# Correct Choices, Failed Sequences: Choice2Seq Benchmarks Correctness Retention Across Output Interfaces

arXiv: https://arxiv.org/abs/2610.16191 · HTML: https://mlsanity.com/paper/2610.16191/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 39 references that are in the map.
- First posted on arXiv 12 Oct 2026.

## Builds on (most cited first, 19 of 19)

- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [Large Language Models Are Not Robust Multiple Choice Selectors](https://mlsanity.com/paper/2309.03882/) (2023)
- [How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering](https://mlsanity.com/paper/2012.00955/) (2020)
- [Towards Scalable Multi-domain Conversational Agents: The Schema-Guided Dialogue Dataset](https://mlsanity.com/paper/1909.05855/) (2019)
- [Why Exposure Bias Matters: An Imitation Learning Perspective of Error Accumulation in Language Generation](https://mlsanity.com/paper/2204.01171/) (2022)
- [Let Me Speak Freely? A Study on the Impact of Format Restrictions on Performance of Large Language Models](https://mlsanity.com/paper/2408.02442/) (2024)
- [JSONSchemaBench: A Rigorous Benchmark of Structured Outputs for Language Models](https://mlsanity.com/paper/2501.10868/) (2025)
- [Exposure Bias versus Self-Recovery: Are Distortions Really Incremental for Autoregressive Text Generation?](https://mlsanity.com/paper/1905.10617/) (2019)
- [CUTE: Measuring LLMs' Understanding of Their Tokens](https://mlsanity.com/paper/2409.15452/) (2024)
- [JEV-as-a-Judge: Accept When Confident, Escalate When Unsure](https://mlsanity.com/paper/2609.26550/) (2026)
- [Models In a Spelling Bee: Language Models Implicitly Learn the Character Composition of Tokens](https://mlsanity.com/paper/2108.11193/) (2021)
- [Type-Safe Is Not Error-Free: Typed Decision Models Follow the Option Name, Not the Definition Bound to It](https://mlsanity.com/paper/2609.26758/) (2026)
- [Same Scores, Different Decisions: Evaluating JEV and Language Models for Legal Document Understanding](https://mlsanity.com/paper/2609.27678/) (2026)
- [Beyond Calibration: Do a Typed-Decision Model's Probabilities Obey the Probability Axioms?](https://mlsanity.com/paper/2609.33209/) (2026)
- [Spelling-out is not Straightforward: LLMs' Capability of Tokenization from Token to Characters](https://mlsanity.com/paper/2506.10641/) (2025)
- [Contextual Metric Meta-Evaluation by Measuring Local Metric Accuracy](https://mlsanity.com/paper/2503.19828/) (2025)
- [Offline Preference-Based Trajectory Evaluation](https://mlsanity.com/paper/2606.17541/) (2026)
- [SecJev: Bringing Security Expertise to System One Decision Models](https://mlsanity.com/paper/2610.03073/) (2026)

## Abstract (arXiv)

A model may select the correct answer yet fail to construct it through a sequence of local decisions. We introduce Choice2Seq, a diagnostic benchmark that measures correctness retention across complete-answer selection and sequential construction in independent runs on the same problems and candidate sets. Sequence Composition Rate (SCR) measures the fraction of correctly selected items that are also answered correctly through local construction. Jev achieves 97.1% and 99.5% selection accuracy on arithmetic and lexical retrieval, yet retains only 33.0% and 2.1% of those correct choices under serial construction, respectively. Substantial gaps also emerge for StartLux and Qwen, including when Qwen emits characters directly. Restricting local options to candidate-supported characters improves construction, while explicit position indexing improves copying fidelity, even for incorrect selections. These interventions distinguish task correctness, plan fidelity, and output compliance. Matched dialogue-state evaluations reveal that local field construction largely preserves Jev's correct states but reduces Qwen's mean-turn accuracy, highlighting task- and interface-dependent differences. Sequence-level confidence and request-cost analyses further characterize completed outputs.