Statistics 1

Forty Eight Hour Checklist

Provenance: compiled 2026-09-29. This document does no new primary-source extraction — it is a synthesis layer sitting on top of work already verified elsewhere in this course, and cites only material that document already established: research/veridian/WST01-verified-facts.md (the provenance-checked facts bank, compiled 2026-09-12, spot-check addendum 2026-09-25), and, in this same folder, forge-exam-technique-guide.md, codex-formula-and-method-bank.md, and signal-frequency-and-boundaries.md. It also draws two real worked mark-scheme lines from content/veridian/question-bank/wst01.ts, both independently re-verified against the genuine June 2019 series (Publications Code WST01_01_1906_QP/MS/ER). Where this document repeats a fact from one of those, it cites the section rather than re-deriving it. It adds two things none of the other four provide: a time-blocked 48-hour schedule built around this specific paper's confirmed mark concentration and weak points, and a retrieval-practice self-test framed as short-answer prompts — distinct from FORGE §4's self-mark checklist, which is run against a script you've already written, not against memory alone. Not affiliated with or endorsed by Pearson Edexcel.

#WST01 — 48-Hour Pre-Exam Checklist

#Pearson Edexcel IAL Mathematics, Unit S1: Statistics 1

#0. What this is, and what it deliberately is not

This is a schedule and a self-test, not a fourth explanation of the paper. Three documents already sit in this folder and cover the material in depth: forge-exam-technique-guide.md (the M/A/B mark-coding system, the statistics-specific notation and "conclusion in context" conventions, a self-mark checklist, and an exam-day protocol), codex-formula-and-method-bank.md (every real formula, tagged by whether it's printed in the yellow booklet or has to come from memory, plus a worked-method reference), and signal-frequency-and-boundaries.md (which spec topics the verified past-paper record actually shows being tested, and the confirmed absence of any real grade-boundary data for this paper). This document does not re-teach any of that. It answers a narrower, time-bound question: given 48 hours left, in what order, and what do you deliberately leave alone.

Everything below traces to the same verified record those three documents already cite — six series with a fully-read examiner report (Jun 2019, Jan 2021, Oct 2021, Jun 2022, Jan 2023, Jun 2024) out of 14 series whose mark schemes were checked for structure (facts bank §3, §8). Nothing here invents a new trap, a new frequency figure, or a grade boundary. Where the underlying research is thin or absent, that is stated rather than filled in.

#1. The 48 hours, block by block

This assumes the exam is on the morning of T-0. Shift the clock, not the order, if yours is an afternoon sitting.

#T-48 to T-30 — the last full re-read (do this once, not twice)

Re-read FORGE §6 (the exam-day protocol) and §4 (the self-mark checklist) in full. This is the section with the highest return per minute spent, because it changes how you write, not what you know — and it is short.
Re-read SIGNAL §1's topic-frequency table once. Two real, defensible conclusions sit in it: Section 2 (representation and summary of data) is confirmed tested in every one of the 6 fully-read series — the safest assumption you can make about this paper — and the Normal distribution (6.1) is confirmed, wherever a full mark breakdown exists, as the single heaviest-weighted question on the paper (17 of 75 marks on the one series fully itemised, Jan 2023 Q5 — SIGNAL §1). Use the table to weight your re-reading, not to predict which topics will appear.
Pull up the curriculum lesson file for your weakest one or two topics only (content/veridian/curriculum/wst01/) and re-read its commonWrongPath / trap-taxonomy entry — not the full worked derivation again, just the named failure mode. This course's 13 WST01 lessons are deliberately organised so you don't need to re-read all of them to be ready in this window.
Do not start this block by opening a past paper cold. Re-reading first, testing second, is the order this whole document is built around.

