Statistics 1

Level Threshold Guide

#WST01 (Statistics 1) — Level/Threshold Guide

Provenance: written 2026-09-29, built entirely on research/veridian/WST01-verified-facts.md (compiled 2026-09-12, spot-check addendum 2026-09-25) — specifically its §4 (mark-scheme conventions), §5 (per-topic verified content) and the "Per-lesson bullet-by-bullet coverage audits" section, which independently re-fetched and re-verified several primary mark schemes directly against rendered PDF page images where pdftotext garbles stacked fractions. Cross-checked against forge-exam-technique-guide.md, forge-battle-card.md, codex-formula-and-method-bank.md and signal-frequency-and-boundaries.md in this same folder to avoid restating what they already cover; where a fact below also appears in one of those documents, this document goes to the next level of granularity that one didn't (a full mark-by-mark ladder, both sides of a boundary, or the exact wording that draws the line) rather than repeating the same sentence. No new PDF fetches or web searches were performed for this document — every claim below traces to a citation already independently verified elsewhere in this course's WST01 research, cited again here by series and question number. Where the real record simply doesn't contain a threshold for something (most of the paper's B1-for-a-diagram-feature and routine M1/A1 arithmetic marks have no examiner-report or audit commentary attached), this document does not manufacture one.

#What "threshold" means on this paper, and what this document refuses to do

WST01 is marked M/A/B, not by level descriptors (1–5 bands, best-fit judgements) the way an essay-marked subject is. There is no qualification-wide grade boundary data anywhere in this course's source record either — signal-frequency-and-boundaries.md §2 checked the facts bank directly for "grade boundary," "boundaries" and "raw mark" and found zero matches, and this document performed no new fetch to manufacture one. So the real, checkable "threshold" structure on this paper is not a percentage-of-marks cutoff for a grade; it is the line the mark scheme itself draws, mark by mark, between a response that earns a specific M, A or B and one that doesn't — and, chained across a whole question, the line between a partial-credit stopping point and full marks. That is what this document maps, exhaustively, using only boundaries a real mark scheme or examiner report actually states. Two hard constraints follow directly from that: no likelihood percentage is invented anywhere below (only "seen in N of M series checked"), and no grade boundary is invented either. Where a boundary below is confirmed on one series only, that is stated, not smoothed into a general rule.

Every threshold below is presented as a pair: a real (or, where marked, a mechanically identical near-miss built from the same verified mechanism) response that just misses the mark, and one that just clears it — followed by the exact mark-scheme wording the boundary itself rests on.

#1. The atomic thresholds: M, A, B, and the cliffs between them

Source for the mark-type definitions themselves: the "General Instructions for Marking" boilerplate opening all 14 mark schemes reviewed, byte-for-byte identical (bar the total-marks figure) across at least 6 spot-checked directly — already summarised in forge-exam-technique-guide.md §2.1. This section goes one level deeper: what actually sits on each side of the M0/M1, A0/A1 and B0/B1 line in real, quoted responses.

#1.1 M0 → M1 — the method-attempt threshold

Rule, verbatim: "A method mark is given for a correct method or an attempt at a correct method."

Real ladder, Oct 2021 Q6(a) (Normal distribution, "show that P(H > 30.8) = 0.15," 3 marks total — mark scheme verified directly against the primary PDF in the facts bank's per-lesson audit of show-that-answer-discipline.ts):

1st M1: "for standardising 30.8 with 25.1 and 5.5 (allow +). Allow use of z = 1.04."
Just clears: the working shows $z = \dfrac{30.8-25.1}{5.5}$ (or the equivalent with a sign slip tolerated), evaluating to 1.04. M1 earned — this M mark rewards the structure of standardising, not a specific decimal precision.
Just misses: the response jumps straight to a final numeric answer with no standardisation expression visible anywhere. This is not hypothetical — the facts bank quotes the real failure mode verbatim: "Some students simply wrote down 0.1500… with no working at all." No standardisation shown means M0 on this line, and because the paper's own general rule states "M0 A1 is impossible," every mark chained to this one becomes unreachable regardless of how correct the final 0.15 is.

