Wisdom in the crowd
BetaCollective intelligence you can trust.
Fairgrade weights every human judgment by demonstrated credibility, so that a crowd of imperfect evaluators reaches a verdict more reliable than any expert working alone. Peer assessment in higher education is where it runs today.
Are you a student? Join a class →
Every dot is an evaluator. Its weight is credibility, earned from the record rather than granted. The centre is the verdict they converge on.
The crisis in education
Assessment is where the university is breaking.
After the cost of tuition, the assigning of grades is the largest source of frustration in higher education, for faculty and students alike. Grading qualitative work well, and returning feedback a student can actually use, takes time that scales with class size. Past roughly twenty-five students per instructor, that feedback stops being meaningful. Courses respond the only way they can: fewer essays, more multiple choice, and less of the thinking education exists to teach.
Then generative AI arrived, and most gradable work stopped being evidence of thinking at all. A model can now produce the essay, the case write-up and the problem set. What it cannot produce is the judgment to tell good work from plausible work. The artifact can be generated; the judgment cannot be outsourced without leaving a trace. So grade the judgment. The moment a course does, the cheating problem becomes a measurement problem, and judgment, the skill that survives automation, becomes the thing the classroom actually teaches.
92%
of undergraduates now use generative AI; 88% have used it for assessed work
2×
the share of teenagers using ChatGPT for schoolwork in a single year
1 in 9
papers submitted for review shows substantial AI-written content
1 : 25
the faculty-to-student ratio past which feedback stops being meaningful
Sources: HEPI / Kortext, Student Generative AI Survey 2025 · Pew Research Center, January 2025 · Turnitin, April 2024: 200 million papers reviewed · Lee et al., CHI 2025, on generative AI and critical thinking. The ratio is the founders’ own threshold from forty years of teaching.
The thesis
The cost of generating an answer is now zero. The cost of trusting one is not.
When information is free, the scarce thing is knowing which of it to believe. Crowds have always been part of the answer: juries, markets and ask-the-audience beat lone experts more often than intuition suggests. But a simple average treats every opinion as equal, and throws away the most valuable thing a crowd produces: its structure. Who has been right before. Who knows the field. Who has an interest in the outcome. How the votes distribute, and where they disagree.
Fairgrade reads that structure. It weights every judgment by the demonstrated credibility of the person making it, and treats disagreement itself as information. The result is not the wisdom of the crowd but the wisdom in it.
A simple average
Fast and honest, and wrong whenever the crowd is uneven. A careless score and a careful one count the same, so a few bad actors move the result.
A black-box model
Data-hungry and hard to explain. Without a theory of how people judge, it needs volumes of history that most real decisions never have.
Credibility-weighted consensus
Grounded in probability theory. It works from a single class of reviews, explains every weight it assigns, and improves as the record grows.
AI has a place in this. It is a capable second opinion set beside the human verdict, for a person to weigh. It is never the source of trust.
The method
Two unknowns, solved together.
Whose judgment to trust cannot be known without knowing the answer, and the answer cannot be known without weighting the judgments. The Marciano Method estimates both at once: scores imply credibility, credibility re-weighs scores, and the loop runs until the two agree. Below is a simplified version, running live. Try to break it.
Dashed line — the grade the instructor gave working alone · scale 55–95
A simplified model of the Marciano Method, running live in your browser. Credibility is earned by agreement with the emerging consensus — and re-earned every time you interfere.
Fairgrade for Education
The first deployment: peer assessment a professor can stand behind.
A Fairgrade assignment is a real assignment, graded by the whole classroom under the instructor’s rubric, weighted by the method, and finalized by the instructor. Students learn the material twice: once producing the work, once judging it. Instructors get their evenings back, and every submission gets more feedback than one grader could ever give.
1Build
Create the assignment and design a rubric: criteria, weights, exemplars.
2Assign
Smart peer assignment distributes reviews fairly and balances workloads.
3Grade
Students give and receive structured reviews on every submission.
4Analyze
Compare the weighted consensus with an optional second opinion, finalize, publish.
Runs inside Canvas, Brightspace, Moodle and Blackboard through LTI 1.3, and returns grades to the gradebook the institution already trusts.
The evidence
A decade of classroom proof.
5,000+
students in pilots
60+
courses at NYU Stern
~60 hrs
saved per instructor, per course
10 years
of research behind the method
$100K
Rennert Prize, NYU's Entrepreneurs Challenge
“The rubric was SUPER helpful, as well as grading other people’s work. I honestly wish other classes let us do this — sometimes we submit our answers and don’t get any feedback at all, which isn’t exactly useful for learning.”
“I think Fairgrade is pretty cool — and the underlying technology obviously has applications outside of the grading functionality. It was an interesting addition to the coursework in lieu of homework.”
Piloted at NYU Stern. Source: NYU Stern, June 2020 →
Beyond grading
Grading is the first proof.
The engine under Fairgrade does not know it is grading essays. It knows how to estimate the truth from many observers of uneven reliability, a problem that appears wherever expert panels, reviews, forecasts or open questions meet a crowd. Education is where the method is proving itself, one classroom at a time, and where its credibility is being built.
- DeployedPeer assessment in higher education
- Research directionExpert panels and peer review, where reviewers of uneven reliability judge the same work
- Research directionForecasting and risk assessment, where track record should outweigh title
- Research directionTalent and performance review, where bias is the known failure
For instructors
Get your grading nights back.
Set up your first assignment in minutes. Free for your first course, no credit card required.
For institutions
Assessment that scales with enrollment.
Structured pilots, FERPA-ready architecture, and evidence from 5,000+ students. We will design one for your department.
Are you a student? Join a class →