Examiner Intelligence

Provenance: compiled 2026-09-29. Source is research/veridian/WMA12-verified-facts.md in full (§1–§9, read start to finish, including the Lesson-audit log's own primary-source re-extractions) — every quote below was already matched against a real Pearson PDF by that document before this pass; no new primary-source fetch was done here. This document was written only after reading the other nine WMA12 technique-hub files in this folder (forge-exam-technique-guide.md, codex-formula-and-method-bank.md, signal-frequency-and-boundaries.md, forge-battle-card.md, forge-forty-eight-hour-checklist.md, forge-exam-day-protocol-extended.md, forge-level-threshold-guide.md, the eleven codex-deep-dive-*.md files and eleven signal-topic-brief-*.md files) and content/veridian/question-bank/wma12.ts, specifically to avoid restating what any of them already cover. None of those asks how the report itself talks — what phrases recur, what tone recurs series after series, and what that implies about how a real script is actually read. That is this document's only subject, matching the sibling signal-examiner-intelligence.md already written for WMA11/WBS11/WBS12/WBS13 in this same course. Direct quotations below are kept under roughly fifteen words each, with the series and question attached; anything longer is paraphrased. No likelihood percentage or grade boundary is stated anywhere below — where a source itself states a number (an examiner report's own "47% scored 0/4"), it is quoted as that report's own words about that exact question in that exact series, never converted into a course-wide or future-series claim. Not affiliated with or endorsed by Pearson Edexcel.

#WMA12 — Examiner Intelligence

#Pearson Edexcel IAL Pure Mathematics 2 — how the report itself talks

#0. What this is

Every other document in this folder tells you what the examiners found — a trap, a mark allocation, a frequency count. This one asks a different question: across the series this course has actually read, does the report's own language repeat? Not the mathematics itself — the phrasing, the emphasis, the moves a report habitually makes that a mark scheme alone would never tell you.

The answer, checked below, is yes, in a specific and countable number of ways. Eight patterns are each grounded in three or more independently-cited series (one exception is noted honestly where it sits at exactly two) and labelled CONFIRMED. Two more are real, quotable and specific, but found only once in this pass — those are labelled SINGLE FINDING and left that way.

Denominator, stated up front (matching signal-frequency-and-boundaries.md's own convention): the facts bank has a question paper and mark scheme for 20 series (June 2019–January 2026) and a published examiner report for 17 of those. This document draws quoted report language from 13 of those 17 sittings: June 2019, October 2019, January 2020, October 2020 (read but not quoted below), January 2021, October 2021, January 2022, June 2022, January 2023, October 2022, October 2023, January 2024, January 2025, and June 2025 (plus its "01a" regional paper and June 2024's "01R" regional paper, quoted once each). Four series with a real, confirmed-present report — June 2023, June 2024's main paper, October 2024 — were not read at the depth needed to mine report language in this pass and contribute nothing below. June 2021 has no report at all (confirmed absent, not merely unread); October 2025 and January 2026 are too recent for one to exist yet. A pattern below citing four series is a real majority of the quoted evidence, not a cherry-picked pair.

---

#1. Confirmed patterns

#1.1 A fixed, graded vocabulary of soft quantifiers is the default register

A WMA12 report almost never gives a number. It reaches for the same small set of hedge words, and the same words recur, sitting after sitting: "most candidates" (June 2019 Q1, June 2022 Q3(i)), "many candidates" / "a significant number" (October 2022 Q1), "a surprising proportion" (January 2020 Q8(i)), "a fairly common mistake" / "a very common mistake" (January 2023 Q4), "some" / "a few" (June 2025 Q8). Six independently-dated series, one shared lexicon. This is not a coincidence of subject matter — WMA11's own sibling document in this course found the identical habit for Pure Mathematics 1, confirming it as a house style of this exam board's report-writing generally, not an accident of which WMA12 series this course happened to read.

#1.2 Hard numbers are reserved for the single most discriminating question of a series

When a report does drop the hedge words for an actual statistic, it is almost always on the one question that series' own commentary treats as the paper's real discriminator — and it then gives several numbers in a row, not one. January 2021 does this four separate times in the same report: Q3(i) drops from "nearly 90% to just over 70%" between its first two marks (only 50% scored all three); Q3(ii) shows "about 80%" using one log law correctly but "only 70%" clearing the equation; Q9 shows "nearly 90%" clearing the first two parts against "fewer than 60%" on the next; Q2 records "only 50% scored" the closing mark. Two further, independent series confirm this is a genuine cross-series habit rather than one report-writer's style: October 2021 Q9 states plainly that "47% scored 0/4", its worst-performing question that series, and January 2022's Q8(d) is independently described as "the first of three discriminating questions at the end of the paper" with a stated mean score of roughly 3.4 out of 9. Three series, one consistent pattern: the ordinary register is qualitative, and a switch to hard numbers is itself a signal — it marks the question the report wants a reader to pay closest attention to.

