How we verify
Every quote, checked against the real document
Nothing in this course is written from memory of what a mark scheme probably says. Every quote, figure and mark allocation is matched against the actual PDF text of the source document — the real spec, the real mark scheme, the real examiner report — before it's used. Where something couldn't be verified in the time available, it's flagged as unconfirmed rather than presented as fact.
The method
- 1. Re-extract the source, every time. The real PDF is converted with
pdftotext -layout— not OCR guesswork, and never a paraphrase carried over from an earlier pass. - 2. Verbatim only. A quote used in a lesson is the source document's own words, not a summary of them.
- 3. Flag what isn't confirmed. If a claim can't be checked against a real primary source in the time available, it's marked unverified rather than quietly treated as true.
- 4. Log every error found and fixed. Verification sometimes catches something wrong in earlier material — a misquote, a wrong date, a mixed-up mark boundary. Corrected, and counted below rather than quietly dropped.
This paper
48 files / 16 series (Oct 2019-Jan 2025), 12 series reviewed.
38
quotes verified
3
errors caught & fixed
Known gaps, stated honestly: Mark-scheme coverage shallow (Jan 2025 only); 4 examiner reports unread.
Across every paper
281 quotes independently verified and 11 real errors caught and fixed, across the 8 papers with a tallied record. 6 more papers (WMA11, WMA12, WMA13, WMA14, WME01, WST01) follow the identical documented method without a single tallied count yet — listed honestly, not rounded into the total above.
| Paper | Quotes verified | Errors caught |
|---|---|---|
| wec11 | 38 | 3 |
| wec12 | 28 | 1 |
| wec13 | 14 | 3 |
| wec14 | 30 | 1 |
| wbs11 | 85 | 0 |
| wbs12 | 45 | 3 |
| wbs13 | 35 | 0 |
| wbs14 | 6 | 0 |
The honest limit
This verification is self-audited — checked by the same pipeline that writes the content, against real primary sources, but without an independent human examiner's sign-off. That's a real, stated limit, not a hidden one: it's why every number above is a count of what was checked, not a claim of zero remaining errors.