ProMAS: Proactive Error Forecasting for Multi-Agent Systems Using Markov Transition Dynamics
Xinkui Zhao, Sai Liu, Yifan Zhang, Qingyu Ma, Guanjie Cheng, Naibo Wang, Chang Liu · 2026
Read ProMAS: Proactive Error Forecasting for Multi-Agent Systems Using Markov Transition Dynamics by Xinkui Zhao, Sai Liu, Yifan Zhang, Qingyu Ma, Guanjie Cheng, Naibo Wang, Chang Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-20260.txt.
Description
Zhao and co-authors forecast errors in multi agent language model systems before the failure lands. One agent's fallacy can spread fast and sink a long joint task. Most tools study failures only after the fact. ProMAS models reasoning steps as Markov transitions in a quantized vector space. Causal delta features capture how meaning shifts between steps. A prediction head with jump detection spots risk acceleration, not fixed thresholds. On the Who&When benchmark it read only 27 percent of logs. Speed of intervention is the main gain claimed.
Catalog record
| CREATOR | Xinkui Zhao, Sai Liu, Yifan Zhang, Qingyu Ma, Guanjie Cheng, Naibo Wang, Chang Liu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45455 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.20260