Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents
Weicheng Xue · 2026
Read Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents by Weicheng Xue (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-2605-28850.txt.
Description
Weicheng Xue studies conduct fit and picture shifts of language agents in market settings. TradeArena, an open trading testbed with risk reports, trade play, memory, and saved runs, lets analysts trace reasons, positions, and nudges under stress. Pre-failure signs emerge: planning vectors drift from centroids, fused plan-risk pictures separate pre-drawdown states, and manifolds show effective-rank squeeze. Tests show squeeze can vanish without reasons, while intent-space signs stay useful, and risk feedback improves tuning for some models and returns for others without fine-tuning.
Catalog record
| CREATOR | Weicheng Xue |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 104704 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.28850