#1.2 The dependent-M (DM) threshold — a second method mark contingent on the first

Same anchor, 2nd line: "2nd M1 for 1 – p (where 0.84 < p < 0.86)." This is a genuinely useful, under-appreciated fact: it is a second M mark, not the accuracy mark that intuition suggests should follow a "1 − Φ(z)" step, and its band is deliberately wide (any $p$ between 0.84 and 0.86 qualifies) rather than pinned to one exact table value. The facts bank's own audit of this lesson found the lesson had mislabelled this line as A1 (requiring 4-d.p. table precision) in three places before correction — a genuine, previously-live error in this course's own material, which is exactly the kind of boundary-drawn-in-the-wrong-place mistake this document exists to prevent repeating.

Just clears: any subtraction of a Φ(z) read-off from 1 landing in $[0.84, 0.86]$ — e.g. using $\Phi(1.04) \approx 0.8508$ to get $1 - 0.8508 = 0.1492$ .
Just misses: omitting the "1 −" step entirely and reporting the raw table value (0.8508) as the answer to " $P(H>30.8)$ " — a real, separately-confirmed error pattern in this exact category (§3 below, "subtract-from-1 direction confusion," independently seen in 3 series).

#1.3 A0 → A1, and the reach-back cso threshold

Final line of the same anchor: "A1cso for an answer of 0.1492… or better (calc: 0.1500…) with evidence of both M's seen." Two separate thresholds are compressed into this one line:

The cso reach: this mark's dependency isn't just on the immediately preceding line — the scheme explicitly requires "evidence of both M's," reaching back across the whole 3-mark chain. A response with a correct final 0.15 but only one of the two method steps visible still scores 0/3, not 2/3, because cso means "there must be no errors or gaps in this part of the question to obtain this mark."
The precision-as-proof threshold: the target value is given in the question (0.15). Writing exactly "0.15" or "0.1500" with no working proves nothing — it could be copied from the question stem. The scheme's own bracketed note, "calc: 0.1500…," makes explicit that the unrounded calculator value is 0.1500 to four places, meaning a candidate who shows 0.1492 or better — i.e., carries at least one more significant figure than the printed target — has demonstrably derived rather than copied it. Just clears: 0.1492 (or any value that rounds to the target but visibly required a computation to reach). Just misses: a bare "0.15" with nothing behind it — the exact real failure this document's §1.1 already quotes.

#1.4 B0 → B1 — the independent/standalone-comment threshold

B marks reward a fact, comment, or diagram feature standing alone, with no method mark to lean on — which makes them unforgiving in a different way: there is no partial credit for "the right idea, wrongly expressed," because there's no M mark upstream to catch a near-miss.

Real boundary, Jun 2022 Q2(e)(ii) (regression reliability judgement, GDP-vs-population data, no min/max range given in the question — 1 mark): credited answer, verbatim, "Unreliable as 7 000 000 is much greater than the mean population/t-bar for the 9 years," with the marking note stating explicitly: "'Extrapolation so unreliable' on its own is B0. Reference to g out of range is also B0."

Just clears: "Unreliable, because $t=7$ is far above $\bar t \approx 0.874$ " — names the specific value, names the specific reference point ( $\bar t$ ), connects them.
Just misses (two independently confirmed ways): (i) "Unreliable — this is extrapolation," with no value or reference point named at all; (ii) naming $g$ (the response variable, the thing being predicted) as what's out of range, when the scheme requires the reference to be to $t$ or $\bar t$ (the explanatory variable the model was actually built on) — a subtler miss than a bare assertion, since it at least attempts specificity, but is explicitly zero-credited by name in the marking note.

#1.5 ft (follow-through) — earning a mark on a value that may itself be wrong

Real boundary, Jan 2023 Q2(d) (tree-diagram conditional probability, "given all 3 same colour, find P(all green)," carried from part (c)'s own answer 61/234): "A1ft for finding the correct numerator... If the answer is incorrect then working must be shown."

