Replayable Financial Agents
Published "Replayable Financial Agents" — 4,705 agent runs at temperature 0.0 across 7 models and 4 providers. By the author's measurement, decision reproducibility showed no correlation with task accuracy (r=-0.11, p=0.63). Large models of 120B+ had reproducibility under 20%
Why it matters
The domain is finance, but the requirement to replay for audits has the same structure as quality audits. Refutes, with 4,705 runs, the common belief that picking a more accurate model brings reproducibility along with it
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/html/2601.15322v2Added in Week 35, 2026 · See that week