Real work (a feature, an integration, an assessment), clear acceptance criteria, production-adjacent standards — toy problems measure toy behavior.
Question quality in week one (good teams interrogate scope), demo behavior (working software or slideware?), how they handle the inevitable ambiguity, and the code review your team performs on the output.
Paid (free trials select for desperation, and you want IP assignment clean), time-boxed, with an explicit evaluation conversation scheduled at the end.
Handover quality if you don't proceed — documentation, repo transfer, graciousness — is the strongest predictor of what year three would have looked like.
Skipping the discipline this article describes until an incident, audit, or stalled project forces it — every practice above is cheaper adopted early than retrofitted under pressure.
Let's discuss how we can help you with trial projects derisking.