Spaces:
No application file
No application file
| title: VisualTraceBench Demo | |
| emoji: 📊 | |
| colorFrom: blue | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: "4.44.0" | |
| app_file: app.py | |
| pinned: false | |
| license: cc-by-4.0 | |
| tags: | |
| - sql | |
| - performance | |
| - benchmark | |
| - visualization | |
| - debugging | |
| - enterprise-software | |
| datasets: | |
| - viperoptic/VisualTraceBench | |
| # VisualTraceBench: SQL Performance Regression Debugging Benchmark | |
| An interactive demo for the VisualTraceBench dataset — 192 production SQL performance regression bugs from a large enterprise column-store database system (2016–2025). | |
| ## Key Finding | |
| **VPER = 27.9%** — Visual execution plan traces improve AI diagnostic accuracy by 27.9% over tabular query statistics alone (p < 0.001). | |
| ## Dataset | |
| - 192 annotated production bug reports | |
| - Each paired with a PlanViz-style SQL execution plan visualization | |
| - Fully anonymized (k-anonymity ≥ 5, no customer data) | |
| 📦 [Dataset on Hugging Face](https://huggingface.co/datasets/viperoptic/VisualTraceBench) | |
| ## Paper | |
| *VisualTraceBench: A Human-Performance Benchmark for Multimodal AI SQL Plan Debugging with PlanViz Visual Traces* | |
| Under review at ICLR 2027 Datasets and Benchmarks Track. | |