# How Output Format Confounds Data Quality and Capability in Instruction Tuning

arXiv: https://arxiv.org/abs/2609.02015 · HTML: https://mlsanity.com/paper/2609.02015/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 14 of its 28 references that are in the map.
- First posted on arXiv 2 Sep 2026.

## Builds on (most cited first, 14 of 14)

- [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)
- [Designing and Interpreting Probes with Control Tasks](https://mlsanity.com/paper/1909.03368/) (2019)
- [LESS: Selecting Influential Data for Targeted Instruction Tuning](https://mlsanity.com/paper/2402.04333/) (2024)
- [Null It Out: Guarding Protected Attributes by Iterative Nullspace Projection](https://mlsanity.com/paper/2004.07667/) (2020)
- [Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style](https://mlsanity.com/paper/2106.04619/) (2021)
- [Measurement and Fairness](https://mlsanity.com/paper/1912.05511/) (2019)
- [LEACE: Perfect linear concept erasure in closed form](https://mlsanity.com/paper/2306.03819/) (2023)
- [The Incomplete Rosetta Stone Problem: Identifiability Results for Multi-View Nonlinear ICA](https://mlsanity.com/paper/1905.06642/) (2019)
- [Two-stage LLM Fine-tuning with Less Specialization and More Generalization](https://mlsanity.com/paper/2211.00635/) (2022)
- [LLMs Are Biased Towards Output Formats! Systematically Evaluating and Mitigating Output Format Bias of LLMs](https://mlsanity.com/paper/2408.08656/) (2024)
- [How Instruction and Reasoning Data shape Post-Training: Data Quality through the Lens of Layer-wise Gradients](https://mlsanity.com/paper/2504.10766/) (2025)
- [GIST: Targeted Data Selection for Instruction Tuning via Coupled Optimization Geometry](https://mlsanity.com/paper/2602.18584/) (2026)
- [Beyond Similarity: A Gradient-based Graph Method for Instruction Tuning Data Selection](https://mlsanity.com/paper/2502.11062/) (2025)
- [GAST: Gradient-aligned Sparse Tuning of Large Language Models with Data-layer Selection](https://mlsanity.com/paper/2603.09865/) (2026)

## Abstract (arXiv)

Instruction-tuning data are judged by quality metrics, and tuned models are judged by benchmarks, but both judgments pass through an output interface: the surface format in which an answer is written. Using gradient signatures across 12 tasks, four semantically equivalent interfaces, three model families, and controlled corruptions, we show that this interface confounds both measurements. Spectral statistics such as effective rank are provably invariant to interface rotation and empirically blind to semantic corruption, while the direction of the update carries the quality signal. The interface-varying residual is not noise: it identifies each unit's own target task perfectly across all three families. Capability itself is stored relative to the training interface: a skill that raises accuracy by more than 40 points under the training format can be nearly invisible under every other, and correcting a single generation budget flips the measured effect of fine-tuning on GSM8K from a gain into a large loss. Pre-registered interventions delimit where this geometry stops short of control. Data quality and model capability are interface-conditioned quantities, and current practice often reports the interface instead of the content.