{"data":{"kind":"file","path":"README.md","version_id":"jbbfiqnz6f2ve6ygp2j1qc18","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1636,"modified_at":"2026-05-11T13:40:07.469000","content_hash":"6708288575f9bac0345d4a3e2ea9d6351ad5acbd046d8347fe2a39e33fa43d96"},"entries":[],"content":"# Data Visualization Critique\n\nA Prime Intellect Verifiers environment for evaluating LLM capabilities on identifying and critiquing issues in data visualizations.\n\n## Overview\n\nThis environment tests the model's ability to analyze chart/graph descriptions and identify common visualization problems:\n\n- **Misleading Axes** — truncated baselines, non-uniform scales\n- **Cherry-Picked Data** — selective time ranges, biased samples\n- **Wrong Chart Type** — pie charts with too many segments, 3D charts\n- **Missing Labels** — absent units, legends, annotations\n- **Color Issues** — accessibility, rainbow palettes, similar shades\n- **Dual Axes** — scale manipulation, false correlations\n- **Statistical Misrepresentation** — correlation vs causation, hindsight bias\n- **Animation/Interactivity** — speed issues, lack of controls\n- **Area/Scale Bias** — geographic area bias, outlier compression\n\n## Scoring\n\nThe rubric combines three reward functions:\n\n1. **Keyword Matching** (25%) — required analysis terms in response\n2. **Issue Identification** (40%) — correctly identifies the type of visualization issue\n3. **Structured Critique Quality** (35%) — checks for issue identification, explanation, and suggestion components\n\n## System Prompt\n\nThe system prompt instructs the model to:\n- List specific problems with the visualization\n- Explain why each issue is problematic\n- Provide concrete recommendations for improvement\n\n## Usage\n\n```bash\nprime env install data-viz-critique\nprime eval run data-viz-critique -m <model>\n```\n\n## Tags\n\n`data-visualization`, `critique`, `analysis`, `single-turn`, `data-quality`\n","encoding":"utf-8","truncated":false,"total_bytes":1636},"status":null}