FeaturesEVO Assess
Assess

The test is built for that role, and they defend their work.

A technical test generated from the role’s competency map, languages with spoken assessment at CEFR level, and a behavioral profile. No webcam, no screen recording: integrity comes from defending your own work out loud.

The problem

Why this exists

How it works today

An off-the-shelf test is the same for a fintech and an agency, and it is exactly the test that cheating tools promise to solve, by name. The market’s answer was more surveillance, but fraud in proctored tests rose from 16% to 35% in one year.

How it works with EVO

The question is original, generated from that role. And the candidate explains their own work live, answering questions built from what they themselves wrote.

How it works

EVO Assess from the inside

1The test

Generated from the competency map, not bought off the shelf.

A deterministic selector chooses what to test from the competencies approved on the role, and the rest of the time budget goes to a real work sample. The model drafts the questions one by one.

The grading rubric is born from the expected deliverable itself, with concrete anchors rather than generic adjectives.

In a controlled experiment with interviewers blind to the purpose, people cheating with AI passed 73% of off-the-shelf questions, and only 25% of original ones, below the 53% of those answering honestly.
Test result · João Lima
Scores proposed by EVOawaiting your review
Continuous discovery78
Autonomy on a lean budget65
Quantitative experimentation (A/B)41

Below what the stage expects: becomes an interview probe, not a cut.

🎯 Becomes an interview probe

"Walk me through the most recent A/B test you designed: the hypothesis, the sample size, and what you decided when the result came back ambiguous."

Adjustments are audited. The score is informative: it never eliminates anyone on its own.

2The defense

Three questions built from their answer, with no going back.

After the submission, the candidate answers live to questions built from what they wrote, and the second reacts to what they said in the first. Fifteen minutes, with no chance to revise.

Whoever did their own work explains it naturally. Whoever did not has nothing to say, and the empty answer is a signal in itself.

Interview sheet · João Lima · technical stage
Question from the guide · Continuous discovery

Tell me about a product decision you changed after talking to a user. What did you believe before, and what changed your mind?

Rubric of anchors
FairCites research, but the decision had already been made
PartialChanged the solution; did not change the understanding of the problem
SuperiorReframed the problem and shows the cost of having pushed on
🎯 Probe generated from this candidate's gaps

"Your assessment flags discovery without an impact metric. In the case you just described, what changed in the numbers afterwards?"

Under the hood

Why no camera is safer, not less safe

Surveillance answers whether someone else is in the room. It does not answer who did the work, and that is the question that matters.

caseDocumented impostors passed screening and two technical interviews with the camera on, with real technical skill
8%In one measurement of a real funnel, 8% of take-home tests were done by someone else
N+1The next question reacts to the transcript of the previous one; an accomplice on a parallel call cannot feed answers in real time
AIThe candidate declares how they used AI, and the declaration costs no points: it becomes context for the grading
LGPDNo score eliminates anyone on its own; human review is mandatory and every adjustment is audited

The psychometrics of the behavioral profile are calculated in code; the model only narrates the result. And behavioral never becomes a cut-off score, it becomes interview material.

What comes next

How it fits into the rest of the journey

Bring a role that is hard to fill

It takes a real role to see EVO work, not a canned example.

Start using it