Design the moderation system that inspects user-generated posts, images, and comments — routing them through automated classifiers and, when needed, a prioritized human-review queue — and enforces a decision of allow, limit, or remove.

Operating context. Every new piece of content is checked. Most are auto-resolved by classifiers; a minority need a human. A finite pool of reviewers works a queue that must be prioritized by severity and by reach, so a high-severity item on a viral post reaches a reviewer fast while low-priority items still get seen eventually and are never starved forever. Decisions must propagate to the content service to enforce visibility, and every decision must be auditable and reversible on appeal.

Out of scope. Training the machine-learning classifiers, the appeals adjudication interface, legal and regulatory reporting exports, and account-level banning policy. Assume other teams own those.

What to produce. A high-level design covering: what runs synchronously before content goes live versus asynchronously after, how automated classifiers are orchestrated and confident cases auto-resolved, the priority queue that feeds human reviewers, how a final decision propagates and is enforced, and the immutable audit trail. Sketch the components and the flow from submission to enforced decision; checkpoints will probe the priority queue and audit design.