Precedent Treatment - High-Level (Few-Shot): leaderboard
Negative precedent treatment classification (Validate Your Authority, 2026): given the passages of a later US court opinion that discuss a precedent, a model labels how the opinion treats it, on citing relationships annotated by Paul Hellyer. This board uses the five high-level treatment categories and a prompt with three worked examples; a prediction counts when it matches any expert-acceptable label. Accuracy in percent; higher is better.
Metric: Instance-based accuracy (%): share of the scored citing relationships (239 in the corpus) whose single predicted high-level treatment category (one of five, from neutral citation to invalidated) matches any of the expert-acceptable ground-truth labels, few-shot prompt with three dataset examples, which are removed from the test set, given the citing-opinion paragraphs that discuss the precedent; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 76.99 | #466 |
| 2 | Gemini 2.5 Pro | 73.27 | #280 |
| 3 | GPT-5 Mini | 70.05 | #373 |
| 4 | Qwen 3 30B A3B | 53.46 | #1165 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=precedent-treatment-high-level-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-22.