Skip to content
ProvenGround

Everyone can use AI now. The question is whether they can direct it to solve a real problem.

ProvenGround measures how candidates work with AI, not just whether they use it.

Get started See how it works
PRD-team-budgets.mdWritten by AI
An AI assistant confidently confirms a customer quote in an AI-written PRD. The candidate asks to see the source, the assistant admits the quote is fabricated, the candidate flags it with proof, and their Record shows they caught the AI's mistake.

Stop hiring for years spent with a tool. Hire for the ability to own what the tool produces.

A take-home used to show whether someone could do the work. Now it mostly shows whether they used AI to do it. The result looks polished either way.

What it doesn’t show is the part that matters: whether the person noticed when the AI was wrong, checked the claims against the sources, and knew whether the work was ready to ship.

ProvenGround tests that part directly.

How it works

From job description to evidence, in three steps.

  1. 01

    Translate the job description

    Start from the role you are hiring for. ProvenGround turns the job description into the failures that role must catch before they reach a customer, a codebase or a board deck.

  2. 02

    Candidates review real AI-produced work

    Candidates get deep, realistic work produced by AI, with faults planted in it. They use AI inside ProvenGround to investigate it, just as they would on the job.

  3. 03

    Code scores it against an answer key

    Scoring is done by code, against a known answer key: what they found, how they verified it, and whether they would sign off on the work.

What we measure

Three layers of ownership

Owning AI-produced work is more than proofreading it. Each layer is a step further from reacting to mistakes and closer to preventing them.

Verification

Catch what AI got wrong.

Check the work against its sources and find the fabrications, contradictions and omissions before anyone relies on them.

Visibility

Build checks that catch what you never saw.

Go beyond reading line by line. Put checks in place that surface problems you would not have spotted on your own.

Control

Coming soon

Define what must always be true.

Set the rules the work has to satisfy every time, so that owning the output does not depend on catching every mistake by hand.

For hiring

See who can own AI-produced work before you hire them.

Invite candidates to an assessment built around your role. Every submission becomes a Record your team can review and discuss, grounded in what the candidate actually did.

  • Assessments matched to the failures your role must catch
  • A Record for every candidate: what they found, what they missed, how they checked
  • Scores tied to evidence and an answer key, not a reviewer’s impression
For training

Build the habit on the team you already have.

Use the same assessments to help people practise reviewing AI output, and to show where their checks hold up and where they need work.

  • Practice on realistic AI-produced work with known faults
  • See exactly which problems were caught and which slipped through
  • Build the habit of verifying before signing off

Hire for judgment, not tool time.

Set up your first role and see how candidates direct AI on work that matters.

Get started