Grading with a rubric: fair and fast assignment feedback

Forty submissions sit in one list. You read the first five closely and give each learner three lines of feedback. By the twentieth the feedback is one line. By the fortieth you type "good, keep going" and put down eight out of ten because you are tired. Two days later a learner asks why his classmate scored nine and he scored eight for much the same work, and you have no answer to give him.

Nothing about that is bad faith, and nothing about it is laziness. It is what happens when you grade without a ruler. This article is about the ruler: how to write one in ten minutes, how to make it shorten your evening rather than lengthen it, and exactly where it stops working.

A mark without a written standard is an impression, and impressions drift

When you grade on overall feel, you are not measuring the work against a fixed level. You are measuring it against the submission you read a minute ago. An average piece that arrives after three weak ones looks excellent. The same piece after three strong ones looks thin. That is not a flaw in you; it is how judgement behaves when it has no external reference to hold on to.

Fatigue makes it worse. The first ten submissions get detailed notes, the last ten get a stock phrase. So does the name on the file: the learner you think of as diligent gets a generous read, the one who was late once gets read while you hunt for the mistake. Put your first mark and your last mark side by side when you finish and the gap in your standard is usually easy to see. While you were grading, it was invisible.

The cost is not only fairness. A learner who cannot predict a mark stops trying to improve and starts trying to negotiate. A trainer with no written reason for a mark spends longer answering objections than grading in the first place. A rubric buys that time back before you spend it.

A rubric is not a list of adjectives, it is an observable scale

Most of what gets called a rubric is a list of pleasant adjectives: organisation, depth, creativity, presentation. Those are not criteria. They are your impression split across four columns. A real criterion answers one question: what can I actually see in this work that puts it at this level rather than the one below?

So a criterion has three parts: a short name, a description of what the work looks like at each level, and a weight saying how much of the total it carries. The description is the part most often skipped, and it is the part doing all the work.

What you usually write	Why it cannot be graded	The observable version
Well organised	Your idea of organised is not theirs	Every section has a heading, conclusions separated from findings
Good analysis	"Good" is not something visible on a page	Every claim backed by a figure or a quote from the case
Followed the brief	The brief itself is not written down in one place	Addresses the three elements named in the task text
Professional presentation	Pure taste, and two people will disagree	One typeface, images no narrower than the stated width

Two graders can agree on the last column. Nobody agrees on the first, and a trainer will not even agree with himself a week later.

Three levels per criterion are enough: above what was asked, at it, below it. Five levels force you to draw distinctions your eye cannot really see, so you end up guessing again inside the very tool you built to stop the guessing.

Write your rubric in ten minutes

1. Write the point of the assignment as one sentence starting with a verb: "the learner analyses an ad campaign and proposes a change grounded in data". If you cannot write that sentence, your problem is the task, not the rubric.
2. Derive three to five criteria from that sentence, no more. Every extra criterion eats into the weight of what actually matters.
3. Describe the top and bottom levels first. The middle then writes itself: less than the top, more than the bottom.
4. Weight by skill, not by visible effort. If the assignment is about analysis, presentation does not take a fifth of the mark.
5. Collapse the formatting rules into one criterion. Typeface, layout and word count belong on a single small line, not three that inflate form over substance.
6. Publish the rubric with the assignment itself. A standard revealed after submission is a penalty. The same standard revealed before it is teaching.

Ten minutes, once, then reused with every cohort you set this assignment for. That reuse is the moment a rubric stops being extra admin and starts being saved time.

