Blog
An AI exam marker marks a paper exam against the mark scheme and answer key you wrote. You upload the exam students sat and the key you trust. You photograph the scripts. Graider marks each script against that scheme. You moderate the stack: review the marks, override any score you would explain differently, sample scripts, and check that the same work received the same call. You toggle feedback, keep the pen, and hand the exam back as a PDF or an email.
Graider is the AI red pen for paper grading, a product of DM Liquidity LLC. The public site is graider.org. The place to run an exam set is AI exam grading. This page stays with the scheme you set up, the method marks and partial credit that follow it, and the moderation you do yourself. Graider does not moderate or second-mark on its own.
Your mark scheme. Their scripts. You moderate. You keep the pen.
Set the mark scheme before you photograph the stack. An AI exam marker follows the rows you wrote. Each question needs the points on offer, the response you accept, and the split when those points come in parts.
The answer key holds the responses you accept. Put the final value on the row, and any equivalent form you already count as the same answer. Put the short phrase, the term, or the date. When two answers are both full credit, list both. A common wrong answer can sit on the key too, with the points it earns and the comment you already use for that mistake.
The mark scheme says how those points divide. A calculation can give marks for the setup, marks for the working, and a mark for the final value with the unit the question asked for. A source question can give a mark for the claim and marks for evidence from the passage. A short explanation can give marks for the reason and a mark for the term that names it. Write the split so you can point at the page for every point during moderation.
One exam can mix those rows. A definition earns full marks for the term you accept, and fewer points for a correct idea in the student's own words. A later sentence may use the student's own earlier number when your scheme allows that follow-through. A labeled sketch is yours to score. Graider does not read or score diagrams, graphs, or drawings. A written scale or a written label stays on the key with the other written answers.
Name the border of each row: which method counts, which unit must appear, and whether a later part may follow an earlier number. A border you leave unwritten gets decided again on every script.
Bring this set to the upload:
Method marks and partial credit are rows on your scheme. The first pass looks for those rows in the writing on the page. A setup that matches the row earns the setup marks. Working that follows the method you named earns the method marks. The final value, with the unit you required, earns the mark you assigned to that line. A correct final value with the working missing earns only the points your scheme gives a bare answer.
Follow-through is a row you wrote, or a gap you left on purpose. When a student carries an earlier wrong number into the next part and uses it correctly, the later method marks follow your rule. When you teach two methods, list both and write the marks each one can earn. When you accept only the method from the notes, write that limit so the first pass has a border.
Put each near miss on the scheme with the points it earns. A missing unit, a sign that slipped on the last line, a technical term one letter off, a reason that covers part of the point you wanted. One row per mistake you already see from this class. Moderation is where you confirm that call or override it.
Paper math tests, including substitution, elimination, and error carried forward, are set out in AI math grading. On any other exam, a method mark should still point at a row you wrote when you moderate.
Upload the exam and the answer key. Photograph the stack. Graider marks the scripts against your own rubric and key. That first pass is what you moderate. It applies the scheme you uploaded to the handwriting on each page.
Read one full script against the scheme before you sample the rest. The mark on each question should name its row: the accepted answer, the points, and the condition for any partial credit. When the line matches a row and you agree, keep those points. When the line matches no row, set the mark yourself. If the same line will show up again, add it to your copy of the scheme and use that row as you continue.
The Low confidence badge is the flag on a hard-to-read answer. Open the photograph, compare the line with your key, and override the score when the page and the first pass disagree. That badge is the only flag.
Moderation is your step. You review the marks. You override the scores you would explain differently. You sample scripts from the stack. You check that the same scheme produced the same call on the same kind of work. Graider does not moderate or second-mark on its own.
Pick a question where the points can split more than one way. Read several scripts that made the same move: the same missing unit, the same alternate method, the same follow-through from an earlier number, the same partial reason on a short explanation. The points should match the row. The comment should match the note you would write in the margin.
When a script in the sample is off, override it. Change the points. Edit the comment so it is the sentence you would write in the margin, in your own words. Carry that decision to the other scripts in the sample that fit it. A script that differs in a way that matters stays on its own page. A second method you accept, a crossed-out line, or a drawing you are scoring yourself belongs on that page alone.
Then sample a second question of another kind, such as a short explanation after a calculation. You are checking that the scheme you wrote is the scheme on the scripts. When the sample and the scheme disagree, amend the row and override the marks already on the stack. Consistency means the same call on the same work, under a scheme you can defend.
Toggle feedback as you go. Turn comments on when they should go home with the score. Leave them off when the class will talk through that question and the score is enough on the handback. The words a student reads are the words you kept. You keep the pen through the whole sample.
When the sample holds and the overrides are in, hand each exam back as a PDF or an email. Students receive the marks you kept and the comments you turned on. You keep the pen until that file goes out.
The ordered read of each script before release is the work of an AI exam checker. Moderation is the sample you take so the scheme stays consistent across the stack.
An AI exam marker marks photographed paper exams against the mark scheme and answer key you provide. Graider is the AI red pen for paper grading. You upload the exam and the key, photograph the stack, and Graider marks against that scheme. You then moderate: review, override, sample scripts, and check that the same work received the same call. You toggle feedback, keep the pen, and hand the exam back as a PDF or an email. Graider does not moderate or second-mark on its own.
Write the points for each question, every answer you accept, and the split when a question has parts. Add method marks and partial credit as a condition, the points, and a short example. Include equivalent answers, a common wrong answer with the points it earns, units, and any follow-through you allow. Upload that scheme with the exam, then photograph the scripts.
Yes. Method marks and partial credit follow the rows you wrote on your mark scheme and answer key. A matching setup, a method you named, a final value with the unit you required, and a near miss you already gave points all keep the points on that row. You can override any mark when you moderate.
You do. Moderation is your review of the marks, your overrides, your sample of scripts, and your check that the same work received the same call across the stack. Graider does not moderate or second-mark on its own. You keep the pen.
The Low confidence badge is the flag on a hard-to-read answer. Open the photograph, compare the line with your key, and override the score when you disagree. Graider does not read or score diagrams, graphs, or drawings. Set those scores yourself. Written labels and values around a drawing stay on the mark scheme.
Set the mark scheme, photograph the scripts, and moderate the stack before the marks go home. Method marks and partial credit follow the rows you wrote. You review, override, and keep the pen.