#T-30 to T-14 — one full timed paper, marked against the real scheme's conventions

Sit one complete past paper under real timing: 90 minutes, 75 marks. Check the front page's own "There are N questions in this question paper" line before starting — the question count is not fixed on this paper (6 across most of the earlier series, 7 or 8 on several later ones, facts bank §2) — and read every question's own printed mark allocation as you go, rather than assuming a flat time-per-question split. Expect the Normal-distribution question specifically to run long (FORGE §1).
Mark it yourself against the real scheme if you have one, applying FORGE §4's self-mark checklist line by line before you look at the scheme's own award of marks.
For every mark lost, sort it into one of two piles: (a) "I didn't know the method" — a genuine content gap, and (b) "I knew it but the mark scheme wouldn't have credited how I wrote it" — a presentation gap. Pile (b) is the one 48 hours can still fix; re-read the relevant FORGE §2/§3 sub-section for each item in it. Large, unfamiliar-topic items in pile (a) go straight to §4 below ("what not to start now") rather than into a same-day cram.
If you don't have a second full paper for T-30 to T-14, don't manufacture timing pressure by re-sitting the same paper again for a "real" score — you already know the answers. Use the extra time for §2 below instead.

#T-14 to T-8 — targeted self-test, no new material

Work through §2 of this document (below) as a closed-book, no-notes retrieval test. Time yourself; each prompt should take seconds, not minutes, if the fact is actually secure.
Run the formula-recall drill in §3 from memory. This is real, checkable leverage: the spec's own "Notation and formulae" box lists specific S1 formulae that will not appear in the yellow booklet and must be reconstructed from memory under time pressure (CODEX §2a) — §3 below is exactly that list, and nothing more.
For anything you can't answer from memory in either drill, that's the item to look up once, now — not to relearn from scratch. A one-line lookup is fine in this window; reconstructing a technique you never solidly had is not (§4 below).

#T-8 to T-1 — wind down, light contact only

One short pass over your own error log from the timed paper (piles (a) and (b) above) — recognition, not new practice. If you're re-solving full questions in this window, you've moved back into T-14–T-8's territory; stop and switch to recognition-only review.
Confirm exam logistics now, not the morning of: candidate number, exam location/time, permitted equipment list.
Check your calculator against the real, verified restriction now, while there's still time to swap it if it fails: no symbolic algebra, no symbolic differentiation/integration, no retrievable stored formulae, no communication with other devices (spec Appendix 6, facts bank §2). This is not a non-calculator paper — an ordinary scientific calculator that has none of those forbidden features is fine.
Sleep. This document will not cite a source for that recommendation because none of this course's research is about sleep — it's stated as ordinary judgement, not as a WST01-specific finding.

#T-1 to T-0 — morning of, before you walk in

Re-read §3 (formula recall) one final time, out loud or on paper, from memory. Do not open the CODEX bank at this point — if a formula isn't secure by now, one more fresh look at the source will not fix that in the next hour, and the attempt eats time better spent calm.
Re-read FORGE §6's exam-day protocol once more (it is nine short lines).
Do not attempt a new past paper, a new topic, or a new practice question in this window. If you have spare time before the exam starts, use it on §2's retrieval prompts again, not on new content.

#2. Self-test: retrieval prompts built from real, recurring losses

Answer each of these from memory, in under 30 seconds each, before you check the source. This is deliberately not the same exercise as FORGE §4 (which is a checklist you run against a solution you've just written); this is testing whether the underlying fact is loaded, independent of any specific question in front of you. Every prompt below is built from a real, cited finding already established in this course's research — none is invented for this document.