Just clears: numerator = the single clean path probability $\frac59 \times \frac48 \times \frac{8}{13} = \frac{20}{117}$ , over the candidate's own (possibly carried-forward, possibly wrong) part-(c) denominator.
Just misses: the real, examiner-report-quoted wrong answers both used the correct denominator (61/234) but folded an extra, spurious copy of that same fraction into the numerator — either added or multiplied in: $\left(\frac59{\times}\frac48{\times}\frac{8}{13} + \frac{61}{234}\right)\big/\frac{61}{234}$ and $\left(\frac59{\times}\frac48{\times}\frac{8}{13} \times \frac{61}{234}\right)\big/\frac{61}{234}$ . The ft mechanism means this mark is judged purely on whether the numerator is the correct path-probability applied to whatever denominator the candidate is carrying — which is exactly why a self-inflicted denominator error elsewhere in the same question would not, on its own, cost this specific mark, while the extra-copy-in-the-numerator error does.

#2. Precision thresholds — where the same rounded value sits on different sides of a mark

This is the single richest thresholds category in the whole facts bank: multiple independent series, the identical underlying mechanism (a rounded intermediate value substituted for an exact one), and — uniquely among the categories in this document — one series gives the exact mark-code boundary in the mark scheme's own words.

#2.1 The 23.6 / 23.60 cliff — a verbatim mark-scheme threshold

Jan 2021 Q3(b) (same Normal-distribution question as §1's anchor, a different part — "find $j$ such that $P(W>j)=0.15$ ," 3 marks: M1 correct table selection, then a middle mark, then a final A1). The mark scheme's own precision note, verbatim: "awrt 23.60 scores 3/3 [but] 23.6 scores M1B0A1 unless 1.0364 or better is seen."

Just clears (full marks): either write "23.60" (two decimal places, proving the rounding wasn't accidental), or write the bald "23.6" but show the intermediate $z = 1.0364$ (or a more precise equivalent) somewhere in the working — either route scores 3/3.
Just misses (2 of 3): write "23.6" with no intermediate $z$ -value shown anywhere. The numeric answer is identical in both cases — correctly rounded to 1 d.p. — but the version with no visible 4-d.p. working loses the middle mark, because a 1-d.p. final answer alone doesn't prove the candidate used the table's precise value rather than a rounded approximation of it.
The lesson this draws out: "23.6" and "23.60" look interchangeable in ordinary written English and are not remotely interchangeable on this mark scheme — the second string is doing real evidentiary work the first doesn't.

#2.2 "Ans only" — the opposite threshold, on an ordinary "find" part

Jan 2021 Q3(a) (same question, the first part — "find $P(W<10)$ ," 3 marks): the mark scheme credits a bare correct final answer with "Ans only 3/3" for any value in the range 0.0692–0.0694 — full marks, zero working required, provided the final value already carries enough precision to prove it wasn't rounded from something less exact.

Just clears: writing "0.0693" with nothing else — full 3/3, because the four-figure precision is itself the proof.
Just misses: this allowance is exactly what §2.1 shows fails one part-letter later in the same question, for exactly the reason given there — a bare answer only survives unaccompanied when its own digits already carry the demanded precision. A bare "0.07" here (2 s.f., not the demanded 3 s.f. minimum) would sit in the same trap as "23.6" does in part (b): correct in value, short on provable precision.
The two together (2.1 + 2.2) are the real, complete version of "when can I skip working on this paper?" — the honest answer is not "never" and not "always," but "only when the digits you wrote down already prove the precision the mark scheme is checking for," which is a genuinely different, narrower rule than either extreme.

#2.3 The 4-decimal-place table-value cliff, confirmed independently three times

Unlike §2.1 (one series, one exact mark-scheme quote), this threshold is confirmed by the same mechanism recurring across three separate series, each with its own quoted rounded-vs-exact pair — strong corroboration that this is a stable paper-wide rule, not a one-series quirk:

Series/Q	Just misses (rounded)	Just clears (table-exact)
Jan 2021 Q3(b)	1.03 or 1.04	1.0364
Oct 2021 Q6(b)	1.64	1.6449
Jun 2024 Q5(b)	0.25	the precise 4-d.p. equivalent

The Jan 2021 examiner report states the mechanism directly: "some students are opting to use calculators for this sort of question and invariably they fail to show sufficient accuracy" — the threshold isn't really about the table specifically, it's that a calculator's own rounded display, or a memorised "nice" value, sits on the wrong side of a line the table's own printed precision defines.

