How does scoring actually work?

You pick the requirements for the job and set a level on each one. You set up the job description, and that JD is scored against too. During the interview the questions adapt to what the candidate says. The score comes back mapped to your requirements and your JD.

Where do the suggested requirements come from?

They are generated from the job description and fill the requirement rows. You review, edit, and set the level on each one before it is saved.

Does the JD really affect the score?

Yes. The JD is one of the two things answers are scored against. A thin JD gives you a thin score.

Are the questions the same for every candidate?

The starting set can be. The follow-ups are not. They are built from what the candidate actually says.

Can I change the requirements after candidates have interviewed?

You can edit the job, but keep a batch consistent. Comparing candidates is only fair when they were scored on the same requirements at the same levels.

Is the score a hiring decision?

No. It is evidence for a person to decide on. There is no automated pass or fail, and integrity flags are always reviewed by a human.

ScoringFor recruiters and in-house teams7 min read

Set requirements and levels

You pick the requirements for the role and set a level on each. The JD is scored against too. Questions adapt to the answer. The score comes back on the requirements you chose.

Before you start

A job with a real job description on it, not a placeholder.
A shortlist of the requirements the role is actually hired on.

Steps

5 steps, in order.

Pick the requirements

Open the job and go to Requirements. Press Add Skill to add a row, or press Suggest requirements to have it read the JD and fill the rows. Until you do either, the empty state tells you to click Add Skill to start. You review the list before anything is saved. Keep the requirements you will genuinely ask about and remove the rest with the X on each row.

Create role

Requirements

Step 2 of 3

Candidate requirements * Add Skill

Weight each skill 0-100. At least one must be 60 or above.

PostgreSQL

Distributed systems

AWS

API design

Kafka

Terraform

Kubernetes

Suggest requirements60 and above is a must-have, 30 is a nice-to-have.

BackContinue

Candidate requirements, each row with a level dropdown, generated from the JD and edited by you before saving.

Six to ten requirements is usually enough. A list of thirty leaves nothing to separate candidates on.

Set a level on each requirement

Each requirement carries a level, picked from the dropdown on its row. The level is how you say what the role stands or falls on. A backend job puts system design and SQL at the top level and drops a nice-to-have framework lower.

There is no percentage and no slider here. The level on each row is the whole of the weighting.

Set up the JD, because it is scored against too

The job description is the second input. Answers are scored against the JD as well as the requirements. This is why a two line JD produces a weak score and a JD with real responsibilities, a real stack and a real seniority produces a useful one.

If you paste a JD from another role and forget to edit it, candidates are scored against that other role.

Let the questions adapt to the answer

During the interview the follow-up questions are built from what the candidate actually says. A vague answer gets probed. A specific answer gets pushed further. That happens whether you picked the questions yourself or took the suggested set.

Coding assessment

Sandboxed coding test

16:07 End

1Coin Change (Minimum Coins)

Python 5 of 5 passed

def min_coins(coins, amount):

dp = [float("inf")] * (amount + 1)

dp[0] = 0

for c in coins:

for x in range(c, amount + 1):

dp[x] = min(dp[x], dp[x - c] + 1)

if dp[amount] == float("inf"):

return -1

return dp[amount]

# greedy fails on coins like [1, 3, 4]

def solve(lines):

n, amount = map(int, lines[0].split())

coins = [int(t) for t in lines[1].split()][:n]

return min_coins(coins, amount)

if __name__ == "__main__":

print(solve(sys.stdin.read().splitlines()))

5/5test cases passedComplexity O(n·k)

Sample 112ms

Sample 29ms

Edge: no solution11ms

Large amount31ms

Single coin8ms

Custom input (optional)

stdin for a manual Run

Run Submit

A live round. What the candidate does here is the evidence the score is built from.

Read the score back on your own skills

The report breaks the score down by the skills you chose, with a bar per skill. Every part of it traces back to the transcript, the code the candidate wrote, the requirements you set and the job description. If a skill looks wrong, open the transcript and check the answer behind it.

Python engineer

Jordan Blake

Download PDF

OverviewAnalysisSubmissionsResume Notes

Flagged for suspicious behaviour3 signals

78Score

Overall performance

Good fit for roleCodingRound 2 of 3

Strong Flask expertise, one unfinished task.

Skill breakdownWeighted against the JD

Flask / REST APIs88

Code quality79

System design74

Communication71

Report summaryPrompt: Flask API over an Elasticsearch job index

Nine years on Python backends, mostly Flask and REST. Built the search route cleanly, then stopped short of the Elasticsearch query and said so.

Strengths

Flask structure modular routes, CORS, jsonify
Production caching Redis layer, eviction reasoning

Areas for growth

Elasticsearch query and relevance sort unfinished
Distributed locks named deadlocks, not the fix

Responses to screening questions

Okay working from the office five days a week?Yes.

Notice period with your current employer?30 days, can be bought out.

Misconduct events

Tab switch × 2Multiple monitors × 1

Resume scoreResume.pdf37

Recording16:41, transcript 96 lines

The skill breakdown on the report, scored from the transcript and the code.

What you get

At the end of this guide.

A requirement list you chose, each row carrying its own level.

A level per requirement that says how much it matters.

A JD that is used as scoring input, not just as an advert.

A per-skill score on every report, traceable back to a line in the transcript.

Common mistakes

What goes wrong, and the fix.

Putting every requirement on the same level.

If everything is Critical, the score cannot tell a strong hire from an average one. Pick the two or three that genuinely decide the role.

Keeping a requirement you never ask about.

Either add a question that tests it or remove the row. A requirement with no evidence behind it is noise.

Reusing a JD from a different seniority.

Edit the JD before you interview. It is scoring input, so the wrong JD scores the wrong role.

Arguing with a score without opening the transcript.

Every score traces to evidence. Open the Analysis tab and read the answer it came from.

FAQ

Questions people ask on this one.

Still stuck? Talk to us and we will walk through your setup.

Related guides.

Job setup

Create a job

Fill in Job details, add the job description, add requirements with a level on each, pick the interview language, and create a job that rounds can be added to.

6 min read

Reports

Read a candidate report

The four tabs, the overall score, the skill bars, the recording and searchable transcript, the malpractice status, and how to share it.

6 min read

Rounds

Add an AI interview round

Walk the Create a round wizard. Name the round on Basic Details, pick the Interview type, set the round type and difficulty, fill the Questions step, and set the score cutoffs on Workflow.

7 min read

All guides

Try this in your own workspace.

The 30-day trial includes 75 Hire Credits, which covers a resume batch and a first round of interviews. Nothing charged for 30 days.

30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days.