Technique Masterclasses

Statistics 1

A Star Gap Analysis

Provenance: written 2026-09-29. This is a synthesis/index document, not new primary-source research. Read directly this session: research/veridian/WST01-verified-facts.md (§2b — the AO-weighting table's own listing of the 14 optional units in this qualification, used only to ground the Part 0 caveat below; §3 — the archive scope and the 6-of-14 examiner-report denominator every WST01 technique-hub document already inherits); content/veridian/curriculum/wst01/ (read as a directory listing only, to confirm no existing lesson file already covers this synthesis — no lesson content is reused here beyond what the technique-hub documents below already cite); and, in full, the following existing WST01 technique-hub documents, which supply every specific claim below: forge-level-threshold-guide.md (the source of every threshold in Part 1), forge-error-pattern-tracker.md (the source of the seven-pattern framing referenced in Part 2 and Part 5's tracking mechanism), forge-exam-technique-guide.md (the M/A/B system and §3.2's "conclusion in context" trigger points, reused in Part 2), forge-battle-card.md §4 (its "Top-5 danger warnings," the closest thing this hub has to a dedicated danger-phrase index, reused in Part 2), forge-vocabulary-precision-list.md (the source of every entry in Part 3), signal-examiner-intelligence.md (the source of every pattern in Part 4), and signal-frequency-and-boundaries.md §2 (the source of the grade-boundary gap stated plainly in Part 0). No independent primary-source PDF was opened this session; every mark-scheme line, examiner-report line, and citation below is reused, with its original attribution preserved, from one of these seven documents. No new diagnostic category is introduced anywhere in this document — every category below already exists, cited, in one of the seven sources.

WST01 — A* Gap Analysis: Self-Diagnostic

What this document is, and isn't

This is not new research. Every WST01 technique-hub document that measures why marks are lost — the level-threshold guide's mark-by-mark boundaries, the error-pattern tracker's seven confirmed mechanisms, the exam-technique guide's "conclusion in context" rules, the vocabulary-precision list, the examiner-intelligence report-language patterns — already exists. What doesn't exist yet is a single place that answers the question a candidate already scoring a solid, comfortable pass actually asks: of everything documented across those five files, which parts are specifically about the gap between a good pass and (near-)full marks, as opposed to the gap between failing and passing at all? This document answers that by re-reading each of the five with that filter and indexing what survives it. It invents nothing new: every citation below already appears, in the same words, in one of the source documents named in the provenance note above.

Part 0 — What this document can and can't say about "A*" specifically

signal-frequency-and-boundaries.md §2 has already checked this paper's own source record directly for grade-boundary data — searching WST01-verified-facts.md for "grade boundary," "boundaries," and "raw mark" — and found zero matches. No Pearson grade-boundary document for WST01 has ever been fetched or verified anywhere in this course's research, so this document states that gap plainly rather than filling it with a plausible-looking claim. WST01 is one of 14 optional units of the same qualification (the facts bank's own §2b AO-weighting table lists it alongside P1–P4, FP1–FP3, M1–M3, S1–S3 and D1, all sharing one assessment-objective structure) — and a comparable unit in that same qualification, WMA12, has had this specific question checked against seven official Pearson grade-boundary documents, finding that A* exists only for the combined six-unit qualification (calculated from UMS totalled across all six units), not for any single 75-mark unit paper. Given WST01 sits in the identical unit structure, the same conclusion is very likely true here too — but that check has not actually been performed for WST01 in this course's research, and this document does not assert it as a confirmed WST01 finding. What this document diagnoses, honestly scoped, is the real and checkable thing underneath the filename: the gap between comfortably clearing a strong pass on WST01's own 75 M/A/B marks and dropping so few of them that this unit is not the one costing the overall qualification's top grade — not a percentage, boundary, or unit-level "A*" this document has verified exists.

Part 1 — The real shape of the gap: seven mark-scheme thresholds, as self-diagnostic questions

forge-level-threshold-guide.md documents places where the real mark scheme draws a line inside a mark or a ladder of marks — not simply right vs wrong, but two responses that can look identical and land on opposite sides. These are exactly where a script that is "mathematically right" can still lose marks, because the maths alone doesn't decide the outcome.

1.1 — Precision-as-proof: does your final answer's own digit count prove you derived it, or could it have been copied from the question? Jan 2021 Q3(a)'s "Ans only 3/3" allowance for 0.0692–0.0694 and Q3(b)'s explicit rule that a bald "23.6" scores only M1B0A1 while "23.60" (or 23.6 with the intermediate z=1.0364z=1.0364 shown) scores 3/3 are the same paper, the same question, drawing this line two different ways depending on whether the digits alone already carry the demanded precision.

1.2 — Show-that reach-back: does your working show both method steps, not just a plausible final line? The Oct 2021 Q6(a) anchor's A1 is explicitly cso, reaching back across the whole chain to require "evidence of both M's seen" — a correct final 0.15 with only one method step visible scores 0/3, not 2/3, and "M0 A1 is impossible" (verbatim, all 14 schemes) means a missing standardisation line makes every mark chained to it unreachable regardless of the final number's correctness.

1.3 — B-mark specificity: does your stand-alone comment name a value and a reference point, or just assert a conclusion? Jun 2022 Q2(e)(ii)'s marking note states outright that "'Extrapolation so unreliable' on its own is B0" and that naming the wrong variable (the response variable instead of the explanatory one) is also B0 — a B mark has no method mark upstream to catch a near-miss, so a plausible-sounding but unspecific comment earns nothing.