You've written out a discrete probability function across four values of X and every probability looks sensibly small and positive. What single arithmetic check would catch a missing domain value, and which specific value is most often the one that goes missing? (Sum every stated probability — if it falls short of 1, a domain value is missing, most often $x=0$ . Confirmed, repeatable real error. Oct 2021 Q4(d) — facts bank §5, CODEX §1.4.)
You're asked for Var(3X − 2). What's the one-step error that has been independently confirmed three separate times across a three-year span on this exact calculation type? (Forgetting to square the constant — using $3 \times \text{Var}(X)$ instead of $3^2 \times \text{Var}(X)$ ; also, no $b$ term survives into the variance at all, whatever its sign. Jan 2021 Q4, Jun 2024 Q2(c), Q3(d) — facts bank §5.1, CODEX §1.4.)
A question tells you one data value was recorded wrong and gives you the corrected value, then asks for an updated box plot. Your first instinct is to redraw only the point that changed and check whether it still counts as an outlier under the old fences. What's wrong with that instinct? (A single corrected value can shift the quartile whose position it crosses — which changes the IQR, which changes both fences, including the one on the side that didn't move. The only safe procedure is to discard the IQR, both fences and the outlier list entirely and rebuild all of them fresh from the two new quartiles. On the one real question this is verified against, most marks were lost to reusing old numbers, not to bad arithmetic. Jun 2019 Q2(c) — question-bank entry wst01-jun19-q2c-box-plot-outliers, facts bank §5.1.)
A question gives you its own rule for identifying an outlier. Do you use that rule, or the $1.5 \times IQR$ convention you probably learned elsewhere? (The question's own rule, every time — this paper's spec guidance states outright that any outlier rule will be specified in the question, and a confirmed real error is substituting a remembered generic multiplier for the one actually given. Jun 2022 Q1(b) — CODEX §1.1; separately, fences point outward from the box — subtract for the lower fence, add for the upper, never reversed, Jan 2021 Q2.)
Two events look like they don't overlap much on a two-way table. Is that evidence they're independent? (No — independence has to be tested algebraically (does $P(A\cap B) = P(A)P(B)$ actually hold), never assumed from how a table or diagram looks. Pearson's own general comment on one series diagnoses exactly this as a paper-wide weak point. Jun 2022, general comment — facts bank §5.1.)
You're drawing a tree diagram for sampling without replacement, second draw. What two things change between the first branch and the second — not one? (Both the running total and the count just removed change on every draw — holding the denominator fixed, or re-using an earlier branch's probability on a later draw after removal, are two separately confirmed real errors on this exact question type. Jun 2022 Q3(d), Jan 2023 Q2 — facts bank §5.1, CODEX §1.2.)
You're computing $S_{xx}$ and you substitute $\bar x^2$ where the correction term should go. What should that term actually be, and is your substitution the same number? (The correction term is $\dfrac{(\Sigma x)^2}{n}$ , not $\bar x^2$ — the two are related but not identical, and this exact substitution error is confirmed on a real question, there labelled $S_{pp}$ rather than $S_{xx}$ (same formula, different letter). Oct 2021 Q2(b) — facts bank §5.1, CODEX §1.3.)
You've found $a$ and $b$ for a regression line and your final answer has a fraction in it. Does that cost you a mark? (Yes — the real mark scheme explicitly does not accept a fraction as appropriate for a final $a$ or $b$ ; decimal form only. Jun 2022 Q2(c) — facts bank §5.1, CODEX §1.3.)
A question asks whether a regression-based prediction outside the data range is reliable. "It's unreliable because it's extrapolation" — does that sentence score the mark? (Not reliably on its own — the real convention this paper rewards names the specific predicted value and its relationship to the specific range boundary (or, with no range given, to $\bar x$ ), not a bare assertion. Jun 2022 Q2(e), Jan 2023 Q6(c) — FORGE §3.2, facts bank §5.1.)
You standardise a Normal variable and reach a table value. Do you subtract it from 1, or use it directly — and is the answer to that question always the same for every question that reaches this step? (No — this is a real, confirmed both-directions error: some cohorts under-subtract (forgetting to subtract from 1 when the region required is the upper tail), others over-subtract (subtracting from 1 when the region required was already the correct one). The same underlying skill produces opposite mistakes on different series. "A simple diagram would have helped many to avoid this error" — Jan 2023 Q5(a); also Jan 2021 Q3(a), Jun 2024 Q5(a) — facts bank §5.6, CODEX §1.5.)
You look up a value from the Percentage Points table and round it to 2 decimal places before using it in the rest of your working. How many decimal places does the real mark scheme actually require? (Four — confirmed as the single most repeated Normal-distribution accuracy trap across three independent series, each time the same mechanism: a rounded value like 1.64 or 1.03 substituted for the table's exact 1.6449 or 1.0364. Jan 2021 Q3(b), Oct 2021 Q6(b), Jun 2024 Q5(b) — facts bank §5.6, FORGE §3.3.)
A Normal-distribution question ends by asking for a probability given that some other event has already happened. What's the single most consistently "most discriminating" mistake across every examiner report that discusses a question like this? (Not recognising that a conditional probability is required at all — e.g. finding $P(L \le 5)$ and stopping, instead of dividing by the conditioning event's own probability. Confirmed across all 5 fully-reviewed examiner reports with a different concrete manifestation each time: Jan 2021 Q3(c), Oct 2021 Q6(c), Jun 2022 Q6(c), Jan 2023 Q5(e). CODEX §1.5/§2.6 has the short method template.)
A multi-step inverse-Normal question (finding an unknown mean or SD from a given probability) gets harder partway through. What did Pearson's own examiner report note candidates actually doing at that point, on one real series — not as a maths error, but as a behaviour? (Giving up partway through, rather than pushing on to the final step. "It was not unusual for students to give up at this point" — Jun 2024 Q5(c), the hardest single Normal-distribution question type in this course's reviewed record by a clear margin. If this question type appears, a partial attempt with a correct diagram and a labelled method line is still worth real marks — finishing badly beats stopping.)

