"System One" decision models such as TypeSafe's Jev and the open‑source Laya can answer typed questions about a text in a single forward pass with calibrated probabilities, but they cannot request missing information: when the first message does not state what separates two departments, they merely guess. LAVOIR (Laya with Value‑Of‑Information Routing) places candidate missing pieces (slots) next to answer options in the input, so a single forward pass returns both the decision distribution and, for each slot, the expected gain in the probability of the correct decision if the user were asked about it. VOI targets require no human labels: gold decisions come from schema rules, an LLM verbalizes messages and answers, a model from another family checks every text, and pairing each message with several profiles enables regression on realized gains to estimate the expected gain. A Gini‑impurity cap bounds the predicted value by what a calibrated model can still gain. In a controlled study, decisions on seen schemas are statistically indistinguishable from the Bayes ceiling; the final model's question policy matches a greedy oracle VOI policy on seen schemas (AUC 0.799 vs. 0.797), and with at most 0.5 questions per conversation it is 14.1 points more accurate than never asking. On real ABCD conversations, one real exchange raises accuracy by 8.3 points where LAVOIR asks and leaves it unchanged where it does not; on SGD the cap lowers the asking rate from 93% to 8.6%. On Laya's twelve benchmarks LAVOIR exceeds Laya's reported scores on seven, and it answers a question in 31 ms (median, GH200).
Review