Everyone builds with AI now, so nobody can tell what anyone is actually good at. Evalor gives you one real problem, judges it against a bar you can read, and keeps a record of which decisions were yours and which the model made. The result is a profile someone can check.
Free to start, no card. Hiring? The second link is what a shared profile looks like.
Priya Nair
Proof profile
“Evaluated artifact, score, and judge notes are now packaged into one profile link.”
Evaluation
Problems taken from how the work actually arrives
The problem
The CV is written by AI. So is the cover letter, the take-home, and half the answers in the screening call. None of it is cheating exactly — it is just that every candidate now produces the same polished surface, and none of it says anything about the person.
So interviews got longer. Take-homes got bigger. And both sides lost: candidates spend weekends on unpaid work that proves nothing, and hiring teams read forty submissions that could have been written by the same model.
What survives is evidence. Real problems, worked in the open, judged against a bar someone can inspect — with a record of which decisions were yours and which the AI made.
Yours, or the AI’s
Everyone works with AI now, and pretending otherwise would make the assessment useless. So the copilot is there — it drafts, it explains, it applies what you ask.
But every change it makes is recorded against the ask that caused it, and every change you make is recorded as yours. On an architecture that is per-block: who added it, and whether a human ever checked it.
On the parts being assessed, the copilot refuses to decide. Ask it what is wrong with the flow and it will tell you that is the call you are here to make.
On this submission
Who it's for
A candidate needs somewhere the work is finally visible. Someone hiring needs to know the work was theirs. It is the same page — drag the profile to read it from either end.
If you did the work
A CV claims. A profile shows — the artifact, the bar it was judged against, and which calls were yours.
One profile
The artifact, the score, the evidence, and who made each call.
If you have to decide
Everyone builds with AI now. The question is no longer what they shipped — it is which decisions were theirs.
If you did the work
A CV claims. A profile shows — the artifact, the bar it was judged against, and which calls were yours.
If you have to decide
Everyone builds with AI now. The question is no longer what they shipped — it is which decisions were theirs.
One profile
The artifact, the score,
the evidence, and who
made each call.
← drag →
How it works
Each problem is worked in the surface it belongs in — an architecture canvas, a document, an API contract — and defended to a stakeholder who pushes back.
A live situation with missing information, a stakeholder who will not volunteer it, and a constraint that changes halfway through. Not a quiz.
Interview the stakeholder. Write the memo, rewire the architecture, build the prototype. An AI copilot works alongside you — and on the parts being assessed, it refuses to decide for you.
Each criterion has four written levels describing exactly what a 1 and a 4 look like. You can read the bar you were measured against, and disagree with it.
The work, the reasoning, the score, and the record of which decisions were yours. One link, and it holds up when someone opens it.
Twenty-six blocks — routers, guardrails, retrieval, human handoff, evals. Run it against sample traffic and watch what it mishandles.
Diagnosis, decision memo, metric definition, the message to the person who will push back. Judged on reasoning, not word count.
Error semantics, idempotency, the migration path — the calls other teams will depend on for years, and the failure nobody wants to own.
A stakeholder with their own agenda, who contradicts themselves under questioning and only reveals the constraint if you ask well.
The result
Why it works
Not a vague rubric. Each requirement has four written descriptions — what a 1, 2, 3 and 4 look like for that specific criterion. When Evalor gives you 68 on Judgment Calls, it quotes the exact decision and shows what 80 would have required.
Judgment calls · Retention challenge
Fix next68 / 100
You correctly identified churn as an activation problem, but jumped to roadmap changes before proving which user segment is most at risk. A stronger answer separates new-user activation from mature-account retention, then names one metric owner for the next sprint.
•Add segment-level diagnosis first
•Name the decision owner explicitly
•Define a two-week success metric
•Cut the low-confidence roadmap bet
Scores, artifacts, judge notes and the attribution record compound into one link — with a case-study video for each challenge and a record of everything you have done. Pick a theme, decide what stays private, and post it.
Four themes, one link
A video of the work
Each finished challenge renders its own case-study walkthrough. Nothing to edit.
A record that accumulates
Every submission, dated. Consistency is its own signal.
Priya Nair
3 evaluations · Product and Technology
82
avg score
Evaluated work
Shipping the refund agent
Eval design
The agent that can spend
Security threat model
p99 tripled at 02:14
Debug & postmortem
What gets measured
Anyone can produce a polished artifact now. What separates people is the thinking underneath — so that is what gets read, against anchors written down before you start. Four of the questions:
What do you do with the information nobody gave you?
Assumes the gap away and moves on.
Names the assumption, says what would disprove it, and goes and asks.
What did you give up to make that call?
Presents the choice as though it were free.
Names the cost, who carries it, and what would make you reverse it.
How do you move someone whose agenda is not yours?
Restates the plan louder to a person who already disagreed.
Finds the constraint behind their position and writes to that.
What happens when the constraint changes halfway?
Defends the original plan because it is the original plan.
Re-cuts the scope and says plainly what is no longer true.
Challenge library
Product and engineering are evaluated the same way — a real problem, a real artifact, and a record of how you decided. The tracks differ; the bar does not.
Judgment under ambiguity: what to build, why, and what you gave up.
Engineering judgment with an AI agent: find the cause, change the code, prove it works.
Working on something real? Bring your own problem into the same workspace. No scoring.
Browse all challenges →Why we are building this
We spent years making calls that mattered and left no trace. The decision lived in a meeting, the reasoning in a doc nobody reopened, and the credit wherever it landed.
Then AI arrived and made everyone’s output look the same. The work got easier to produce and impossible to read.
So we are building the thing we wanted and never had: somewhere the work is judged against a bar you can read, and the record says which parts were yours.
The Evalor team
Founder-operated · Amplifigen Product Innovation Inc.
Evalor beta
Do a real problem, get judged against a bar you can read, and keep a record someone can check. Free to start.
Free to start, no card. Hiring? The second link is what a shared profile looks like.
Beta access · Free evaluation access · No credit card required