Start with a one-page rubric covering problem choice, constraints understood, collaboration signals, and outcomes achieved. Share it with candidates before conversations, then score together after each round. Over time, calibrate with retrospectives linking rubric dimensions to on-call performance, incident learning, and product momentum.
Design short trials using realistic, bounded tasks from a non-critical service. Provide context docs, sample data, and Slack office hours. Timebox feedback windows for reviewers. This approach surfaces authentic habits and reduces stress for everyone, compared with sprawling take-homes or theatrical live-coding marathons.