Knowledge editing modifies model parameters so that a requested fact is updated while unrelated behavior stays unchanged. Editing is not only a write problem but also an address problem: deciding which hidden states should receive the new residual. An update that activates too narrowly memorizes a single prompt, while one that activates too broadly disrupts neighboring knowledge. Parametric editors encode this scope implicitly, whereas memory‑based editors make the selection explicit but keep it outside the edited model. We propose ALOE (Addressed Low‑rank Operator for Editing), which learns semantic addresses from paraphrases and hard same‑subject negatives, aligns them with autoregressive hidden states through rollout refinement and gate calibration, and embeds the resulting gated low‑rank operator within a single MLP layer. The deployed model thus runs in one forward pass with no external retriever or auxiliary router. Evaluations on CounterFact, ZSRE, and KnowEdit across three 7–8B model families show efficacy between 0.955 and 0.999 and locality between 0.981 and 1.000. Mechanistic analyses confirm that the learned geometry separates competing edits and that calibration suppresses out‑of‑scope activation. Remaining errors concentrate on paraphrase coverage and write fitting.
Review