Grade in the right order, then write feedback that changes the next submission

A rubric on its own is not enough if the order you grade in fools your eye. Six habits move the result more than any tool does:

• Skim the whole list before awarding a single mark. Ten minutes gives you the real range in this cohort, so you do not spend your top marks on the first acceptable piece you open.
• Grade one criterion across every submission before moving to the next. Holding one criterion in your head is sharper than switching between four on every piece, and if you do drift, the drift stays inside one criterion instead of spreading across the whole mark.
• Score each criterion before you think about the total. Start from the total and the criteria will quietly rearrange themselves to justify it.
• Do not look at the name while you are judging. Look at it afterwards, when you write the feedback: the feedback needs their history, the mark does not.
• After the last submission, reopen the first three. If you change your mind about them, you drifted along the way, and the later judgement is usually the better one because you now know the task.
• Do not split one assignment across two sittings days apart. The day in between hands you a different ruler.

This is where something that does not grade for you genuinely helps: a first read that shortens the skim and flags the passages worth your attention, after which you read and decide. AI-assisted review on Dwrrah works on exactly that logic. It comes before your decision, never in place of it.

Then there is the feedback, which is the other half of the job. The mark tells a learner where they stood; the feedback tells them how to move, and most written feedback does neither. Build it from three parts and stop there:

1. What specifically worked. Not "excellent" but "you tied the drop in conversion to the page change, and that link is the point of the task".
2. What cost them the mark. In the language of the rubric, so they can see the line between what they read beforehand and what they got.
3. One step for next time. One they can actually carry out, not five pieces of advice that evaporate together.

Four lines that get read beat ten that get ignored. And do not rewrite the work for them in the comment: if you fix the paragraph, you are the one who learned something.

Where the system saves you time, and where it stops

Everything above works on paper too. What the assignments and grading system removes is the friction around it. The rubric is written once alongside the assignment, so your learner sees it before starting. Work arrives as a file, a link, or text typed inside the platform, depending on what the task actually is. Submissions collect in one list that shows who handed in and who is late, with no counting by hand in your inbox. You set a due date, and reminders reach your learners before it without any chasing from you.

An assignment can belong to the whole course, to a session, or to one class inside a multi-meeting program, so it reaches only the learners that class covers. Every mark and comment stays on the learner's record beside their progress and their quiz results, so when you ask two months later whether they earned the certificate, the answer sits in one place.

The difference between the two tools deserves a sentence. A quiz marks itself and measures understanding with defined answers. An assignment collects work a human judges against a rubric that same human wrote. Confuse them and you end up using a quiz to measure a skill no question can reach.

So that nothing here reads as a promise no one made, these are the limits, stated plainly:

• It does not remove your judgement. Written descriptions reduce drift; they do not end it. Your first cohort will expose a vague phrase or two in your own rubric, so fix those before the next one.
• The AI does not set a mark. The first read shortens the review and points at what deserves attention. The final mark and the written feedback are yours alone.
• There is no automatic late penalty. The system shows you who was late and when. Whether you accept late work is your policy, so state it in the assignment text from the start.
• There is no originality check. Nothing here tells you the work is not theirs. The fix is design, not technology: tie the task to the learner's own situation or to data you handed them, so copying is worth nothing.
• A rubric cannot rescue a vague task. If the assignment text does not say precisely what is wanted, no scale repairs that.

What ruins a rubric

First: nine criteria. A long list feels rigorous and is the opposite. Each weight shrinks until none of them moves the mark, and grading takes longer for nothing. Three to five, then stop.

Second: equal weights for everything. When formatting weighs the same as analysis, you have announced that form is half the skill, and your learners will believe you and spend their time where the marks are.

Third: writing the rubric after reading the first submission. You are not writing a standard then, you are describing that one piece of work, and everyone who follows is measured against it through no fault of their own.

Fourth: deducting for something you never mentioned. If it is not in the published rubric, do not take marks for it. Add it to the next version and flag it in the next assignment instead.

Start with one assignment you set every cohort. Give it three criteria with level descriptions and publish it with the task. Grade one cohort that way and measure two things: how long grading took, and how many objections landed in your messages. Those two numbers will tell you what no article can.

Assignments and grading sit on the Pro plan, alongside quizzes and certificates. Before committing to anything, you can try the rubric, the grading list and the AI first read during the free trial, which runs for fourteen days and needs no card.