#2.4 The sign threshold — same magnitude, wrong side of zero

Jan 2021 Q3(b) (again) and Oct 2021 Q6(b), independently: a candidate reaches the exact right 4-d.p. magnitude and then applies it with the wrong sign — Jan 2021: "others gained this mark but were using −1.0364"; Oct 2021: "using the wrong sign for 1.6449... giving 34.1, a value higher than the upper limit." This is a real, distinct threshold from precision itself: +1.0364 clears the mark that −1.0364 misses, even though both numbers demonstrate the same correct table read-off. Pearson's own stated fix in both reports is identical: draw a diagram first, so the sign is decided by a picture rather than a rule remembered under pressure.

#3. Form and simplification thresholds

#3.1 The unsimplified-fraction cliff

Jan 2021 Q1(c) (Venn diagram, three events, conditional probability $P(A\mid B')$ , mark scheme independently re-verified against the primary PDF for this course's audit of elementary-probability-and-conditional-notation.ts): the M1 line accepts $p/0.4$ for any $0<p<0.4$ or the specific working value $0.15/0.4$ — but the A1 line requires "=3/8", and the scheme is explicit that an unreduced $0.15/0.4$ standing alone does not earn it.

Just clears: $\dfrac{3}{8}$ , or its decimal equivalent $0.375$ .
Just misses: $\dfrac{0.15}{0.4}$ left exactly as computed — correct in value, wrong in form.
Where oe actually starts, not where it stops: contrast this with the real, confirmed scope of oe elsewhere in the archive — Oct 2022 MS credits "= 40/117 oe... Allow awrt 0.556 or awrt 55.6%" as the same mark. oe covers interchangeable forms of an already-simplified value (fraction ↔ decimal ↔ percentage); it does not retroactively rescue an unreduced ratio that never got simplified in the first place. The two facts together draw the actual boundary precisely: simplify first, then any equivalent form of the simplified value is safe.

#3.2 The fraction-in-a-final-regression-line threshold

Jun 2022 Q2(c) examiner report, verbatim: "it is also important to note that fractions are not accepted in a final regression equation; the values of a and b are both estimates and so a fractional answer is not appropriate." This threshold is unusual among the ones in this document: it is a rule about form, applied irrespective of whether the fraction is even fully simplified — a correct, simplified fraction for $a$ or $b$ still misses this specific line, where the identical value written as a decimal clears it. No exact numeric pair for this specific series is quoted in the facts bank beyond the rule itself, so none is invented here — the rule is stated exactly as strongly as the primary source states it, no more precisely.

#3.3 A non-threshold, stated for contrast: coding and $r$

Jan 2024 MS Q2(d), verbatim: "r not affected by (linear) coding." Worth including here specifically because it is the mirror image of every boundary above: recomputing $r$ from scratch on the original variables after a coding step, versus reading off the already-computed $r$ from the coded variables directly, land on the same side of every mark — there is no threshold here to cross, only wasted time in choosing the harder route. Stated explicitly so it isn't mistaken for one.

#4. Structural thresholds — where a multi-mark ladder actually breaks

The M/A/B system chains marks together, so the more interesting real thresholds are often not "did you get this one mark" but "how far up a several-mark ladder did you get before something broke." Two full real ladders, independently re-verified in this course's research, are reproduced here completely because nowhere else in the technique-hub gives either one mark-by-mark in full.

