How do you upload handwritten student work for marking?

Why does handwritten work still need to be digitised?

A good deal of secondary assessment is still done on paper. Final written examinations are sat by hand, in-class essays are often handwritten to keep the conditions honest, and junior classes draft by hand as a matter of course. Handwriting itself is part of what English asks students to develop. Any marking routine built only around typed files leaves a large share of real student work sitting outside it.

Digitising a script is not about replacing the paper. It is about getting a text version you can work with: something you can search, quote back to a student, attach comments to, and still have after the exercise book goes missing. Treat the scan as the record and the transcript as the working copy. The transcript is only ever as good as the scan it came from, which is why the first ten seconds of the process decide most of the quality.

How do you scan handwritten work so it can be read accurately?

Scan in even light with the page flat and the camera square to the paper. Contrast and sharpness matter far more than file size. Handwriting recognition works from the shapes of letters, so a crisp, high-contrast page reads far more accurately than a dim or skewed one. Dark ink on plain white paper is the easy case. Pencil on lined paper, faded photocopies and glossy pages under a ceiling light are the hard ones.

Use the document scanner built into your phone rather than a plain camera photo. It squares up the edges, flattens the page and exports several pages as one PDF, which is exactly the shape you want. One physical page should map to one page in the file. If you do photograph a page instead, keep the whole response inside the frame with a small margin around it.

These habits cost almost nothing once they become routine, and they pay for themselves across a whole class set.

• Fill the frame with the page and leave out the desk, your hand and the sheet underneath.
• Use daylight or bright, even room light, and move to kill glare and hard shadows.
• Keep the page flat and hold the camera directly above it, not at an angle.
• Save as a PDF at normal or high quality rather than a heavily compressed image.
• Check every page is the right way up before you upload.
• Keep one student's pages in one file, in the order they wrote them.

What does handwriting recognition usually get wrong?

Recognition is strong on ordinary connected prose and weaker almost everywhere else. The predictable trouble spots are proper nouns, subject-specific vocabulary, numbers, units and dates. Crossed-out words, carets and arrows pointing to inserted text are read literally or dropped altogether. Tables, columns, diagram labels and mathematical notation often come back out of order, or not at all.

The awkward part is that a transcription error does not look like an error. The text comes back clean and confident, with a plausible word standing in for the one the student wrote. A misread figure in a data response, or a scientist’s name spelled three different ways, quietly changes what you are marking. Whether the task is a short-answer set or expert essay marking, the judgement has to rest on what the student actually put on the page.

How should you check a transcript before you mark?

Read the transcript beside the original scan before you mark anything. On a two-page response this takes under a minute. Check the page count first, then the first and last line of each page, which is where truncation shows up. Then look for the categories that break: names, numbers, dates, technical terms, and any sentence where the meaning suddenly goes strange.

Correct recognition errors only. This distinction matters more than it sounds. If a student wrote “definately”, that is their spelling and it is assessable, so it stays. If the software turned “hegemony” into “harmony”, that is the machine’s error and you fix it. Tidying a student’s prose while you clean up the transcript changes the evidence before you mark it, and nobody will spot it later.

With thirty scripts, put the effort at the front. Compare the first three line by line. If those come back clean, the scans are good enough that a spot check on the rest is reasonable. If they come back messy, the problem is the scanning rather than the students, and rescanning the set is faster than correcting thirty transcripts.

What is a workable routine for a whole class set?

Keep one student per file. That single rule prevents most of the mess. A class set is a series of individual uploads, not one merged document, because a merged file cannot be split back into the right students without manual work. Collect the scripts in roll order, scan each student’s pages into a clearly named file, and upload them in the same sitting.

Use a naming convention you will still understand next term: surname, first name, task code. Hold on to the paper until marks are finalised and moderation is done, because a scan on its own is a thin record if a result is queried. Student work is personal information, so keep it in the systems your school has approved rather than a personal cloud account.

The transcripts are reusable once they exist. A strong paragraph becomes an exemplar for next year, and the errors that repeat across the set tell you exactly what belongs in the study guides you hand back to the class.

Why do some scans fail to upload or process?

Most failures come down to size, orientation or network. A ten-page scan at 300 DPI carries far more image than a two-page one, so it takes longer to upload, longer to transcribe, and it is the file most likely to time out. Pages saved upside down or rotated sideways usually come back as nonsense rather than an outright error, which is worse, because nonsense looks like a result.

School networks are the other common culprit. Content filters on school wi-fi sometimes block file uploads before they ever reach the application, which shows up as a progress bar that never moves and no error message at all. Sending one file over a phone hotspot is the quickest way to find out whether the network is the problem.

• Rescan any page that is blurry, dark, curled or cut off at the edge.
• Split a very large scan into two or three smaller files.
• Rotate every page the right way up before uploading, not after.
• Try a different network, such as a phone hotspot, if an upload stalls with no error.
• Scan at about 200 to 300 DPI, which is plenty for handwriting and much faster to move.

How does an AI marking platform handle a scanned script?

Everything above applies whatever tool you use. This last part is specific to Jeddle, the platform behind this site, so you can see how those steps map onto one implementation.

You upload a student’s PDF or photo to their own submission. JeddAI transcribes the handwriting into editable text before any marking begins, and the original scan stays stored so you can read it beside the transcript and fix misreads yourself. Blank or genuinely unreadable pages are flagged rather than filled in with invented content, which is the behaviour to look for in any tool doing this job.

Once the text is confirmed, the work is marked the way typed work is: against your own rubric, success criteria and comment banks. Feedback and a suggested mark are drafted against those criteria, and you review, adjust and approve every piece before a student sees it. The saving is in not retyping and not starting each comment from an empty box. The judgement stays yours, which is the only arrangement that survives a moderation meeting.