What a Context-Blind Proposal Costs: Structured Operators versus Selection in Event-Announcement Forecasting

Text-conditioned forecasting predicts a series from its numerical history together with natural-language context describing events the history cannot reveal. The dominant training-free recipe keeps a time-series foundation model as the numerical engine and lets a language model select among sampled trajectories. We show this carries a cost whenever the proposal is context-blind: a selector can only return a trajectory the proposal happened to draw, and the budget needed before an announced deviation appears grows exponentially in how surprising that deviation is—a statement about sample complexity, not impossibility. On the Context-is-Key subset whose context announces a dated, signed, magnitude-specified deviation ($13$ families, $195$ instances), an oracle selector over $256$ candidates reaches only $0.754$ of the context-blind error, while on the same samples a five-parameter operator reaches $0.285$, a richer edit schema $0.084$, and a frozen $14$B model that writes that schema zero-shot—never touching the ground truth—reaches $0.224$, beating the selection oracle on $169/195$ instances. Two oracle-guided sequential proposals do no better ($0.782$, $0.884$). The regime is narrow: across all $71$ families the operator ceiling no longer dominates ($0.382$ against $0.471$), and a surprise statistic computed without consulting either branch classifies Time-MMD as out of scope, where the ordering reverses. We have no forecasting-time scope detector; three attempts fail. The contribution is an action-space analysis with a characterised regime, not a general text-conditioned forecaster.