Blog
Partial credit grading with AI follows the splits you write before a single paper is marked. A method mark, an incomplete answer, and an accepted variant each need a line on your answer key or rubric: what is on the page, and the points that case earns. Graider marks the photographed papers against that writing. You review the score, override when you disagree, and keep the pen.
Write the split once on the key or rubric. A wording you accept in full sits at the full total. You check each mark at review against the same key and rubric.
Graider is the AI red pen for paper grading, a product of DM Liquidity LLC. You upload the test and your answer key or rubric, photograph the stack, and Graider marks each question against that document. You review and override, toggle feedback, keep the pen, and hand the work back as a PDF or an email. Paper tests follow that path in AI test grading.
Your key. Their papers. You review every mark. You keep the pen.
Graider returns one score and one comment per question. A whole essay gets one score and one comment. Write each partial as that finished score. A line that says "formula and substitution, wrong final value: 3 of 4" is the score for a paper that matches it. The method, the missing piece, and the variant are how you chose 3.
Keep each line at or under the question total. Use the fullest line the paper meets, and use that line's points as the score. When two lines could both fit, name the one you want applied in the wording of the line.
Number the question the same way on the test and on the key. If part (b) is its own question, give part (b) its own total and its own lines. If you score the parts together, keep one number and list the full-credit combination and each partial combination under that number.
A method mark is credit for a step a colleague can find on the page. Name the step and the finished score for a paper that shows it, including a paper whose final value is wrong.
On a 4-point item, "Find the area and show the formula," the key can read:
A paper matches one of those lines, and that line's total is the score. When you accept two methods, give each a finished score: "Uses the source and states a claim: 3 of 4" and "Uses the source and leaves the claim unfinished: 2 of 4."
Algebra keys often carry several method marks, units, and follow-through on one item. How to write those rows is in AI math grading for paper tests.
An incomplete answer has some of what you asked for and leaves a required piece out. The line names what is present, what is missing, and the finished score. "Names the treaty and the year: 2. Names the treaty and leaves out the year: 1."
If a missing unit costs one point and keeps the rest, the line shows the lower total. If a missing unit costs the whole question in your class, write that response as 0. Graider returns one comment for the question. Edit that comment at review so it sounds like you.
When a blank earns nothing, the key can say "blank: 0". When you later accept a response the key left out, override that paper at review, and list that wording on your key for the next test.
An accepted variant is a wording or a form you count. Put it on its own line with the points it earns. Variants you accept in full sit at the full total. Variants that are close, and still short of the full response, sit at the lower total. The points on the line are the split.
On "Explain one way a plant gets food, and name the plant part that does that work," worth 4 points, the key can read:
The 4-point lines are full-credit variants. The 2-point lines are incomplete answers. The 1-point line is a start. A paper that matches a 4-point line earns 4, because that is the fullest line it meets.
Spelling and units use the same kind of line. "Color: 2" and "Colour: 2" when either spelling earns the mark. "Receive: 2" and "Recieve: 1" when a missing letter costs a point. "12 cm: 2" and "12: 1" when the bare number is a partial.
A variant you have not written is a score you set at review. Override that paper at review, and list that wording on your key for the next test. Whether those lines live on a fixed answer key or on a rubric is covered in answer key grading vs rubric. On a key, each response is a line with points. On a rubric, each level names what is present, what is missing, and the one score that level earns.
Photograph every page a student used, including the back when a student continued there, and confirm the roster match.
Keep a score when it matches a line on your key. Override it when you disagree, and edit the comment so it names the line that matched. Toggle feedback on when that comment should appear on the PDF or the email, and off when you will go over the item with the class.
The only flag is the Low confidence badge on an answer Graider found hard to read. Open every badge. Read the student's answer as text, compare it with the paper and your key, and override when you disagree.
Graider does not read or score diagrams, graphs, or drawings. When a question asks for a sketch, a graph, or a labeled drawing, set those points yourself inside the one score for that question. A written label you can list, such as "nucleus" when the word is what you score, can stay on the key inside that same score and comment.
Before you finish review, check:
Hand each paper back as a PDF or an email after you finish review. The score is the one you kept. When feedback is on, the comment you kept sits with that question.
Student papers and account data aren't used to train models or for advertising.
Partial credit grading with AI marks photographed paper tests against the lines you wrote on your answer key or rubric. Those lines name method marks, incomplete answers, and accepted variants, each with a finished score. You review the one score and one comment per question, then hand the work back as a PDF or an email.
Give the question a total. Write every full-credit response at that total. Write each method case, incomplete answer, and lower variant as its own line with a finished total at or under that number. A colleague with the key and the student page should award the listed cases the same way you would.
They go in the answer key or rubric text you upload with the test. A method mark names a visible step and the finished score for a paper that shows it. An incomplete answer names what is present, what is missing, and the points. An accepted variant is its own line at the points you chose.
Open every Low confidence badge. It is the only flag, and it sits on a hard-to-read answer. Compare the read with the paper and your key, override any score you disagree with, and set diagram, graph, or drawing points yourself. Then toggle feedback and hand the work back as a PDF or an email.
Check every Low confidence badge, keep the partial-credit lines on your key, and hand the papers back when you can stand behind the scores. You keep the pen.