HOW IT WORKS

Two phases. One token budget. Real measurement.

Every candidate works through the same structured assessment. Same AI. Same fixed token budget. We measure who builds the most with the least — no pass/fail, just progress comparison.

01 — THE PHASES

From logic to building with AI.

Each phase tests a different skill that modern developers need. The whole session tells the story.

01LOGIC LIVE
No AI. Candidate writes code alone. Proves they can think through problems, design systems, and write working code without assistance. The foundation everything else builds on.
AI OFFEDITOR EDITABLE
02AI BUILD LIVE
AI enabled. Editor read-only. Fixed token budget visible in real time. The candidate must direct AI to build the right thing through clear, specific prompts. Vague in, vague out.
AI ONEDITOR READ-ONLY
02 — FULL FLOW

Four phases. One score.

The full assessment includes four phases. Debug and Review are live and included in every assessment.

LIVE
03DEBUG
We inject 3-5 hidden bugs into the AI-generated codebase — security vulnerabilities, logic errors, edge case failures, performance issues. The candidate must find and fix them. Tests AI judgment: can you spot when AI is wrong?
AI ONEDITOR EDITABLE
LIVE
04REVIEW
Editor locked to read-only. We ask 5-10 questions about the codebase: architecture decisions, edge cases, potential improvements, security concerns. Multiple choice + short answer. Tests comprehension, not just construction.
AI OFFEDITOR READ-ONLY
03 — WHAT HAPPENS IN A SESSION

Every interaction is logged.

For companies, every prompt, every AI response, every code edit, and every test run is captured in a structured replay. You don't just see the final score — you see how they think.

01
START
Candidate receives a challenge brief. The token budget is fixed and visible. Time limit is set. The workspace opens with a full IDE, AI panel, and terminal.
02
BUILD
Candidate prompts AI to generate code. Editor is read-only. They iterate through prompts, testing each change. Token count drops in real time. Terminal shows test results.
03
SCORE
AI evaluates the session across two signals: Understanding and Token Efficiency. Every claim is backed by evidence from the chat log.
04 — THE AI IS RESTRICTIVE BY DESIGN

~30% restrictive. That's the point.

The AI mirrors candidate quality. Vague prompt → minimal stub. Precise prompt → precise code. The candidate has to notice what's missing and request it.

ON A VAGUE REQUEST
Like "build a booking API" — the AI builds a minimal working version: basic endpoints, no validation, no conflict detection, no error handling. The AI never adds features the candidate did not ask for.
WHAT IT NEVER DOES
It does not coach. It does not suggest next steps. It does not warn about problems. The candidate has to notice what is missing, frame a follow-up prompt, and direct the AI precisely.

Ready to see it in action?

Create an assessment in minutes. Send one link. Get real measurement.

FOR COMPANIES →PRACTICE FREE ■