You didn't sign up to spend your weekend ticking boxes.
Grading eats the hours you don't have, and what comes out the other end is often a number with no explanation for anyone. Arc drafts the rubric, evaluates against it with visible reasoning, and leaves every override in your hands.
The feedback loop was never actually closed.
A rubric turns a nuanced answer into a checklist — a tick for each line item, no room for the fact that a student got the concept right but slipped on notation. Applied by hand, at speed, across fifty sheets, that isn't evaluation. It's sorting. And it takes just as long as real evaluation would.
The result is a feedback loop with only one working half. The student gets a score. They rarely get a reason. By the time the next test comes around, nobody quite remembers what went wrong on the last one — least of all the student who needed to.
Meanwhile, the hours that grading eats are hours that don't go into what actually moves a class forward: a concept re-taught, a harder problem set, a conversation with a student who's quietly struggling. Time spent marking is time not spent teaching.
What changes when reasoning comes with every mark.
Marking alone
- Every sheet marked from scratch, rubric held in your head
- A mark with no explanation attached to it
- Partial credit decided a little differently sheet to sheet
- A parent query means re-deriving why a mark was given
- Evenings and weekends absorbed by the pile
Marking with Arc
- Rubric drafted by AI per question — you confirm it once
- Every mark, full or partial or zero, comes with a written reason
- The same rubric applied identically across every sheet
- Every deduction already documented, ready to explain
- You review and override in a fraction of the time
Every mark, explained — right there in the review screen.
This is the actual review layout: AI's mark, the rubric line it maps to, and the reasoning for anything less than full credit. You accept it, edit it, or override it — nothing publishes silently.
Arithmetic slip in the final step — arrived at 24 instead of 12.
Unit missing from final answer.
Not a black box. A second pair of hands you can check.
Every item here is something you can inspect, edit, or override before it ever reaches a student.
AI-drafted rubrics
For every question, AI proposes a model answer and a partial-marking scheme. You confirm it, edit it, or regenerate it — nothing locks until you say so.
Reasoning behind every mark
No answer is scored in silence. You see exactly which rubric line was met, which wasn't, and why — before you publish anything.
You always have the last word
Disagree with an AI mark? Override it with a required note. That override is logged and carries into how disputes get resolved.
Class and topic analytics
Topic-mastery heatmaps, difficulty breakdowns, and Bloom's-level coverage — so you see where the class actually struggled, not just what they scored.
Reports that write themselves
Per-test, monthly, and yearly reports generate in the tone of the reader — analytical for you, encouraging for the student, plain-English for the parent.
A closed re-evaluation loop
Students and parents can contest a mark; you see the original reasoning beside their objection and respond in one place.
See where the class actually struggled.
A topic-mastery heatmap across recent tests — so a dip reads as 'this class needs another pass at fractions,' not just a lower average.
From blank paper to published feedback.
The same loop whether you're creating a paper from scratch or uploading one you already wrote.
Create or upload the paper
Describe the test you want — grade, subject, chapters, question mix — and AI drafts it. Or upload one you already wrote.
Confirm the rubric
AI drafts a model answer and partial-marking scheme per question. Review it once, edit anything, lock it in.
Evaluate with visible reasoning
Upload answer sheets. AI extracts and evaluates each answer against your locked rubric, reasoning shown for every mark.
Review, override, publish
Walk through each sheet beside the scan. Accept, override with a note, or flag for a second look — then publish.
This is what reviewing a single answer looks like.
The actual scanned answer sits next to the extracted text — nothing is evaluated off-screen, out of view.
"I used to spend Sunday afternoons on a stack of fifty sheets. Now I spend twenty minutes checking what the AI already reasoned through — and actually have Sunday back."
Class 7 Science Teacher
Illustrative example, not an attributed customer quote.
Get your evenings back.
Try the actual teacher workflow — rubric drafting, evaluation review, and publishing — on real seeded data.
Open the teacher demo