1.4 — ft mechanics: on a carried-forward value, is your numerator built from the real path probability, or padded with an extra copy of the denominator? Jan 2023 Q2(d)'s A1ft judges purely whether the numerator is the correct path-probability applied to whatever denominator is being carried — the real, examiner-quoted wrong answers both used the correct denominator but folded a spurious extra copy of it into the numerator.

1.5 — Structural completeness: when one part of a dataset changes, have you re-derived every downstream quantity — including the side that didn't itself change? The full 7-mark June 2019 Q2(c) box-plot ladder shows a single amended value moving one quartile forces both outlier fences to be recalculated, including the fence on the side nothing changed numerically, because the IQR itself changed. The examiner report confirms this costs 5 of 7 marks in one block, as the dominant real failure, not a rare one.

1.6 — Two-unknowns traps: does "find aa and bb" actually require two independently-solved equations? The June 2024 Q2(c) 4-mark ladder is two separate two-mark chains bolted into one part — finding aa from the variance equation does not follow automatically into finding bb from the mean equation, and this course's own material once stopped exactly halfway up this ladder before correction.

1.7 — Backward thresholds: could writing more, not less, cost a mark the default rule would otherwise protect? Jan 2024 MS Q2(c) explicitly switches off the normally-forgiving isw ("ignore subsequent working") default for one specific line; separately, the same mark scheme that credits a box-plot's lower whisker at either the true data value or the calculated fence credits the upper whisker only at the true value — the same scheme, the same question, two different rules for two whiskers.

Part 2 — The invisible-answer trap: the candidate who can see the answer is the one it catches

This is the mechanism forge-battle-card.md §4 ranks its #5 danger warning, and forge-error-pattern-tracker.md §1 ranks its #1 pattern by cross-series confirmation: "show that" working left too thin. The real, quoted failure mode from Part 1.2's own anchor is blunt: "Some students simply wrote down 0.1500… with no working at all." A weaker candidate who genuinely needs the standardisation step to reach the answer produces it by necessity. A candidate strong enough to see that P(H>30.8)≈0.15P(H>30.8)\approx0.15 almost by inspection is the one who can skip writing the intermediate z=1.04z=1.04 and the "1−1-" step down — and is therefore precisely the one this cso rule catches, because "M0 A1 is impossible" makes the omission fatal regardless of how right the final number is.

The gap-closing behaviour this points to is not "get better at the statistics" — by construction, the candidate already has that. It is tracking which command word governs which rule on that exact part: Part 1.1 already shows a bare correct answer can score full marks on an ordinary "find" (Jan 2021 Q3(a)) while the identical skill under "show that" wording requires the full method shown (Oct 2021 Q6(a)) — the same underlying calculation, two structurally different schemes, decided entirely by one word in the question.

Part 3 — Vocabulary precision as a last-mile differentiator

forge-vocabulary-precision-list.md documents distinctions the mark scheme is exact about that a mathematically-sound script routinely blurs anyway — which is exactly why blurring them costs marks a purely computational check wouldn't catch. Three entries reused directly rather than re-derived: command words map onto different mark types, not interchangeable phrasing — "write down"/"state" buys a no-method B mark, "calculate"/"find" buys M+A with a bare-answer route often available, and "show that" looks like "find" but requires both method steps shown, no bare-answer route at all (§1 of that document). The word "independent" carries three unrelated technical meanings on this one paper — independent events (spec 3.3), the independent/explanatory variable in regression (spec 4.2), and "independent" as the mark-scheme's own name for a B mark that needs no prior method — and nothing in the spec or any mark scheme warns a candidate the same word is being reused across all three. Var(aX+b)'s squaring requirement recurs under two different names in two different lessons — spec 5.3's linear-transformation formula and spec 2.2's coding-decode step both punish forgetting to square the scaling constant, confirmed independently across a three-year span, but a candidate who has mastered one lesson's version has not automatically recognised the other's as the same trap.

Part 4 — What the examiner reports themselves say about where the gap sits

signal-examiner-intelligence.md finds that four of the six fully-read reports shift into a distinct "students should be encouraged/reminded/advised" register whose recommended fix is consistently procedural — draw a diagram, use the table's exact value, show the standardisation step — never "know more content" (§1, §5). The same document finds a second, narrower signal worth acting on directly: when a report reaches for "only the [most/more] able candidates," it is flagging a genuine grade-discriminating step, confirmed in two instances attached to the single hardest sub-part of a multi-part question — a context-scaled gradient interpretation (Jun 2022 Q2(d)) and a multi-stage inverse-Normal problem candidates gave up on partway through (Jun 2024 Q5(c)) (§3). A third finding matters for how a strong candidate should read this whole hub: the same specific mechanisms recur, named again, years apart in fresh cohorts — rounded-z-value, subtract-from-1 direction, and the Var(aX+b) squaring slip are each independently confirmed in multiple series spaced years apart (§4) — meaning competence at the underlying maths does not, on this record, make a candidate immune to these specific named traps; only the deliberate habit-check does.

Part 5 — Using this as an actual diagnostic, not just a reading list

This document adds no new tracking mechanism of its own — forge-error-pattern-tracker.md already has the fillable tables (its §3 structural-pattern log, its §5 topic-specific trap log, its §6 running tally across papers). What this document adds is a filter to apply before logging a lost mark there: for any mark lost on a question you were otherwise getting right, check it against Part 1's seven thresholds first, rather than defaulting to "silly mistake." A mark lost to the cso reach-back rule or the isw override (Parts 1.2, 1.7) is not the same category of problem as not knowing a method at all, and the fix is a habit — write the intermediate step down, check which command word is governing this exact part — not more practice of statistics the candidate, by definition, already has.

Flagged gaps — what this document does not claim

Sources used