Mathematics: The handwriting-to-feedback loop

What you'll be able to do: photograph the working you did on paper and have the error found in your own handwriting: then redraw the corrected step yourself.

Why this matters more in mathematics than anywhere else

Mathematics happens on paper. Not in a text box, on paper, with crossings-out, arrows, a substitution written in the margin, and two abandoned attempts above the one that worked.

Every other route into a tool requires transcription, and transcription is actively harmful here for a reason that isn't obvious: you re-derive as you type. The error you most needed found gets silently fixed on the way in. You paste a clean version of working you never actually did, get told it's correct, and learn nothing about the mistake that's still in your hand.

Photographing removes that. And in mathematics it removes it for the majority of the work, because the majority of the work was never going to be typed.

The caution belongs here, not at the end. Reading handwritten mathematics is one of the less reliable things these systems do. A misread exponent, a 7 taken for a 1, a subscript lost, and you get a fluent, confident, specific correction of a mistake you never made. You'll believe it, because it's specific.

So the loop always starts with a transcription check. Ten seconds, every time.

The five steps

1. Work the problem properly on paper. Don't tidy it. The mess is the data.
2. Photograph it flat, in good light, whole page.
3. Ask for a line-by-line transcription first, and check it. If a digit is misread, correct it and re-run before accepting anything.
4. Ask for the first error and the rule you appear to hold.
5. Redraw the corrected step by hand, on the same page, below the original.

Step 5 is the one people skip and the one that does the work. Being shown a correction produces recognition. Redrawing it produces the motor and visual trace of having done it right, on the same page as having done it wrong, and that comparison is what changes what you do next time.

This photo is my handwritten attempt. Step 1: transcribe exactly what you see, line by line, and flag anything you're unsure you read correctly. Step 2: after I confirm, tell me the FIRST line that's wrong and what rule I appear to be applying. Don't give me the full solution.

Where it pays across mathematics

Long algebraic manipulation. Six lines of rearrangement with one sign lost in line three. Nothing after line three is wrong given line three, which is exactly why marking the final answer wrong teaches nothing.

Integration by substitution. The error is usually in the du line, and it's usually written in the margin. Transcription matters here, margin notes are where misreading happens.

Simultaneous equations. Errors are arithmetic and positional. A photograph catches the line where a term was carried down wrong, which a typed version would have silently repaired.

Geometry proofs. The reasoning is annotated on a diagram. There is no verbal form of "you've labelled the wrong angle as the included one": it only exists as a picture.

Matrix work. Positional errors are the whole failure mode. Typing a matrix re-orders your attention and fixes them.

Induction proofs. The structure is visual, base case, assumption, step. A photograph shows that you wrote the assumption and never used it, which is the most common invisible failure in induction.

Statistics by hand. Standard error and degrees of freedom errors live in subscripts, which is where misreading is most likely. Check the transcription especially carefully.

Graph sketching. The sketch is the answer. Photograph it and ask what's wrong with the shape, the asymptotes and the intercepts, three separate questions, because they fail independently.

The version that finds patterns

Here are photographs of five problems I got wrong this week. Transcribe each, let me confirm, then: don't correct them individually. What single misunderstanding explains the most of them?

Handwritten work makes this better than typed, because the pattern of where on the page things go wrong is visible. Errors that cluster at the end of long manipulations are a stamina and care problem; errors that cluster at setup are a comprehension problem. They need different remedies and a typed transcript hides which you have.

Pitfalls specific to this subject

1. Skipping the transcription check. The defining risk. A misread exponent produces a confident correction of an error you didn't make.
2. Photographing a tidy copy. You'll have fixed the error in transit.
3. Not redrawing. Reading the correction is the thinnest possible version of this.
4. Bad photographs. Shadow across the page and an angled shot cause misreads, and misreads cause wrong corrections.
5. Asking for the full solution. Converts the loop into answer-getting.
6. The tell: a page covered in corrections and you can't state the rule you'd been getting wrong.

Try this today

Do one problem on paper, properly, including the messy bits. Photograph it.

Ask for the transcription first. Check it. Then ask for the first error. Then redraw that one step by hand.

Five minutes, and you'll finish with a named misrule instead of a corrected answer.