If more than two or three of these take longer than 30 seconds, that's real, specific information about where the remaining hours should go — re-read the cited lesson file's trap-taxonomy entry for that item, not the whole file.

#3. Formula recall drill — say each one from memory before you check it

The yellow "Mathematical Formulae and Statistical Tables" booklet gives you a genuine amount of S1 content: the addition rule, the general (non-independent) conditional-probability and Bayes-ratio forms, the base definitions of $E(X)$ , $\text{Var}(X)$ and $E(g(X))$ , both forms of $S_{xx}$ , $S_{yy}$ and $S_{xy}$ , the regression coefficients and line, the PMCC formula, and both Normal tables (CODEX §2a). The items below are specifically the ones the spec's own "Notation and formulae" box states will not appear in the booklet — reconstruct each one from memory before checking against codex-formula-and-method-bank.md Part 1:

$\bar x = \dfrac{\Sigma x}{n}$ or $\dfrac{\Sigma fx}{\Sigma f}$ ; the variance/SD formula in both its deviation and computational forms; $IQR = Q_3 - Q_1$ .
$P(A') = 1 - P(A)$ .
The independence-specific forms only: $P(B\mid A) = P(B)$ , $P(A\mid B) = P(A)$ , $P(A\cap B) = P(A)P(B)$ — the general form $P(A\cap B) = P(A)P(B\mid A)$ is given, only the independence special-case has to come from memory.
$E(aX+b) = aE(X)+b$ and $\text{Var}(aX+b) = a^2\text{Var}(X)$ — the shortcut forms; the base definitions above them are printed in the booklet.
$F(x_0) = P(X \le x_0) = \Sigma p(x)$ for $x \le x_0$ — note the $\le$ , not $<$ .
$Z = \dfrac{X-\mu}{\sigma}$ .

This drill is short on purpose. It is the one place in this whole 48-hour window where "know it cold" and "roughly remember it" produce genuinely different outcomes on the page, because there is no booklet entry to fall back on if memory fails on any of these six.

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

Hypothesis testing — $H_0$ / $H_1$ , a test statistic, a critical value, a conclusion against a significance level. This content does not exist on this paper. It is confirmed absent twice directly in the spec's own guidance (items 2.2 and 4.3, both stating explicitly that significance testing is not required for S1) and by a full-text search of the facts bank returning nothing under "hypothesis," "H0," "H1," "critical value" or "test statistic" beyond that guidance line (facts bank §5, FORGE §3.2). If a habit or a resource from a different Statistics unit pulls you toward this in the next 48 hours, stop — it is not examinable here.

The Normal p.d.f., its derivation, derivation of its mean/variance/CDF, or table interpolation. All three are explicitly stated as not required by the spec's own guidance (CODEX §1.5). Time spent deriving where the Normal tables come from is time not spent on the two skills that are actually examined and repeatedly missed: standardisation discipline and recognising when a probability is conditional.

Derivations of the regression coefficients $a$ , $b$ or $r$ , or a significance test of $r$ . Spec guidance states plainly that derivations are not required for 4.2, and that derivations and tests of significance will not be required for 4.3 (CODEX §1.3). Know how to use the formulae; do not spend this window trying to prove where they come from.

