DeEscalWild - ROUGE-L: leaderboard

DeEscalWild (2026) civilian role-play over 150 real police-civilian interactions, scored by ROUGE-L overlap (0-100) between the generated and the real civilian turns; higher is better.

Metric: ROUGE-L (0-100) of the generated civilian turns against the real civilian turns, on DeEscalWild's 150 held-out police-civilian interactions transcribed from real social-media footage: at each turn the model, prompted zero-shot with the situation and a civilian persona, receives the gold officer utterance and generates the civilian reply (autoregressive simulation); mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Llama 3.2 3B Instruct11.3#7241
2Qwen 2.5 3B Instruct11.2#5392
3granite-3.0-2B-instruct10.2#8277
4Falcon3-3B-Instruct9.2#8313
5Gemma 2 2B (IT)7.2#7054

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=deescalwild-rouge-l · How It Works · Data refreshed daily, snapshot 2026-11-21.