NumericJev: Jev-like LLM Numerical Decoding with Multiway Decision Trees
Abstract
Large language models can interpret natural lan- guage, yet robust decisions remain challenging. Jev-like models expose structured choices, but these interfaces do not directly provide numeri- cal values at a requested precision. We propose NUMERICJEV, a training-free numerical decod- ing algorithm that enables numerical output from any LLM with a Jev-like structured-choice in- terface. Surprisingly, on our arithmetic bench- mark, it outperforms direct selection from a can- didate list containing the correct answer by 2.93 percentage points (Figure 1). Our motivation comes from the observation that numerical range selection is itself a decision problem that Jev- like LLMs can address. NUMERICJEV recur- sively refines a range through a multiway deci- sion tree while retaining the original question in context, without parameter updates or hidden- state access. On a 100-value grid, a ten-way tree requires only two decision rounds. Range- normalized MAE is 1.84% versus 5.18% for di- rect choice. A separate three-date historical- index study yields 4.58% mean relative recall er- ror and 0% readout error when the value is sup- plied. Code is available at https://github. com/Bring-AI/jev-numeric.
Related
- Decision trees with optimal joint partitioning
- Decision trees, Frobenius traces, and Weierstrass coefficients of elliptic curves
- Optimal Multi-way Decision Trees for Stratified Sampling in Online Controlled Experiments
- On Stopping Sets and Continuous-Time Decision Trees
- Multiparameter counting of numerical semigroups: recurrences and leaf-discriminating trees
BibTeX
@misc{ye2026numericjev,
title = {{NumericJev: Jev-like LLM Numerical Decoding with Multiway Decision Trees}},
author = {Weiwei Ye and Hangchen Liu and Renhe Jiang},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.28587},
eprint = {2609.28587},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.28587},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - NumericJev: Jev-like LLM Numerical Decoding with Multiway Decision Trees
AU - Weiwei Ye
AU - Hangchen Liu
AU - Renhe Jiang
PY - 2026
DA - 2026/09/23
PB - arXiv
DO - 10.48550/arxiv.2609.28587
UR - https://arxiv.org/abs/2609.28587
AB - Large language models can interpret natural lan- guage, yet robust decisions remain challenging. Jev-like models expose structured choices, but these interfaces do not directly provide numeri- cal values at a requested precision. We propose NUMERICJEV, a training-free numerical decod- ing algorithm that enables numerical output from any LLM with a Jev-like structured-choice in- terface. Surprisingly, on our arithmetic bench- mark, it outperforms direct selection from a can- didate list containing the correct answer by 2.93 percentage points (Figure 1). Our motivation comes from the observation that numerical range selection is itself a decision problem that Jev- like LLMs can address. NUMERICJEV recur- sively refines a range through a multiway deci- sion tree while retaining the original question in context, without parameter updates or hidden- state access. On a 100-value grid, a ten-way tree requires only two decision rounds. Range- normalized MAE is 1.84% versus 5.18% for di- rect choice. A separate three-date historical- index study yields 4.58% mean relative recall er- ror and 0% readout error when the value is sup- plied. Code is available at https://github. com/Bring-AI/jev-numeric.
ER -