#1.3 A named wrong strategy is priced at flat zero — distinct from a slip inside the right one

Reports draw a hard, explicit line between an execution error inside a valid method (which keeps partial credit elsewhere) and choosing an entire strategy the mark scheme does not credit at all, and they say so directly. June 2022's proof report states that algebraic approaches to a counter-example question "score no marks unless they substitute numbers" — the technique itself, not a slip within it, is what forecloses credit. January 2024's Q5(i) is explicit that a calculator-derived sum-to-infinity value earns nothing: "no marks were scored" even when the number was correct. January 2023's Q9(c) names the same cliff for an area question: integrating the curve alone "gained no credit as it was not a correct strategy" for that region. A specific instance of this same cliff recurs on its own across three separate series — candidates defaulting to "integrate the curve alone" and ignoring a bounding line entirely, independently confirmed in January 2020 Q4, January 2023 Q9, and June 2025 Q3 — which is itself evidence that this is a standing category of failure the reports watch for by name, not a one-off.

#1.4 The report localises the discriminating step to one exact sub-operation, never the whole question

Rather than saying a topic was hard, a WMA12 report typically names the single specific move inside a multi-step question that actually separated candidates, while confirming everything around it was fine. January 2021's Q9 states that "nearly 90%" of candidates cleared the circle's opening parts, and it was identifying the centre specifically — one sub-step — that then dropped success below 60%. January 2022's Q8(d) mixed-series question is not hard because either the arithmetic or the geometric formula is hard; the report's own commentary (cited in §6.4 of the facts bank) isolates the single "overlap term" between the two halves as the discriminator. October 2023's Q6 trapezium-rule report states directly that candidates "found extracting information from a graph much harder than extracting it from a table" — the discriminator is the input format, not the trapezium formula itself, which the same report treats as understood. Three series, three different topics, the same rhetorical move: point at one sub-skill, not the subject.

#1.5 Equally-credited methods are named as empirically unequal — as commentary on candidates, not on the mark scheme

Several WMA12 questions accept two or more genuinely different valid routes, and the reports repeatedly volunteer which one real candidates handle worse, even though the mark scheme itself does not penalise the choice. January 2024's Q1 states plainly that candidates "who chose to use division gave themselves a more difficult task" than the remainder theorem, and were less successful for it. January 2021's Q9 notes that candidates attempting implicit differentiation on the circle "often made little progress" despite the method marks being available. January 2023's Q6 traces a harder algebra path directly to which variable a candidate chose to substitute for. June 2025 confirms the same habit twice in one sitting: the main paper's Q1 notes candidates mostly reached for the more direct combinatorial binomial form rather than the general series form, and the regional "01R" paper's Q7 states outright that candidates "struggled to make a start" with the harder, general form. Four independent series (one contributing two papers): the reports are not just recording that a method is credit-worthy, they are recording, unprompted, which credit-worthy method actually works under exam conditions.

#1.6 A correct-looking final line is not credited without the specific shown method — and calculators are the report's own named suspect

Distinct from 1.3 (a whole wrong strategy) and 1.5 (two right strategies of unequal difficulty), this pattern is about a right answer reached invisibly. October 2019's Q10 records that candidates "simply stated there were no solutions, without showing how they knew this" — a true conclusion, unscored, because the discriminant working the mark scheme actually asks for was never shown; the same report attributes this directly to over-reliance on calculators. January 2020's Q4 makes the identical point for a definite integral: a calculator-only exact value scores nothing where the exact-answer rule applies. January 2024's Q5(i) repeats it for a sum to infinity: a calculator's summation button can reach the right number and still score zero. Three series, one repeated diagnosis. Pearson's own general marking guidance corroborates that this is a recognised, standing theme rather than an inference of this document's own: its "quote the formula first" clause states that this is "recent examiners' reports" — plural — telling candidates the formula should be written before the numbers.

#1.7 The report occasionally steps outside diagnosis into direct, forward-looking advice — in its own recognisable register

Most report prose is retrospective ("candidates did X"). A smaller, distinct register is explicitly prescriptive, telling a future candidate what to do differently, and it recurs across series that are otherwise unrelated in topic. January 2020's Q8(i) closes its sum-formula-proof commentary with the advice that "students would be well advised" to know the standard proofs. October 2019's Q10 closes on the same register for "show that" rigour generally, naming that exposure to proof rigour "would be of assistance" to struggling candidates. January 2021's Q9 volunteers, mid-diagnosis, that "it is advisable for candidates to use a sketch" before starting a circle-geometry question. Three series, three different topics, one shared move: the report pauses its own retrospective account of what happened and speaks directly to the reader about what to do next time.

