TraceCAD: Trace-Guided Repair for Agentic CAD Generation
Quick Answer
TraceCAD introduces a recovery layer for LLM-based CAD agents, enhancing repair reliability and geometric quality on DeepCAD benchmarks.
Quick Take
It maintains persistent state, achieving competitive results with reduced retries and costs, demonstrating significant improvements in CAD generation.
Key Points
- TraceCAD links features, modeling steps, and failure evidence as persistent state.
- Achieves competitive geometric quality with improved IoU, Chamfer, and Hausdorff distances.
- Removing persistent state nearly halves the recovery score.
- Localized search more than doubles geometric regression and code-agent invocations.
- Initializing skill store on disjoint models reduces retries, token costs, and latency.
DeepSignal Analysis
What happened
TraceCAD introduces a recovery layer for LLM-based CAD agents, which enhances the reliability of repairs and the geometric quality of outputs. The system maintains persistent state and demonstrates competitive results on DeepCAD benchmarks, achieving improvements in CAD generation efficiency.
Key evidence
- TraceCAD links requested features, modeling steps, failure evidence, and candidate outcomes as persistent state, which aids in diagnosing faulty operations.
- On DeepCAD-derived benchmarks, TraceCAD achieves competitive geometric quality, measured by metrics such as IoU, Chamfer distance, and Hausdorff distance.
- Removing persistent state reduces recovery scores by nearly half, while eliminating localized search more than doubles geometric regression and code-agent invocations.
Why it matters
The introduction of TraceCAD could significantly impact the efficiency of CAD generation by improving the reliability of repairs. This is particularly relevant for industries relying on precise CAD outputs, as enhanced geometric quality can lead to better design outcomes and reduced costs associated with retries and errors.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract:LLM-based CAD agents produce executable parametric programs, but their correction loops may lose evidence about satisfied requirements, faulty operations, and prior repairs. We introduce TraceCAD, a recovery layer that links requested features, modeling steps, failure evidence, and candidate outcomes as persistent state. TraceCAD diagnoses likely faulty operations, searches bounded edits in their dependency regions, validates candidates through execution and preservation checks, and retains successful and failed repair outcomes in reusable skill memory. On DeepCAD-derived benchmarks with 200-model ablations and a 1K-model comparison, TraceCAD achieves competitive geometric quality in terms of IoU, Chamfer distance, and Hausdorff distance. Removing persistent state nearly halves recovery score; removing localized search more than doubles geometric regression and doubles code-agent invocations. Initializing the skill store on disjoint training models further reduces retries, token cost, and latency. These results demonstrate that persistent, localized, and reusable recovery improves final CAD quality and repair reliability.
| Comments: | 18 pages, 7 figures; includes supplementary material |
| Subjects: | Artificial Intelligence (cs.AI); Graphics (cs.GR); Software Engineering (cs.SE) |
| Cite as: | arXiv:2608.03062 [cs.AI] |
| (or arXiv:2608.03062v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2608.03062 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Fan Fengxiao [view email]
[v1]
Tue, 4 Aug 2026 03:23:11 UTC (5,658 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →HOBA: Hierarchical On-Policy Bidding Agents for Adaptive Online Advertising
HOBA (Hierarchical On-policy Bidding Agents) is a novel hierarchical reinforcement learning framework that enhances online advertising bidding systems by improving adaptability and reducing hyperparameter tuning costs. It utilizes a for hyperparameter inference, a SARSA agent for expert model selection, and a dynamic expert pool for bid execution, achieving a +3.6% increase in target cost during large-scale deployment and outperforming state-of-the-art baselines on AuctionNet.