Challenge library
Each challenge is the job in miniature: a messy brief, real constraints, and a genuine trade-off in every part. Scored on decision density rather than output volume, which is why 90 minutes is enough.
60–90 min. Take-homes that eat a weekend select for free time.
Full criteria visible before you start.
Openly. Using it well is part of the skill.
Do it once; the proof counts everywhere.
Design an agent that processes refund requests — then prove you know where it will fail and what stops it.
A support bot is giving wrong answers with total confidence. Find out why — the answer is not in the prompt.
You have capacity for two of five things. Decide, and tell the person who loses.
One vague request, one Slack thread of contradictions. Produce a spec a team can build without you.
There is an obvious optimisation available. It is not the constraint. Find the one that is.
An intermittent production failure with an error message that points at the wrong thing.
Questions people ask
- How long does a challenge take?
- 60 to 90 minutes, stated before you start. Going over is recorded but never penalised — the cap is a promise about our demands on your time, not a trap.
- Can I use AI tools?
- Yes, openly. For these roles, using AI well is part of the job. The rubrics target judgment an assistant will not supply for you: what you refuse to do, what you would cut, and how you would verify you were right.
- Do I see the rubric before I start?
- Always, in full, on the public challenge page. Hidden criteria produce anxiety, not signal.
- Can I retake a challenge?
- Yes. You see your result before deciding whether to publish it, and a weak result stays private.