DiagChain - Noisy Evidence: leaderboard
The noisy-evidence profile of DiagChain (Tsinghua University, 2026): attack-chain reconstruction when sampled benign or distracting activity is mixed into the telemetry. Percentage of the 283 coverable reference steps reconstructed with no failure at any stage; higher is better.
Metric: Correctly reconstructed reference steps (%; share of the 283 coverable reference steps of the 23 noisy-evidence cases (sampled benign or distracting activity added); attack-chain reconstruction from heterogeneous telemetry with the ECRAG evidence-retrieval scaffold; a step counts when its evidence is observed, used, fully attributed and correctly ordered, with no earlier failure; temperature 0). Source: arxiv.org. Saturation forecast: Around April 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 41.7 | #49 |
| 2 | GLM-5.2 (High) | 26.5 | #138 (GLM-5.2) |
| 3 | GLM-5.2 (Non-reasoning) | 24.03 | #138 (GLM-5.2) |
| 4 | DeepSeek V4 Pro (Non-reasoning) | 18.73 | #188 (DeepSeek V4 Pro) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=diagchain-noisy-evidence · How It Works · Data refreshed daily, snapshot 2026-11-13.