Design content moderation
Hard45 minFree, no account
The errors are asymmetric, the policy changes weekly, and humans are part of the architecture.
The question
Design moderation for a platform taking a million posts an hour across text, image and video.
Say where humans sit and how the system changes when policy changes.
Functional
- Classify against several policies: violence, harassment, nudity, self-harm, spam.
- Act automatically on high confidence; queue the rest for review.
- Appeals, and feeding the outcome back into training.
Non-functional
- 1M posts/hour; the worst categories need action in seconds.
- Policy definitions change frequently.
- Decisions must be explainable to a reviewer and to the user.
45:00Commit to an answer before you open the solution. Reading it first teaches you to recognise good answers, which is not the skill being tested.
Stuck?
0 of 3 hints takenThe worked solution
written by a person · not a gradeScore yourself
0 of 5 marked- A cost-ordered cascade rather than one model25
- Framed the system as allocating scarce human review25
- Per-policy thresholds justified by asymmetric harm20
- Versioned policy and labels, with a rules layer for speed20
- Fed appeals back as training data10
We run no AI here and nothing on this page grades you. The score is yours, and the useful number is the one you get on the same problem a month from now, cold.
kept in this browser only