pipette
ESEspañol

Answer Delivery Under a Shared Token Budget: An Audited Study of Stopping and Finalization

Yupeng Chang, Linyuan Gao, Zhiyuan Li, Yuan Wu, Yi Chang

Preprint

In the authors' words

An inference budget is useful only when the system delivers an answer within it. We study a fixed-budget policy combining first-valid-box stopping with a charged finalization cue. On 100 fresh, independently audited GSM8K test questions per model, it improves delivered endpoint accuracy over first-box stopping by 8.0 percentage points for Qwen3-4B-Thinking-2507 and 7.0 for DeepSeek-R1-Distill-Llama-8B. A prelocked repeat on the same questions gives 7.0 and 10.0 points; neither seed shows a gross harm relative to first-box stopping. The point estimates of incremental physical-slot savings are below one percent. A finalization-only ablation separates cue recovery from stopping, and a fixed scalar reader tests extraction as an alternative explanation. The cue is prior art; the contribution is a component and readout analysis with charged injections and verified paths. Results cover two checkpoints, one task, one prompt family, and two fixed seed conditions.

Main resultLimitation the authors admit

Appeared: Tuesday, October 20. arXiv. Preprint, not yet peer-reviewed.