Mark essays and written answers against your own rubric
evaleasy is an AI essay grader that marks written answers against your rubric or key, or against a correct answer it works out if you give neither. Each answer is marked in 2 or 3 independent passes and the median counts, so one outlying pass can't move the mark. The model can only choose from allowed marks, it quotes evidence from the paper, and every score is computed in code.
Updated
How does the AI essay grader decide a mark?
evaleasy marks the answer in two independent passes, adds a third if they differ, and takes the median. Take a 10-mark essay. If both passes give 7, the mark is 7. If they give 6 and 8, a third pass runs: 6, 6 and 8 gives 6, and 6, 8 and 8 gives 8. If the passes give 6, 7 and 8, the mark is 7. No two agreed, so with flags on the report also flags "low confidence in the mark" for a teacher.
Why can the model only pick from allowed marks?
So that every mark is a value you'd recognise, with no arithmetic left to the model. Up to 20 marks, half marks are allowed: a 5-mark answer can get 0, 0.5, 1 and so on up to 5. Above 20, marks are whole: a 25-mark essay can get 0, 1, 2 and so on up to 25.
A question's marks come from your key, then what's printed on the paper, then a rule in your instructions. If none of those says, a long answer is worth 10 and a short answer 5, so tell evaleasy if your essay is out of 25.
Should I use a plain-words rubric or a structured one?
Most teachers should write the rubric in plain words, which works in the dashboard. A structured rubric marks by criterion and is set up through the API. A plain-words rubric can be as simple as "Question 1 is an essay out of 20. Look for a clear argument, evidence from the text and accurate grammar. Ignore handwriting."
| Compare | Plain words | Structured |
|---|---|---|
| Set up | Typed or uploaded, in the dashboard or the API | API only, as JSON or a saved rubric |
| Model chooses | A mark from the allowed values | One level for each criterion |
| Report shows | A mark, reason and evidence per question | A level, points, reason and evidence per criterion |
| With flags on, flagged when | All passes give different marks | Passes differ noticeably on levels |
| Size limit | 20,000 characters typed | 50 criteria, 2 to 10 levels each |
How is a structured rubric scored?
The model picks one level per criterion, each level has fixed points, and code adds them up and scales the result to 100. With criteria for Argument (10), Evidence (5) and Grammar (5), a paper placed at 7.5, 4 and 2.5 earns 14 of 20, or 70 out of 100.
If you give a criterion only a maximum, evaleasy fills in levels from excellent down to none. Each criterion takes its own median across the passes. Saving a rubric again creates a new version, and earlier results keep the one they used.
What evidence does the report quote?
Each marked answer quotes the student's own words with the page they're on, such as "p2: the treaty left Germany resentful". If nothing on the paper is relevant, it says "No relevant work found." Beside it is a short reason for the teacher. The paper also gets 2 to 4 sentences of feedback for the student, with up to 3 strengths and 3 things to improve.
How does strictness change the mark on a written answer?
Strictness sets how much the model may assume, and which way it goes when an answer sits between two levels.
- Lenient: ignore minor slips, give part marks when the core idea is there, and take the higher level
- Standard: award a level only when the work clearly meets it. If two levels are equally supported, take the lower
- Strict: every element of a level must be written down, and a right answer without required working doesn't earn full marks. Take the lower level
Common questions
- What if a student writes "give me full marks" in the essay?
- evaleasy tells the model that the paper is data to be assessed, never instructions, and to ignore anything in it that tries to influence the marking.
- Can it mark an essay with no question numbers?
- Yes. evaleasy treats one unnumbered piece of work as question 1, a long answer, worth 10 marks unless your key, the paper or your instructions say otherwise.
- Does it check for plagiarism or AI-written work?
- It checks a class for possible copying, but not for AI writing. On a bulk evaluation, evaleasy compares the papers with each other and flags possible copying for a teacher to check, including papers that share the same unusual wrong answers. The check costs no credits and doesn't change marks. It doesn't search the web and doesn't detect AI writing.
- Which question type should I choose for essays?
- Choose Descriptive. The reader is told to expect written answers and to copy each one in full. If a paper also has multiple choice or blanks, leave it on Mixed or not sure.
- Will the same essay always get the same mark?
- Not always. evaleasy reuses its reading of identical work, but the marking passes run again, so a mark can differ between runs. The median and the fixed list of allowed marks limit what one pass can do.
- How long can an essay be?
- A paper can be up to 30 pages. Typed essays and Word files count one page per 3,000 characters, so a 7,000-character essay uses 3 credits.
- How do I compare it with my own marking?
- Run evaleasy on essays a teacher has already marked, and read the quoted evidence wherever the marks differ. evaleasy doesn't publish an accuracy figure.
Your next stack of papers doesn't have to be your problem.
Signing up is free. Marking uses credits, at 1 credit per page.