general marking guidance printed at the start of every scheme reviewed defines exactly three mark types and states the gating rule verbatim: "M0 A1 is impossible" — an accuracy mark literally cannot exist without its own named method mark first being separately awarded. A B mark, by contrast, is deliberately independent of any method: "independent accuracy marks where there is no method." There is no fourth category, and no mechanism anywhere in the reviewed schemes for a mark to be awarded on overall impression rather than a named trigger.
  3. The archive contains zero evidence, across 8 series read in full for content and 11 for mark-scheme convention, of a level-descriptor or best-fit judgement anywhere in a WMA13 scheme. This is worth stating as a checked absence rather than an assumption: every single mark cited anywhere in the verified-facts bank underlying this hub is a discrete M, A, dM, ddM, or B, tied to a specific line of working or a specific stated fact — never a banded judgement of the response as a whole.

Why this matters beyond terminology. Because credit is tied to named steps rather than overall coherence, a script's actual mark total can diverge sharply from how "good" it reads. The clearest single proof of this in the archive is a "show that" question, where the final numerical or algebraic answer is printed on the question paper itself — the candidate already knows what they are supposed to reach. On such a question, the final answer carries no marks of its own at all; every mark is for the justification that gets there. Verified verbatim, a rational-division "show that" question (spec 1.1, Oct 2020 Q9(a)) where a candidate is asked to show a remainder equals 5: the mark scheme's last two marks are M1 for writing the expression in the required factorised form and A1 for the final line, explicitly annotated "Q = 5 is given so it must be shown from correct work, not just stated." A candidate who writes the correct value 5 with the working skipped scores neither mark, despite stating a mathematically true fact — because on a discrete scheme, a true statement with no corresponding named step behind it earns nothing. A holistic reader might credit "the candidate clearly knows the answer"; a discrete scheme, mechanically, cannot.

2. What a marker checks first, in the order the mark scheme itself enforces

An honest limit before this section starts: the primary-source archive underlying this hub contains the printed general marking guidance that opens every WMA13 mark scheme — the document a marker is actually handed and instructed to apply — but it does not contain any account of the separate human training process around it (standardisation meetings, marker recruitment, live calibration against seeded scripts). No sentence below should be read as describing that human process, because no source for it was located. What follows is built only from the printed guidance itself, treated as exactly what it is: the written instruction a marker is trained against, not a narrative reconstruction of a marking day.

Read in that light, the printed guidance sets a strict order of operations, not a flat checklist:

First: is a method identifiable at all — before the final answer is looked at. Because "M0 A1 is impossible" is a hard gate, not a guideline, a marker's first question on any line of working is not "is this correct" but "is a correct method, or an attempt at one, present here." The M-mark definition is deliberately generous on this point — verbatim, "a correct method or an attempt at a correct method" — so a marker's first pass credits an identifiable line of reasoning even where the numbers afterwards go wrong. Only once that gate is passed does the accuracy mark chained to it become checkable at all.

Second, for Pure Mathematics specifically: a small set of generic, topic-independent triggers, applied before the question's own specific numbers are engaged with. The general principles printed for Pure Mathematics marking (verified verbatim, appearing under "General Principles for Pure Mathematics Marking" in every scheme reviewed) name the method-mark trigger for differentiation as "power of at least one term decreased by 1," and for integration as "power of at least one term increased by 1" — a marker checks for that structural signature first, independent of which specific function is being differentiated or integrated, before assessing whether the resulting coefficients are right. The same section separately defines, in comparable generic terms, what counts as a valid method for solving a 3-term quadratic (factorising, the formula, or completing the square). These are the literal, generic pattern-matches a marker is instructed to look for before turning to a question's own particular content.

Third: whether a method mark not explicitly written out can still be inferred — and, if so, how far the marker is instructed to keep checking before granting it. The guidance's own wording is explicit that an unquoted formula does not automatically fail: "the method mark can be gained by implication from correct working with values." But it immediately narrows that generosity: it "may be lost if there is any mistake in the working" that follows. Read from the marker's side rather than the student's, this is an instruction to keep reading forward past the point where the mark would already be earned if quoted — an unquoted method claim is provisional until the marker has checked every subsequent line for an error that would retroactively withdraw it. A quoted formula, by contrast, closes that check immediately: the mark is secured at the line it appears on, regardless of what goes wrong afterwards. This is the same underlying rule forge-exam-technique-guide.md §2.3 already gives as advice to a student ("quote it first"); stated here from the other side of the page, it is a real fact about how much extra reading a marker has committed to doing on every unquoted method line in a script.

Only after those checks does a specific numeric or symbolic accuracy mark get evaluated against its own named acceptance criteria — and, per §4 of the verified-facts bank, that evaluation is itself frequently a checklist of separately-required components rather than a single correct/incorrect judgement. The clearest verified example of this final layer is the paper's own root-location question (spec 6.1), where the accuracy mark for justifying a sign change is defined, verbatim, as requiring three separately-named components in the same line: "both f(1) and f(2) correct... reason, which must mention continuity and state or indicate sign change in some way... conclusion, 'hence root'" (Oct 2023 Q1(a) mark scheme). A marker checking this line is not asking "did the candidate show they understand the idea" — they are checking off three specifically named sub-components, and the mark is withheld if any one of the three is missing, independent of how clearly the other two are argued. This is the discrete architecture from §1 operating at its most granular: even a single mark can itself be a small, itemised list.

3. Marking in practice: what examiner-report commentary reveals about the mechanism running

Three real, cited fragments — each attributed to a named series, each under 15 words as quoted — that show this discrete architecture visible in the examiners' own account of a live series, rather than only in the printed scheme:

1. The M-then-A chain is not just a rule on paper — examiners describe candidates' scores as literally following from it. On a log-manipulation question (spec 3.3, June 2023 Q2), the report states plainly: "If the method marks were achieved, then the answer was usually correct" (June 2023 Principal Examiner Feedback). This is an examiner describing, in their own summary of a real cohort's results, the exact mechanical consequence of "M0 A1 is impossible" from §1 — not as an abstract rule, but as an observed fact about how that series' marks actually distributed. Where the same report locates the point of failure, it is equally specific rather than general: the dependent method mark, not the surrounding steps, is named as where scripts split.

2. A B mark's "independent accuracy, no method" status is enforced against a specific list of acceptable and unacceptable wordings, not a judgement of whether the idea came across. On a temperature-modelling question testing why an equation has no solution (spec 4.4, January 2024 Q5(d)), the mark scheme names three acceptable phrasings and explicitly excludes two named incorrect ones; the examiner report then confirms the rejection was applied as written: "Common incorrect answers were to give the minimum value as 8 or 18" (January 2024 Principal Examiner Feedback). A candidate offering either of those two numbers had engaged with the model and produced a plausible-sounding boundary value — the kind of response a holistic reader might partially credit for "understanding the shape of the problem." A discrete B mark has no such middle ground: the wording either matches a named acceptable form or it does not, and the report's own account of that series confirms the mark was withheld exactly that cleanly.

3. Even a "show that" question's marking is described by examiners in terms of the missing step, not the missing understanding. On the same rational-division question already cited in §1 (October 2020 Q9(a)), the report's closing sentence states the actual marking principle directly rather than just describing an error: "it is important to justify the answer" (October 2020 Principal Examiner Feedback). The report is not saying candidates didn't know the remainder was 5 — most demonstrably did, since they wrote it down. It is stating, in the examiner's own words, exactly the discrete-marking fact from §1: on this kind of question, knowing the answer and being credited for it are two separate things, and only the visible justification is what the scheme can actually reward.

4. Honest limits

Sources