Deletion Cost, Not Salience: Causal Measurement for Context Eviction in LLM Agents

Agents with finite context must decide what to retain. Common eviction rules use proxies for how much an element still matters: age, message type, attention mass, or a model's judgement. We measure deletion cost directly: perturb one element, replay the identical continuation under teacher forcing, and read the change in log-odds on the decisions it governs, with a placebo control at each distance. The survey's position-controlled power-law exponents lie within $\pm 0.177$, and deleting a standing constraint remains costly through $1153$ steps. In separate probe schedules, attention half-lives span $1.13\times$, while plan and constraint differ by $8.8\times$ in fitted causal half-life. Across eight open-weight models, all $48$ (model, probe-channel) cells remain on the relevant side of the fitted power-law boundary. Finally, scoring all $720$ static orders on a six-probe BF16 evaluation, we find that a frozen $4$B order based on directly measured deletion costs beats the evaluated six-text and fixed-prompt task-informed judges on two Qwen models. A third model's informed judge recovers the direct order, but too few trajectories pass its gate for a stable cross-guise comparison. Fitting supplies no resolved advantage over the direct order. These results provide a measurement for pricing context deletion; how often an element will still be needed remains a deployment input to be measured.