A closed-form variance shortcut for the discrete uniform distribution. This is a genuine, flagged gap in the reviewed record, not a shortcut worth memorising fresh: no equivalent-to- $\frac{n^2-1}{12}$ variance formula is confirmed anywhere in the facts bank for this paper, and every discrete-uniform variance calculation found in the record sums $x_i^2p(x_i)$ term by term (CODEX §1.4). If this sub-topic appears, use the general $E(X)$ / $\text{Var}(X)$ definitions, not a remembered closed form that has never been confirmed as something this paper's mark schemes expect.

A search for a standalone "mathematical modelling" past-paper question. No such question exists in the 14-series archive this course has reviewed (facts bank §8) — modelling content (spec 1.1) is examined embedded inside other questions (e.g. "assume the die is fair"), funded by an unusually high AO3 allocation (15–20 of 75 marks, facts bank §2b) rather than as its own question. Don't spend part of this window hunting for a modelling-only worked example that doesn't appear to exist; the relevant skill — naming an assumption, or judging a model's suitability in context — is exactly what FORGE §3.2's "conclusion in context" checks (regression gradient, reliability, distribution comparison) already cover, and is better practised through those.

A computed numeric skewness coefficient. Spec item 2.4 lists "skewness," but no computed skewness-coefficient formula is confirmed anywhere in the facts bank's reviewed record — it is treated on this paper as a comparative, descriptive judgement (mean vs median position, or quartile spacing), not a statistic calculated from a memorised formula (CODEX §1.1). Don't spend this window trying to recall a formula that isn't confirmed as belonging to this paper.

A grade boundary number. No real grade-boundary data exists anywhere in the source material used across this course's WST01 research — checked directly against the facts bank for "grade boundary," "boundaries" and "raw mark," with zero matches (SIGNAL §2). Don't let a target score drive how you spend the last 48 hours; there is nothing real to calibrate against here.

A new calculator, or new calculator settings. If you haven't used a given model or settings through your actual past-paper practice, the 48 hours before the exam is not the time to switch. Not a WST01-specific finding — the same reasoning as the "don't start a new topic" point above, applied to equipment.

Re-deriving your own revision notes from the curriculum files a second time. If you already made a summary earlier in the course, trust it in this window rather than rebuilding it "to be thorough" — that is T-48–T-30's job, not something to repeat in every later block too.

#5. Flagged gaps and honest limits

This document's factual claims are only as good as the series they trace back to. Six series have a fully-read examiner report (Jun 2019, Jan 2021, Oct 2021, Jun 2022, Jan 2023, Jun 2024); the other 8 of the 14 series whose mark schemes were checked (Jun 2021, Jan 2022, Oct 2022, Jun 2023, Oct 2023, Jan 2024, Oct 2024, Jan 2025) were searched for an examiner report but none was conclusively located and confirmed (facts bank §8). Every "confirmed" or "real" claim above traces to one of the six read series — see facts bank §3, §5 and §8 for the exact scope.
The one targeted spot-check this course ran on a previously-unread series (Jun 2019) found two new, previously-undocumented gaps from that single read — a strike rate the facts bank itself flags as suggesting the true gap count from the other 8 unread series is likely higher, not lower (facts bank §8). This document's self-test in §2 should be read the same way: a lower bound on real, confirmed traps, not a finished inventory.
"Confirmed in N of 6" is not a prediction. Following SIGNAL's own convention, it means a specific question/mark-scheme citation exists in that many of the six fully-read series — not that a topic is N/6 likely to appear again on the next sitting. This document repeats that warning deliberately, because a 48-hour checklist is exactly the kind of document a reader might otherwise misread as forecasting the paper.
This document does not contain, and will not manufacture, a grade boundary, a pass mark, or a probability that any specific topic appears. See §4 above for why.
The self-test in §2 draws only on findings already named in the facts bank, FORGE or CODEX — it introduces no new trap discovered specifically for this document. The addition this document makes is the retrieval-practice framing and the 48-hour time structure, not new underlying content; the facts themselves are inherited, cited, and unchanged from their original sources.
The T-48-to-T-0 block structure and its specific hour ranges are this document's own judgement, not a cited finding — no WST01 examiner report or spec document makes a claim about optimal revision timing, and none should be read into the section headers above.