Before a tag is offered as busy elsewhere, a sample of its posts is counted by author server, and a tag carried by too few of them is dropped as a bot wave. That count never asks whether the operator has blocked the servers those authors live on — only the server it queried is checked. So an instance the operator has shut out still supplies a distinct author server, still counts towards the diversity that clears the gate, and still enters the share that decides whether the traffic is bots. The operator notices nothing; they blocked an instance and it keeps voting.
A blocked instance should not count as evidence about a tag, the same way it does not appear as a post. The shape of the fix is the one the pull already uses; what it needs first is a measurement in both directions over real timelines, because tightening it drops tags as well as bot waves.
Where: authors/2 in Vutuv.Tags.ExternalTagClient, beside the check parse/2 already does.
An AI agent wrote this text in my name. I know that is problematic.
Before a tag is offered as busy elsewhere, a sample of its posts is counted by author server, and a tag carried by too few of them is dropped as a bot wave. That count never asks whether the operator has blocked the servers those authors live on — only the server it queried is checked. So an instance the operator has shut out still supplies a distinct author server, still counts towards the diversity that clears the gate, and still enters the share that decides whether the traffic is bots. The operator notices nothing; they blocked an instance and it keeps voting.
A blocked instance should not count as evidence about a tag, the same way it does not appear as a post. The shape of the fix is the one the pull already uses; what it needs first is a measurement in both directions over real timelines, because tightening it drops tags as well as bot waves.
Where:
authors/2inVutuv.Tags.ExternalTagClient, beside the checkparse/2already does.An AI agent wrote this text in my name. I know that is problematic.