#4.1 The complete 7-mark ladder — June 2019 Q2(c), an amended box plot

Real question, verbatim task: "Albert's score was changed from 25 to 37. Beth's score was changed from 54 to 60. (c) ... draw an updated box plot. Show clearly any calculations that you used." Original five-number summary (verified from part (a)'s own mark scheme): min 11, $Q_1=33$ , median 42, $Q_3=47$ , max 54, IQR 14. The full real mark scheme, in order:

Rung	Mark	What earns it
1	B1	$25\to37$ ⇒ new $Q_1=35$
2	M1	Recognising the new IQR (12) must be recalculated, and that fences must follow
3	A1	New outlier fences: $>47+18=65$ or $<35-18=17$
4	M1	A box with only two whiskers, median still at 42
5	A1ft	Lower quartile at "their 35" ( $\ne 33$ ), upper quartile unchanged at 47
6	A1	Two outliers correctly identified at 11 and 15
7	A1	Lower whisker at 18 (or 17), upper whisker at 60

The examiner report confirms exactly where real candidates fell off this ladder, and it isn't rung 1: "The major problem was that few candidates realised that the lower quartile would change and increase to 35. Most of those who spotted this realised that this would affect the interquartile range... most others usually just obtained a mark for drawing a box plot with an unchanged median."

Just clears rung 1 only (1/7): redraws the original box plot with just the two amended data points moved to their new positions on an unchanged frame — median unchanged, both quartiles unchanged, no new fences computed. This is explicitly the modal real wrong answer, not a constructed worst case. Score: 1/7 (rung 4's B1-for-median is separate and likely still earned since the median itself didn't move — the point is that rungs 2–3 and 5–7, five of the seven marks, are lost together as one block the moment the IQR isn't recomputed).
Just clears the whole ladder (7/7): recomputes $Q_1=35$ (rung 1) because one of the two amended values fell in the lower half of the data; recognises this changes the IQR to 12, forcing both fences to be recalculated from scratch — including the upper fence, on the side that didn't change (rung 2); finds the new fences 17 and 65 (rung 3); redraws with two whiskers, median fixed at 42 (rung 4); places $Q_1=35$ , $Q_3=47$ unchanged (rung 5); identifies both new outliers, 11 and 15 — values that were always in the data but only became outliers because the fence moved past them, not because they themselves changed (rung 6); draws whiskers to 18 (or the fence, 17) and 60 (rung 7).
The real threshold this exposes, stated as its own rule: a single amended data value can move one quartile while leaving the other unchanged, and that alone is enough to force both fences to be recomputed — including the fence on the side nothing changed on numerically, because the fence's width term (the IQR) changed even where its anchor quartile didn't. Missing this one structural fact costs 5 of the ladder's 7 marks in one go, confirmed as the dominant real failure mode, not a rare one.

#4.2 The split 4-mark ladder — June 2024 Q2(c), two independent equations in one part

Real question: $Y=aX-b$ ( $a,b>0$ ), given $E(Y)=13.4$ and $\mathrm{Var}(Y)=66.24$ , alongside $E(X)=4.6$ and $\mathrm{Var}(X)=4.14$ (from an earlier part), "find the value of $a$ and the value of $b$ " — 4 marks. Mark scheme, verbatim:

Rung	Mark	What earns it
1	M1	"a² × 4.14 [=66.24]" — the squared-constant variance equation, correctly set up
2	A1	"a = 4"
3	M1	"E(Y) =13.4 = a × 4.6 − b" — the separate mean equation, correctly set up
4	A1	"b = 5"

This ladder is structurally different from §4.1's: it isn't one chain where an early slip cascades, it's two independent two-mark chains bolted into a single question part, and the facts bank's own audit of this lesson found real teaching material (before correction) that closed only the first chain and told the reader the second one "plays no part in it at all" — a genuine, previously-live example of stopping exactly halfway up a ladder that looks, from the first chain alone, like it's already finished.

Just clears rungs 1–2 only (2/4): correctly solves $a^2 \times 4.14 = 66.24 \Rightarrow a=4$ , and stops — a fully complete, fully correct piece of mathematics that answers only half the question asked ("find the value of $a$ and the value of $b$ "). The examiner report's own quoted common error for rung 1 — "a × 4.14 = 66.24 rather than a² × 4.14 = 66.24" — is a different, earlier failure (forgetting to square $a$ at all, which is M0, not a stopped-halfway case); the stopped-halfway case is a distinct, separately real way to lose exactly the same two marks.
Just clears the whole ladder (4/4): having found $a=4$ , separately sets up the mean equation $13.4 = 4\times4.6-b$ and solves $b=5$ — a mark-scheme-confirmed, genuinely independent second step that does not follow automatically from finding $a$ , and is not "the same skill applied twice."