#1.8 The report explicitly names a question's own novelty or familiarity as an explanatory factor

A distinct habit, confirmed across three series, is commentary on the question's own design — not the candidates — as part of the explanation for how it was answered. October 2023's exhaustion question is introduced as "a familiar context for a question", immediately followed by many candidates scoring full marks. January 2025's trapezium-rule question is introduced the opposite way — "a novel way of testing the trapezium rule via logarithms" — as the report's own explanation for why it discriminated. June 2025's Q9 circle question gets the same framing again: "another unexpected question that candidates did not seem fully prepared for". Three series, the same explanatory habit: before describing what candidates got wrong, the report first places the question itself on a familiar-to-novel scale, as if that scale is doing real explanatory work of its own.

---

#2. Single findings — real, but not (yet) a pattern

• The report explains its own marking accommodation directly, not just what candidates did. January 2021's Q9 notes that some candidates confused the point O (labelled on the diagram) with the circle's centre rather than the origin, and states plainly that "a special case was permitted" for this specific misreading. This is the report speaking about its own scheme's design decision, not about candidate behaviour — a distinct register from every pattern in §1. No second series in this pass showed a report explaining a marking accommodation this explicitly; it may well recur, but this pass found it once.
• A speculative causal hypothesis, offered in parentheses, distinct from simply naming the error. October 2019's Q10 report does not just state that candidates failed to show discriminant working — it adds its own guess as to why, in parenthesis: "(presumably due to an overreliance on calculators)". That is a different rhetorical move from §1.6 (which is about the mark consequence): this is the report speculating about cause, in its own voice, rather than only describing the observed effect. Found once in this pass; §1.6's broader pattern is the safer generalisation, this specific speculative-parenthesis move is not.

---

#3. What this implies for how you write an answer

Briefly, because the technique itself already has a home in forge-exam-technique-guide.md and forge-level-threshold-guide.md — this section only states what follows directly from §1, not a new technique layer:

• Per §1.3 and §1.6: a plausible, even correct, final answer is not what is being credited — the specific working the mark scheme names is. Where a question bans calculator-only working or requires an exact form, treat that instruction as a direct description of exactly what a real report will check for, not boilerplate.
• Per §1.4: revising by topic is not the same as revising by discriminating sub-step. The reports repeatedly show that within a question you can already do, one specific operation (identifying a centre, finding an overlap term, reading a graph) is where marks are actually lost — practise that step in isolation, not just the question type around it.
• Per §1.5: when a question allows more than one valid method, the mark scheme's silence on which is "better" is not the same as the methods being equally safe under time pressure — the reports themselves repeatedly name which real candidates handle worse.
• Per §1.7: where a report repeats the same forward-looking advice across unrelated series and topics (know the standard proofs, sketch before starting circle questions), that repetition is Pearson's own evidence of a standing expectation, not a one-off house style.

---

#4. Series ledger for this document

(Note on the "June 2024 '01R' Q7" row: the 01R regional paper is dated ahead of June 2024's main sitting; its binomial-expansion Q7 report language is used in §1.5 alongside June 2025's own regional "01a" paper's Q1 — two different sittings, two different regional variants, not the same paper twice.)

---

#5. Flagged gaps

• This document draws on 13 of the 17 series with a published report. Four confirmed-present reports (June 2023, June 2024's main paper, October 2024) were not read at the depth needed to mine report language for this pass — a genuinely recurring habit that shows up only in one of those three would not appear anywhere above, and nothing here claims otherwise.
• §1.2's "hard numbers on the discriminating question" pattern leans heavily on January 2021, which alone supplies four of its examples; October 2021 and January 2022 each supply one. This is stated as it is in §1.2 itself — a real three-series pattern, not inflated by treating one heavily-quoted series as three.
• As with the WMA11 sibling document, the hedge-word scale in §1.1 is described qualitatively, not measured — no report in this pass defines its own words against an actual candidate count, so "many" vs "some" vs "a few" is read from context, not from a stated ranking.
• This document does not attempt to quantify how often each pattern recurs within a single report (what fraction of a report's paragraphs use the strategy-cliff framing of §1.3, for instance) — only that the pattern recurs across series. A future pass reading the remaining four unread-at-depth series in full could sharpen "recurs" into an actual count, and might also surface a second instance of either §2 single finding.