ChartRevise
ChartRevise: A Dataset and Evaluation Protocol for Exact Chart Editing via Code
A dataset of 92,438 records covering exact chart editing operations across code generation tasks.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info chartrevise
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
5 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| qwen3.5-9B | 94.3 | N/A | Independent | medium | Sep 29, 2026 | arxiv.org |
| Qwen3.5 4B | 93.8 | N/A | Independent | medium | Sep 29, 2026 | arxiv.org |
| Granite | 92.3 | N/A | Independent | medium | Sep 29, 2026 | arxiv.org |
| Phi-3.5-V | 91.7 | N/A | Independent | medium | Sep 29, 2026 | arxiv.org |
| InternVL2.5 | 90.3 | N/A | Independent | medium | Sep 29, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
Content & Construction: The benchmark introduces a structured dataset derived from the grammar of graphics to systematically cover chart-editing operations. It includes source-program checks and atomic audits for completion, coupled updates, and preservation.
Scoring & Findings: A reference-free evaluation protocol measures requirement recall, exact-edit rate, and gratuitous changes. Fine-tuning on this data yields relative gains in mean requirement recall (16%) and mean exact-edit rate (22%).