Hire engineers for how they work with AI, not whether the tests went green
Candidates solve a real problem alongside Claude Code. You review the Replay — every Instruction, Approval and edit — with evidence for every claim.
export function take(b: Bucket, now: number) {
const elapsed = now - b.updatedAt;
b.tokens = Math.min(b.capacity,
b.tokens + elapsed * b.refillPerMs);
b.updatedAt = now;
if (b.tokens < 1) return { ok: false };
b.tokens -= 1;
return { ok: true, remaining: b.tokens };
}✓ bucket refills over time
✓ rejects when empty
2 passed · 0 failed
Everything you need to run a technical round
From the Brief to the Decision, in one place.
Catalog of Assessments
Pick a real problem with a Brief, limits and Expected insights. Every edit ships a new version.
Invites that reserve Credits
One Credit per Session, reserved on send and released if you revoke.
3 Credits will be reserved · 39 available
A lobby without accounts
Candidates read the Brief, limits and what we record, then consent. English and Spanish.
I understand the clock never pauses, even if I disconnect.
Approvals, not autopilot
The Agent proposes. The Candidate approves, amends or rejects — and that choice is the signal.
Hidden checks
Candidates can run them any time and see only the count. You see which ones passed.
Replay
Every event on one timeline, coloured by who acted.
Dimensions with evidence
The Evaluator scores each Dimension 1–4 and cites the Replay events behind it.
Disagree, on the record
Your score sits next to the Evaluator's. Neither replaces the other.
A Decision only you make
Advance, Pending or Don't advance. Never pre-selected.
The Agent does the typing. The Candidate does the engineering.
A Session is a real workspace with Claude Code — editor, terminal, Preview. What you learn is how someone steers it: what they ask for, what they check, what they refuse.
instructionagent_editapprovalcommandchecks_runBuilt to be fair to both sides
Three rules the product won't bend, for Reviewers and Candidates alike.
No global score
Five Dimensions, each on its own. No totals, averages, rankings or colour grades to anchor on.
Evidence for every claim
Every Evaluator statement cites the Replay events behind it, one click away.
People decide
Only a Reviewer sets a Decision. Your disagreements sit next to the Evaluator’s scores, never over them.
Every feature, on every plan
Plans grow with the Sessions you run, not the size of your team.
- Full Catalog
- Unlimited Reviewers
- Replay
- Evaluator scoring
- Compare
- Final diff
- EN + ES lobby
Questions
Is there an overall score?
No. Result is the Hidden-check count, and Direction, Verification, Efficiency and Technical judgement are each scored 1–4 on their own. We never sum, average or rank them — comparing engineers is a judgement call, and it stays yours.
What exactly do you record?
Instructions, Approvals, edits, commands, focus loss and large pastes. No camera, no screen recording, no keystrokes. Focus loss and pastes are shown as neutral signals, never warnings.
Can the Agent see the Brief or the Hidden checks?
No. The Candidate reads the Brief and decides what to tell the Agent. Hidden checks run on the platform; Candidates see only how many pass.
Who sets the Decision?
Only a Reviewer. The Evaluator scores Dimensions and cites Replay events, but it never suggests Advance or Don’t advance.
What is a Credit?
One Credit is one Session. Sending an Invite reserves a Credit; revoking the Invite or letting it expire unused releases it.
What if a Candidate disconnects?
They can reconnect from the same link and pick up where they left off. The clock never pauses, and the lobby says so before they start.
Can I compare Candidates on different versions of an Assessment?
Comparisons stay within one Assessment version. If a Round mixes versions, we show a warning instead of lining them up.
Review the work, not the vibe
Pick an Assessment, send three Invites and read your first Replay this week.