mlsanityData to 7 Oct 2026

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

arXiv 2607.23191 ↗First posted 25 Jul 2026cs.LG cs.GRcited by 0

Not on the map. Why?

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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 24 of its 26 references that are in the map. First posted on arXiv 25 Jul 2026.

Builds on

24 of 26 references.

References · 24

  1. 2022Classifier-Free Diffusion Guidancecited by 5,717
  2. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,355
  3. 2024Qwen2.5-Coder Technical Reportcited by 1,151
  4. 2021DeepCAD: A Deep Generative Network for Computer-Aided Design Modelscited by 266
  5. 2022SkexGen: Autoregressive Generation of CAD Construction Sequences with Disentangled Codebookscited by 121
  6. 2023Stay on topic with Classifier-Free Guidancecited by 90
  7. 2023Hierarchical Neural Coding for Controllable CAD Model Generationcited by 84
  8. 2022SolidGen: An Autoregressive Model for Direct B-rep Synthesiscited by 78
  9. 2025CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generationcited by 65
  10. 2024CAD-Recode: Reverse Engineering CAD Code from Point Cloudscited by 63
  11. 2025Retrieval-Augmented Generation with Conflicting Evidencecited by 60
  12. 2025CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Rewardcited by 53
12 more
  1. 2025Text-to-CAD Generation Through Infusing Visual Feedback in Large Language Modelscited by 43
  2. 2025Text-to-CadQuery: A New Paradigm for CAD Generation with Scalable Large Model Capabilitiescited by 43
  3. 2024FlexCAD: Unified and Versatile Controllable CAD Generation with Fine-tuned Large Language Modelscited by 33
  4. 2025CADmium: Fine-Tuning Code Language Models for Text-Driven Sequential CAD Designcited by 23
  5. 2025CodeCrash: Exposing LLM Fragility to Misleading Natural Language in Code Reasoningcited by 14
  6. 2026Text-to-CAD Evaluation with CADTestscited by 9
  7. 2025Aligning Constraint Generation with Design Intent in Parametric CADcited by 8
  8. 2026Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generationcited by 7
  9. 2026TOOLCAD: Exploring Tool-Using Large Language Models in Text-to-CAD Generation with Reinforcement Learningcited by 6
  10. 2025How does Misinformation Affect Large Language Model Behaviors and Preferences?cited by 5
  11. 2025On Code-Induced Reasoning in LLMscited by 5
  12. 2026No-Worse Context-Aware Decoding: Preventing Neutral Regression in Context-Conditioned Generationcited by 1

Built on by

Not cited yet.