Moderation queue

A moderation queue is the workspace where content that cannot be auto-decided waits for a human. When an automated layer is confident, it approves or blocks on its own. Everything in between, the flagged, the borderline, the genuinely ambiguous, lands in the queue for a moderator to arbitrate.

What it holds

The queue is not just a list of pending comments. It is the place where editorial judgement happens, so it has to surface enough context to decide in seconds rather than minutes.

• Toxicity score. A numeric signal from the model, usually with a category (hate, spam, off-topic, threat) so the moderator knows why the item was held.
• Model reasoning. A short explanation of what the classifier flagged, which lets the human agree, override or correct.
• Article context. The piece the comment responds to, because the same sentence can be fair under an opinion column and out of bounds under an obituary.
• User history. Past decisions on the same account, so a first-time slip is not treated like a repeat offender.

Best practices

A good queue is measured in throughput, not in features. Two things matter most.

• Keyboard shortcuts. Moderators should never reach for the mouse. Approve, reject and skip on single keys turn a review session into a rhythm.
• Multi-selection. Spam waves and brigading come in bursts. Selecting many items and acting on them at once keeps the backlog from growing faster than the team can clear it.

The goal is to spend human attention only where it adds value, and to make each decision traceable for later audit.

How Logora handles it

Logora routes the roughly 15% of contributions the model is unsure about into a human queue, while clean content auto-publishes and clearly toxic content is auto-blocked.

• Keyboard-first review. Shortcuts including Ctrl+A and Ctrl+S keep moderators in the flow.
• Six DSA rejection reasons. Each rejection is tagged with one of six structured grounds, so every decision feeds clean records for notice-and-action and transparency reporting.

Related concepts

• Content moderation, the practice the queue serves
• Human in the loop, why the ambiguous cases need a person
• Pre-moderation, when everything goes through the queue first
• Notice and action, the DSA process the rejection reasons feed