Week 9: The queue at 2am

What the work does to the people who do it, why mod teams collapse in predictable ways, and how a rota is a piece of moral engineering

3 May 2027

Nine volunteers, 900 removals a day, no training, no screening, no employee assistance programme, and a queue that is worst at the hours when the fewest of them are awake. This is a normal mid-sized moderation team and it describes working conditions that would be illegal if anyone were being paid.

The failure modes are well documented and they arrive in an order. First the team over-relies on the one member with insomnia. Then that member burns out and leaves without a handover, taking the AutoModerator rationale with them. Then the remaining team, short-handed, tightens the rules to reduce volume, which increases appeals, which increases volume. Then somebody who enjoys the power more than the work becomes the most active moderator by default, because they are the only one still showing up.

That last step is how communities are captured, and it is not a conspiracy. It is an attrition gradient.

The counter-measures are unglamorous and they are the actual content of this week: caps on shift length, mandatory co-signature above a severity threshold, rotation off the worst queues, a written record of why each rule exists, and recruitment before you need it rather than after somebody quits.

Outline#

• the volume: what a mid-sized queue actually looks like per person per day
• documented harms in commercial moderation, and what transfers to volunteers
• the attrition gradient and how capture happens without anyone intending it
• rota design as moral engineering: caps, rotation, co-signature, review
• the question the course keeps returning to — should this work be paid, and what breaks if it is