Why it is a question
A21 §2: q_t → Z_t → b_{t+1} → q_{t+1}. “An early query selects particular evidence. That evidence influences the agent’s interpretation. That interpretation determines the next query.” The owner’s point, which A21 granted: “A lot of what gets called agentic search, context engineering, or memory orchestration is a new implementation of older problems.” What is new, if anything, is that the agent “can change not only the query terms but its task interpretation, its stopping criterion, and even its definition of relevant evidence.”
It refines QUESTION-109 from one retrieval to a sequence of them. CLAIM-301 is the thesis and CLAIM-292 the granted limit. A21 §7 proposed the design: fixed query, adaptive single trajectory, independent trajectories, and “intent-anchored” adaptive search, at matched budgets, scored on coverage, calibration and “the rate at which conjectures become unstated premises.”
The exchange was itself a case. A2 framed the record as “an epistemic record”; the framing set what was searched for next, and by A12 and A13 the questions were about the record’s software. The owner named it at U14 (workbench entry of 2026-10-15).