For teams hiring engineers

Look under the submission.

Replace your take-home task with a realistic trial. You get a report on how your candidate worked with an AI agent and what they submitted.

Your candidate works on their own computer in their chosen harness, such as Claude Code, Codex or OpenCode.

Choose a layer
01SubmissionSample
src/events/handler.ts
src/events/dedupe.ts
src/refunds/status.ts
test/duplicates.test.ts
docs/runbook.md

The files your candidate handed in. Nothing about how they got there.

01Submission
02Session recordSample
Candidateok, continue
AI agentTwo of seven fail: the daily totals and the refund status.
02Session record
03Observations · VerificationSample

The evaluator uses the evidence to choose a level for each dimension.

03Observations
04ReportSample
Understanding of the problemlevel 3 of 4
Analytical thinkinglevel 3 of 4
Decision making · not enough evidence
Own ideaslevel 2 of 4
VerificationChecked and followed through · level 3 of 4Read Verification in the sample reportAbove level 2: Jonas read what the checks showed and had a failing one fixed and checked again. Below level 4: one failing check was left as it was, and the agent's assumption about refunds went unquestioned.
Meets the requirementslevel 2 of 4
Quality of executionlevel 2 of 4
Overall score52/100
04Report

One trial, four layers. A take-home task shows you only the top one. Open a layer to read it, or follow a quote down to the level it supports.

How it works

A trial builds up layer by layer

See how your candidate explains a problem to an AI agent, steers it and checks what it made.

  1. 01

    Assign a trial

    Add your candidate to an opening. PairTrial draws a task from its pool and emails the invitation. You can also choose the task yourself.

  2. 02

    Give your candidate a realistic task

    The exchange captured during the trial becomes the session record. The candidate hands in their work as a submission.

    Adds 01 Submission · 02 Session record

  3. 03

    See what the evidence shows

    Your report draws on the submission and the session record. The evaluator, an AI agent, reads the submission without running it and quotes the evidence behind its observations.

    Adds 03 Observations

  4. 04

    Read the report and decide what's next

    See an overall score out of 100, a level for each dimension with enough evidence, and a rationale that explains each level. The decision stays with your team.

    Adds 04 Report

For the tech lead

Check the evidence yourself

Open a quote to see its context. In session record quotes, you can see what the candidate asked the AI agent to do and how the agent responded.

For the recruiter

Understand how your candidate worked with AI

Start with the written summary. You'll see what the candidate checked, where the work fell short, and what the overall score covers.

What the report looks at

These are the default weights. A task's rubric can give some dimensions more weight than others.

How the candidate worked · 60%

  1. 15%Understanding of the problem
  2. 10%Analytical thinking
  3. 10%Decision making
  4. 10%Own ideas
  5. 15%Verification

What the candidate submitted · 40%

  1. 25%Meets the requirements
  2. 15%Quality of execution

Book a demo

Walk through a realistic engineering trial and see how to read its report with your hiring team.

Enter your work email address.