All six dimensions
AI collaboration
Every engineer now works alongside a model. This dimension scores how well, across five separate sub-tests, from the prompt they write to the code they accept, reject, and rewrite on top.
What a candidate is asked to do
- 01Write the prompt you would actually send, given a feature spec.
- 02Read AI-written code that runs, and say what you would change.
- 03Find and patch exactly one planted bug, without refactoring around it.
- 04Critique code containing several planted flaws, and catch all of them.
- 05Work a final question with an AI sidebar open, with every exchange recorded.
How it is scored
What earns points
- Prompts that carry context, constraints, edge cases, and acceptance criteria
- Naming the existing pattern or file the work should follow
- Rejecting a plausible suggestion and saying why
- Minimal, targeted fixes
What loses points
- Verbosity mistaken for detail
- Accepting the first answer without reading it
- Broad refactors that miss the actual defect
- Stopping at the first flaw and calling the rest fine
Why it predicts the job
The bottom quartile of every team takes the model output as given. The top quartile catches the invented import and the over-engineered class. Two engineers of the same nominal seniority can differ sharply here, and nothing else on a CV tells you which one you are hiring.