If you'd be happy to photocopy from it, it will convert. That's the rule, and most of what follows is that rule with the reasons attached.
The four things a scan needs
| It needs to be | Because |
|---|---|
| Straight | Skewed text is read as skewed text. A degree or two is fine; a page photographed from the side isn't |
| 300 dpi or better | Below that, small type and subscripts blur into each other. Most copiers default to 300 |
| The whole page | A cropped margin is a cropped question. Number 12 is usually at the bottom |
| Evenly lit | A shadow across half the page reads as a dark band, and whatever's under it is lost |
An office copier set to scan-to-PDF produces all four without being asked. That's why most centres already have what they need. Put the worksheet on the glass, press the button, and the file that comes out will convert cleanly.
A phone photo usually fails at least two. It's taken at a slight angle, the lighting is whatever the room had, and the edges are cut off or in shadow. A scanning app on the phone fixes most of that by flattening and cropping the page, and a scan taken that way is usually fine. A photo from the camera app, held over a desk, usually isn't.
What converts well and what doesn't
Typed documents, clean PDFs, decent scans and spreadsheets of questions all convert well. That covers most of what a centre has: the Word file the worksheet was written in, the PDF that got printed, the scanned copy of the paper original.
Handwriting doesn't, in general. A handwritten worksheet, or a typed one with handwritten annotations you want kept, is the case where a re-type is faster than a re-scan. Phone photos at an angle don't, for the reasons above.
Neither is a surprise on the day. The upload and convert step tells you which documents will fight you before conversion starts, not after a student has seen a garbled question.
The fidelity score
Every document you upload gets a score before anything is made from it. It measures how faithfully the page was read: how much of the text came through cleanly, whether the formulae are intact, whether the diagrams were captured with the questions they belong to. The fidelity score entry in the glossary has the definition.
A clean typed worksheet scores in the high nineties. A tired photocopy of a photocopy scores lower, and when it does, the pages that dragged it down are flagged for a re-scan. Not the whole document, just those pages. Nothing under the line is silently accepted and pushed through, because a question that came out wrong and got published under your logo is worse than a question that was never converted.
In practice the score is the first thing an owner looks at in a new batch, and the second thing they stop looking at, because after a few documents from the same copier it's the same number every time.
Maths is the special case
Maths material has a problem that English material doesn't: a fraction is a picture, not a word.
Ordinary text recognition reads a page as a stream of characters. Run it over three-quarters and you get a 3, a line and a 4, in some order, and the question is now wrong. Run it over an exponent, a square root, a long division layout or a matrix and the result is worse.
So maths goes through a specialist step where formulae are read as formulae. A fraction comes out as a fraction. An exponent stays raised. The question a student sees on their phone is the question you wrote, and the answer key it's marked against is the one on your sheet. If your centre is mostly maths, this step is most of the reason the conversion works at all, and it's why a skewed phone photo hurts maths more than it hurts a comprehension passage.
Diagrams and graphs are captured with the question they belong to, and shown exactly as they are on your worksheet. They don't need to be redrawn and they don't need to be described.
The practical run
Before a first batch, walk through the folder with one question: would I photocopy from this? For most pages the answer is yes and there's nothing to do. For a few it's no, and those are the ones to re-scan from the original, or to re-type if the original is lost.
If the whole folder is paper and the scanner is the problem rather than the pages, the digitising your teaching material guide covers how to get through a drawer in an afternoon, and there's a prep service if you'd rather post the folder and have it scanned, sorted and curated before conversion.
The scan quality entry in the glossary is the short version of this page, if you want something to send a tutor who's been asked to scan the Year 9 folder.
This week
Pick the folder you'd convert first and scan three pages from it on the office copier. Open the PDF on a phone. If you can read every subscript and the page is straight, that's the standard, and the rest of the folder scanned the same way will convert. If you can't, the copier settings are the fix, not the pages.