How a doubt gets solved from a photograph, step by step
The important step is the one that looks like a delay: before anything is solved, the page is read back to you line by line so you can correct the reading, because a confident answer to a misread question is the worst thing this could produce.
Photographing a page of homework and getting help with it is an old idea. The difference between a useful version and a frustrating one is almost entirely in what happens in the first few seconds, so here is the whole sequence, in order.
1. The photo is prepared on the device
The picture is turned the right way up, scaled down to a sensible size and re-encoded before it leaves the phone. A page photographed at an angle in poor light is the normal case, not the exception, and doing this work on the device rather than on a server means the thing that travels is small and the thing that arrives is upright.
A photo that is too large is refused on the device with a plain message rather than sent and rejected later.
2. It is screened before it is read
Before any reading happens, the image is checked. A face, another child's name, a home address or a live examination paper comes back as a refusal, and nothing is kept, not even in memory on the device.
That is a deliberate cost. It means an innocent photo occasionally gets refused, and we would rather explain a refusal than store a photograph of somebody's child.
3. The reading is shown back before any answer
Vision misreads. A 3 becomes an 8. A minus sign vanishes. An exponent lands on the wrong term.
So the page comes back as a list of lines, each one editable in place, with the question stated as it was understood: this is what I read, is that right, fix anything I got wrong first. Nothing is computed until that is settled, and the code will not let it be: the state machine simply does not accept an instruction to explain until a reading exists and has been shown.
This is the step people expect to be able to skip, and it is the step that decides whether the rest is worth anything. Correcting one character here is quicker than discovering, four lines into a confident solution, that the whole thing was answering a different question.
4. The explanation happens on your page
Once the reading is confirmed, the working is explained on the photograph itself. The ink lands on the line it is talking about, at the moment it is being talked about, and it circles a line rather than pointing at a pixel, because a line is a thing you can find again and a pixel is not.
The rule underneath this is small and it is enforced in the code: every mark made on the photo has to fall inside the window of the sentence that explains it. Stroke by sentence. A drawing that arrives before or after its explanation is decoration, and it is exactly the failure that makes annotated answers feel like something being performed at you.
If the idea needs more than a mark on the page, it opens somewhere with more room, which is the only time that happens.
5. It joins the rest of the learning
A solved photo that leaves no trace is a missed lesson. So the doubt is placed on the topic it belongs to, the topic moves to started on the learner's own map, and it is scheduled to come back later to see whether it stuck.
What it does not do is record it as mastered. A doubt explained is not a topic learnt, and there is no answer from the learner to be right or wrong about, so nothing is claimed. The map moves forward and never backwards, and never all the way.
Under the surface it also counts as a miss on that concept, so if the same idea comes up again the next explanation changes its method rather than repeating the one that has already failed once.
Where this still fails
Handwriting that is genuinely hard to read is genuinely hard to read, and no amount of asking politely changes that. A page with several questions on it will be read as several questions and the useful move is to photograph one. A diagram-heavy page in dim light is the hardest case there is.
The honest behaviour in all of those is the same: say the page could not be read, say why, and ask for another photograph. A guess at an unreadable page is the one outcome that wastes a child's evening, because it looks like an answer.
Why the boring step is the whole thing
If you take one idea from this: the reading is shown before the answer, and it is a rule rather than a courtesy. Every other part of this can be rebuilt. That one cannot be skipped without turning a tutor into a machine that is confidently wrong on your child's homework, which is worse than nothing at all.
How this was written. Drafted with AI help, from source documents and notes a person gathered, then read and edited by a person before it went up. Every figure in it is one we can show you the source of.
Works on how an explanation is built, what happens when one misses, and what the tutor should refuse to do.