AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Correct but Unselectable: The Hidden Interface Tax in Multi-Candidate Reasoning

Qianli Ma, Zhiqing Tang, Fanshuai Meng, Zhi Yao, Hanshuai Cui, Weijia Jia

NeurIPS 2026 · Large language models › Reasoning and chain of thought · approach: Efficient and adaptive thinking · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Identifies an "interface tax" in multi-candidate generate-then-rerank for small language models: correct answers can appear in the candidate pool yet fail to be selected, so more candidates do not necessarily improve accuracy.

Problem

Small language models under cost, latency, or privacy constraints need test-time scaling, but generate-then-rerank with more candidates does not reliably raise accuracy because correct answers in the pool often fail to be selected.

Approach

Identifies a previously overlooked bottleneck between candidate generation and selection (the interface between them) where a correct candidate is present but unselectable, and frames this as a hidden tax on multi-candidate reasoning.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add CACC interface layer to generate-then-rerankQwen3-1.7B + fixed verifier · GSM8Kfinal accuracy: 0.2873 -> 0.3632Improved
add CACC interface layerQwen3-1.7B + fixed verifier · competition_math (numeric)final accuracy: 0.1519 -> 0.2579Improved
add CACC interface layerQwen3-1.7B + fixed verifier · GPQA Diamondfinal accuracy: 0.0556 -> 0.1919Improved
add CACC interface layerQwen3-1.7B + fixed verifier · MMLU-Profinal accuracy: 0.2362 -> 0.2277Hurt
add proposer-side strengthening (Qwen3-8B) to CACCCACC · GPQA Diamondfinal accuracy: 0.1919 -> 0.2576Improved
add proposer-side strengthening to CACCCACC · MMLU-Profinal accuracy: 0.2277 -> 0.2936Improved
add answer-mode alignment (benchmark-locked choice-letter readout) (ablation)CCR-only pool · MMLU-Profinal accuracy: 0.0096 -> 0.2362Improved
add candidate hygiene (SCH) on top of CCR+AMC (ablation)CACC · MMLU-Profinal accuracy: 0.2270 -> 0.2277No clear effect
add hygiene to CACC pool with second-family generator (ablation)CACC · MMLU-Pro confirm-512final accuracy: 0.1992 -> 0.2246Improved
add conversion-aware rescue after CACC+SPCACC+SP · MMLU-Profinal accuracy: 0.2936 -> 0.3320Improved
raise retained cap from 8 to 16 (ablation)CACC · GSM8Kfinal accuracy: 0.3632 -> 0.3298Hurt
sample 32 candidates instead of 4 (ablation)CACC · GSM8Kfinal accuracy: 0.3753 -> 0.2904Hurt

Surprising findings

Experimental setup

Metrics: accuracy

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy