# Wrong Design Intent Is Worse Than Never Conditioning: A Derangement-Control Diagnosis of Header Conditioning in CAD Program Completion

arXiv: https://arxiv.org/abs/2607.23191 · HTML: https://mlsanity.com/paper/2607.23191/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 26 references that are in the map.
- First posted on arXiv 25 Jul 2026.

## Builds on (most cited first, 24 of 24)

- [Classifier-Free Diffusion Guidance](https://mlsanity.com/paper/2207.12598/) (2022)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Qwen2.5-Coder Technical Report](https://mlsanity.com/paper/2409.12186/) (2024)
- [DeepCAD: A Deep Generative Network for Computer-Aided Design Models](https://mlsanity.com/paper/2105.09492/) (2021)
- [SkexGen: Autoregressive Generation of CAD Construction Sequences with Disentangled Codebooks](https://mlsanity.com/paper/2207.04632/) (2022)
- [Stay on topic with Classifier-Free Guidance](https://mlsanity.com/paper/2306.17806/) (2023)
- [Hierarchical Neural Coding for Controllable CAD Model Generation](https://mlsanity.com/paper/2307.00149/) (2023)
- [SolidGen: An Autoregressive Model for Direct B-rep Synthesis](https://mlsanity.com/paper/2203.13944/) (2022)
- [CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation](https://mlsanity.com/paper/2505.04481/) (2025)
- [CAD-Recode: Reverse Engineering CAD Code from Point Clouds](https://mlsanity.com/paper/2412.14042/) (2024)
- [Retrieval-Augmented Generation with Conflicting Evidence](https://mlsanity.com/paper/2504.13079/) (2025)
- [CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward](https://mlsanity.com/paper/2505.19713/) (2025)
- [Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Models](https://mlsanity.com/paper/2501.19054/) (2025)
- [Text-to-CadQuery: A New Paradigm for CAD Generation with Scalable Large Model Capabilities](https://mlsanity.com/paper/2505.06507/) (2025)
- [FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Models](https://mlsanity.com/paper/2411.05823/) (2024)
- [CADmium: Fine-Tuning Code Language Models for Text-Driven Sequential CAD Design](https://mlsanity.com/paper/2507.09792/) (2025)
- [CodeCrash: Exposing LLM Fragility to Misleading Natural Language in Code Reasoning](https://mlsanity.com/paper/2504.14119/) (2025)
- [Text-to-CAD Evaluation with CADTests](https://mlsanity.com/paper/2605.07807/) (2026)
- [Aligning Constraint Generation with Design Intent in Parametric CAD](https://mlsanity.com/paper/2504.13178/) (2025)
- [Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generation](https://mlsanity.com/paper/2602.03045/) (2026)
- [TOOLCAD: Exploring Tool-Using Large Language Models in Text-to-CAD Generation with Reinforcement Learning](https://mlsanity.com/paper/2604.07960/) (2026)
- [How does Misinformation Affect Large Language Model Behaviors and Preferences?](https://mlsanity.com/paper/2505.21608/) (2025)
- [On Code-Induced Reasoning in LLMs](https://mlsanity.com/paper/2509.21499/) (2025)
- [No-Worse Context-Aware Decoding: Preventing Neutral Regression in Context-Conditioned Generation](https://mlsanity.com/paper/2604.16686/) (2026)

## Abstract (arXiv)

Fine-tuned code LLMs are routinely conditioned on a design-intent specification, but the correctness axis of such a signal -- a wrong intent rather than an absent one -- has not been tested, and the benefit of conditioning is usually scored with the same detector that defines the signal. We study CADCON, a five-feature design-intent header prepended to CadQuery-style programs during LoRA fine-tuning of Qwen2.5-Coder-1.5B, scoring adherence with executable geometric assertions that share no code with the header-defining extractor. On a pre-registered sample of 400 deduplicated held-out programs stratified over eleven intent profiles, at 40% prefix and three seeds, a semantically wrong header degrades adherence below the never-header-trained baseline on 3/3 seeds under both tokenizations at the program level, and on 3/3 token and 2/3 text seeds at the 298 distinct model inputs they present. Wrong-header executability is not depressed relative to that baseline. A derangement control, retrained so every program receives another program's header -- holding the header marginal fixed while destroying its correlation with the program -- saw the same programs, indices and wrong headers. Its correct-to-wrong change is -0.006/+0.016/-0.003 against 0.124/0.241/0.230 for the standard model, and the interaction is significant on 3/3 seeds (p <= 5.9e-7), so the model's sensitivity to whether the header is right or wrong requires the learned mapping. The control sits below the baseline by the same margin under a correct as under a wrong header, so we claim that sensitivity and not the below-baseline level. On features the true intent lacks, the standard model realizes a feature far more often when the wrong header names it; the control does not. Ground truth itself scores only 0.567 here, the scale on which arm levels should be read. Wrong design intent is not inert: it actively misdirects generation.