#4.3 "Show that N outliers exist" — the list-them-out completeness threshold

Oct 2021 Q3(c): "having gained the correct limits some did not list the 3 outliers in this part in order to show there are 3 outliers."

Just clears: state the correct fences, then explicitly write out all three qualifying data values.
Just misses: state the correct fences (the harder half of the work, and the half that earns the method credit) and stop — the mark scheme still wants the count demonstrated, not merely implied by a correct boundary.

#4.4 The probability-function domain-sum self-check threshold

Oct 2021 Q4(d): "the most common error was to miss out the value x = 0... had students checked whether the sum of their probabilities equalled 1 they may have realised that they had missed zero out."

Just clears: every value in the stated domain (including $x=0$ ) is assigned a probability, and the set sums to exactly 1.
Just misses: an otherwise fully-worked probability function that omits one domain value — the arithmetic on every other value can be entirely correct and the response still fails, because the function as a whole no longer sums to 1. This is the one threshold in this document candidates can check for themselves before submitting: sum what you've written before moving on.

#4.5 The fixed-vs-shrinking-denominator threshold in sampling without replacement

Jun 2022 Q3(d) (four counters drawn without replacement, "P(all four the same result)" scored on a scheme explicitly acknowledging a named special case): "the most common response was scoring 1 mark for the special case using (62/88)⁴" — keeping the denominator fixed at 88 across all four draws — against the full-marks route that decreases the denominator by one on each successive draw.

Just clears: $\frac{62}{88}\times\frac{61}{87}\times\frac{60}{86}\times\frac{59}{85}$ (each factor's denominator one smaller than the last).
Just misses, but not to zero: $\left(\frac{62}{88}\right)^4$ — mechanically simpler, and explicitly named in the mark scheme as a recognised "special case" worth exactly 1 mark, not 0. This is one of the few thresholds in this document where the mark scheme itself names the near-miss and gives it partial credit by design, rather than the near-miss being an unplanned examiner observation.

#4.6 The stop-at-recognition threshold in conditional probability under a Normal model

Jan 2023 Q5(e): "a common error was to find P(L≤5) and go no further" — without dividing by the conditioning event's own probability.

Just clears: recognises a conditional probability is required, computes the numerator $P(L\le5)$ , and divides by the conditioning probability to complete the ratio.
Just misses: computes $P(L\le5)$ correctly (a real, non-trivial piece of Normal-distribution work) and reports that number as the final answer — earning the recognition/method credit for the numerator alone while missing the mark(s) attached to completing the division.

#4.7 The E(X)-vs-P(win) threshold — a boundary drawn by which quantity you cite, not by arithmetic

Jun 2019 Q5(e) (two-stage spinner game; $E(X)=\tfrac12$ , $P(\text{Amar wins})=\tfrac{7}{16}$ , both independently verified elsewhere in the question; the question's own fictional character states "he will play the game" because $E(X)>0$ ): mark scheme, verbatim, "[Although E(X) > 0 since] P(win) < 0.5 ... Amar should not play the game or 'disagree'" (M1 A1, 2 marks).

Just clears: disagrees with Amar, citing $P(\text{win})<0.5$ as the deciding quantity.
Just misses: agrees with Amar (or disagrees for the wrong reason), citing $E(X)>0$ as if a positive expected value implied a better-than-even chance on a single play. The examiner report confirms this was the dominant failure, not a rare one: "There was little success in part (d) [sic — see the facts bank's own note on this citation] with many attempts focussing on the fact that E(X) > 0 rather than the critical issue that P(Amar wins) was less than 0.5." Both quantities can be — and in this question, were — computed with equal difficulty and equal correctness earlier in the response; the mark lives entirely in which one gets named as the reason, because a positive long-run average and a sub-50% single-play probability are different calculations that can genuinely disagree about the same game.

