The Object the Field Keeps Reaching For

The map essay. Where the two foundational companions argue the criterion at a single layer — context selection at inference — this one generalizes the recognition across the whole AI pipeline. One operation recurs at a dozen points: context selection at inference, KV-cache eviction, pretraining-data curation, RAG reranking, chain-of-thought compression, distillation, quantization, memory consolidation, active learning. Each is handled by separate teams as a separate problem. In every case the far end of the channel is fixed and the budget is finite, and the thing being chosen is what enters that channel. The claim is specific and current: at this moment, in the most recent literature on each layer, the field is independently converging on the need for a decision-theoretic selection criterion and reaching for adjacent-but-degenerate objects — the information bottleneck, causal sufficiency, influence functions, KL heuristics — because the correct object (Le Cam deficiency together with the Vereshchagin–Vitányi structure function relative to a fixed decoder) has never been named in this setting. The convergence is the evidence; the misreach is the gap. KV-cache eviction is the showpiece: an information objective correctly identified, the information bottleneck reached for, and artificial noise injected to repair a degeneracy the correct object never incurs. The essay does not re-derive the foundation; it walks the layers and locates, in each, the adjacent object the field grabbed instead.