Judgment =
• choosing the user + job-to-be-done
• defining the outcome (what changes for them?)
• cutting scope to the smallest shippable proof
• picking a success signal (activation, repeat use, $)
• holding the quality bar (trust, errors, accessibility)