Research · Pilots open
The interview panel is a crowd.
Résumés, take-home tests and interview answers have become cheap to produce and expensive to trust. The scarce signal in hiring is judgment: how well a candidate evaluates work, and how well the people evaluating candidates evaluate. The Marciano Method measures both at once, from a single hiring drive. This page sets out the problem, what the method changes, what has already been shown, and the pilot we are seeking partners for.
The problem
Every signal a recruiter relied on is getting cheaper to fake.
The artifact is free.
A model writes the résumé, completes the take-home and rehearses the interview. For professional and technical roles, the document in front of a recruiter no longer says who produced it or what they know.
The judges are never measured.
A hiring drive runs hundreds of interviews, and nobody records whose judgment proved right. Some interviewers are harsh, some lenient, some favour a college or an accent. Their verdicts are averaged as if they were equal.
The cost lands later.
A wrong hire shows up as early attrition, a replacement search and a client who trusts the agency a little less. By then the interview scores that led there have been forgotten.
What the method changes
Let the pool judge, and judge the pool.
Fairgrade asks every candidate in a drive to do two things: produce a sample of real work, and evaluate the work of several peers against the same rubric. Recruiters and hiring managers grade a sample of the same work. The Marciano Method then estimates two things together, how good each piece of work is and how credible each evaluator is, letting each estimate sharpen the other.
The result is a hiring signal that is hard to fake and cheap to collect. A candidate's evaluation of others' work is evidence of their own mastery: a person cannot reliably tell strong work from weak without understanding the material. A recruiter's record of agreement with the eventual outcome is evidence of their judgment. Both are measured on the same scale, from the same drive, and both tell the hiring team whom to trust and whom to match to the opportunity.
Most of the evaluation moves to the pool itself. Recruiters stop grading everyone and start grading a sample, and their hours go to the candidates the method has already placed at the top. Fairgrade does not try to detect whether a model helped write the résumé or the sample. It measures what cannot be delegated: the quality of a person's judgment.
The three dimensions of credibility, applied to hiring and staffing
Judgment
Whose verdicts tracked what happened. An interviewer whose yes predicted a candidate who stayed and performed earns weight. One who ran consistently harsh or lenient is adjusted for it.
Mastery
Demonstrated expertise, in the work sample and in the evaluation of other people's samples. A candidate who can tell strong work from weak has shown something a résumé cannot.
Bias
The tilt toward a college, a company name, a manner of speaking, or work that resembles one's own. The method looks for the pattern and discounts it rather than averaging it in.
How credibility is surmised, on the Science page →
Where this applies
The same pool, the same work sample, the same ground truth, in five settings. Each is a line here rather than a page, because the problem does not change shape when the industry does.
Staffing drives
Professional and IT mandates, where proxy interviews and inflated résumés cost the most and a client's trust is on the line with every placement.
Hire-train-deploy and apprenticeship cohorts
A cohort is already the shape the platform runs. Trainees produce and judge work for weeks; the deployment decision is the hire.
Campus hiring
Hundreds of near-identical résumés from the same few colleges. The pool itself ranks the candidates, and the recruiters grade a sample.
Cross-border talent mobility
An employer abroad decides on a person it cannot interview properly, from a credential system it cannot read. A record of measured judgment travels where a certificate is not trusted.
Internal mobility
Promotion and transfer panels, where the judges are colleagues with histories and the ground truth arrives in the next review.
Try it
One work sample, twelve evaluators, and the trouble you add.
Below, a pool of candidates and recruiters scores one work sample. Add a careless evaluator, a harsh one, or two who agree with each other and no one else, and watch what each does to a simple average and to the weighted estimate. It is a simplified model, rigged in one direction only: against the bad actors you add.
Simple average
—
Marciano Method
—
A simplified model of the Marciano Method, running in your browser. The dashed line is a senior reviewer’s careful score, the benchmark the estimate is checked against. Credibility is earned by agreement with the emerging consensus, and re-earned whenever the pool changes.
What is already shown
The method has a record. Hiring is where it is tested next.
The Marciano Method was built and proven in the field where the answer can be checked most easily: university grading, with the instructor's own grade as the benchmark.
5,000+
students assessed at NYU Stern
60+
courses, checked against the instructors' own grades
10 years
of research behind the method
$100K
Rennert Prize, NYU's Entrepreneurs Challenge, 2020
No hiring result appears here yet. When the first pilot reports, its numbers replace these.
What a pilot would measure
Five questions with checkable answers.
A pilot is worth running only if it can be wrong. These are the questions it would answer, each against an outcome the partner already records.
- 1
Does the weighted estimate of a candidate predict retention and early performance better than the unweighted average of the same evaluations?
- 2
Does a candidate's evaluation of peers' work predict their own performance on the job?
- 3
How many evaluators does a calibrated pool need before another one stops changing the verdict?
- 4
Does credibility weighting reduce measurable bias by college, gender or background, compared with the average?
- 5
How many recruiter hours per hire does the pool absorb, at equal or better accuracy?
What a pilot looks like
One drive or one cohort. Twelve weeks. No charge.
The pilot runs on the Fairgrade platform as it exists today, inside one hiring drive or one training cohort, and reports against the partner's own outcomes.
- Who judges
- The candidates in the drive, reviewing one another's work samples, and a small number of recruiters or hiring managers grading a sample of the same work.
- What they judge
- A work sample produced inside Fairgrade against a rubric the hiring team writes: a case, a specification, a piece of code, a client email.
- The ground truth
- The hiring decision, client acceptance, and retention and performance at ninety days, recorded after the fact and compared with every evaluator's verdict.
- Duration
- Six to twelve weeks from the first drive to the first outcomes.
- Cost
- None. Fairgrade runs the pilot in exchange for a case study, published with the partner's approval.
- What the partner keeps
- A calibrated view of its own evaluators, a ranked pool, and the first option on commercial terms when the method is offered in this field.
Run the pilot with us.
We are seeking two or three staffing firms or employers for research pilots in 2026 and 2027. If your organisation hires through many evaluators and wants a signal it can check, write to us.
General enquiries: info@fairgrade.com · All research →