#5. Thresholds that run backward — where writing more can cost a mark you'd already earned

Every threshold above rewards showing more work or more precision. This paper also contains the opposite: real, mark-scheme-confirmed cases where the default assumption that extra working is harmless is explicitly switched off.

#5.1 The isw override

Default rule, verbatim: "ignore subsequent working" — wrong working after a correct answer normally doesn't cost the mark. Jan 2024 MS Q2(c) switches this off for one specific line: "(do not isw if any further calculation is done after multiplying by 9/5)."

Just clears: reaches the correct value at the "multiply by 9/5" step and stops there.
Just misses: reaches the identical correct value at that step, then keeps calculating and arrives somewhere else — on this specific mark, that costs the mark isw would ordinarily have protected, precisely because the scheme has named this line as an exception to the default.

#5.2 The box-plot whisker asymmetry — the same mark scheme, two different rules for two whiskers

Jan 2021 Q2(d), verified directly against a rendered image of the mark-scheme page (the facts bank flags pdftotext as unreliable here): the 2nd A1 line reads, verbatim, "for upper whisker ending at 97 and lower whisker ending at 54 or 50 and only one outlier, shown at 43. Allow ± 0.5 of a square for accuracy."

Upper whisker threshold: credited only at 97 — the true highest non-outlier data value. Drawing it to 98 (the calculated fence) is the real, examiner-quoted error and is not credited.
Lower whisker threshold — genuinely different: credited at either 54 (the true lowest non-outlier data value) or 50 (the calculated fence itself). The same mark scheme, the same question, the same diagram, draws the "fence is not good enough, use the real value" line for one whisker and does not enforce it for the other.
The rule to internalise, stated exactly as strongly as the evidence supports and no further: drawing to the real data value is the safe default that clears the mark on both whiskers in this question — drawing to the fence clears the mark on one side here and not the other, which makes the fence an unreliable shortcut rather than a rule broken cleanly in half. Treat "draw to the real data value" as the rule to follow, not "fences are fine" — but the asymmetry itself is real, confirmed, and worth knowing rather than smoothing into a false universal.

#6. What this document deliberately does not contain

No qualification-level grade boundaries. Confirmed absent from every source consulted for this document and for signal-frequency-and-boundaries.md before it — not a gap in this document's research, a gap in what Pearson has published freely and what this course has verified.
No invented likelihood percentages. Every frequency claim above is stated as "seen in N of M series checked" (and several thresholds above are stated as confirmed on one series only, explicitly not generalised).
No thresholds for the parts of the paper this course's examiner-report coverage doesn't reach. Examiner-report coverage across the whole facts bank remains 6 of 14 mark-scheme series (Jan 2021, Oct 2021, Jun 2022, Jan 2023, Jun 2024, and the 2026-09-25 spot-check of Jun 2019) — the facts bank itself states this as a lower bound, not exhaustive coverage, and every threshold in this document inherits that same caveat. The routine B1-for-a-correctly-drawn-diagram-feature and M1/A1 arithmetic marks that make up most of this paper's 75 marks have no examiner-report commentary attached anywhere in this course's research and so have no threshold entry here — this document covers where the record is rich, not where the paper's mark allocation happens to be heaviest.
No re-derivation of thresholds already fully covered elsewhere in the technique hub without new granularity. Where forge-exam-technique-guide.md, forge-battle-card.md or codex-formula-and-method-bank.md already state a rule (the misread penalty, the general marking principles, the crossed-out-work convention), this document either adds the specific ladder or boundary-pair those documents don't work through in full, or omits the point rather than restating it a fourth time.

Word count and sourcing note for this document specifically: every numeric value, mark code and quotation above is reproduced from research/veridian/WST01-verified-facts.md's own already-verified citations (each traceable there to a named series, question number and, in several cases, a rendered-PDF-page re-check the facts bank records explicitly) or from the corresponding lesson-audit entries in that same file's "Per-lesson bullet-by-bullet coverage audits" section. No new primary source was fetched to write this document; none of its content was generated to hit a target length.