Replayable Financial Agents: A Determinism-Faithfulness Assurance Harness for Tool-Using LLM Agents

TL;DR

This paper introduces DFAH, a framework for measuring determinism and faithfulness in tool-using LLM agents in finance, revealing that models can be deterministic without being accurate and vice versa.

Contribution

The paper presents DFAH, a novel multi-dimensional evaluation framework for assessing determinism and faithfulness in financial LLM agents, supported by extensive empirical analysis.

Findings

Determinism and accuracy are uncorrelated in financial LLM agents.

Small models achieve high determinism but low accuracy.

Frontier models show moderate determinism with variable accuracy.

Abstract

LLM agents struggle with regulatory audit replay: when asked to reproduce a flagged transaction decision with identical inputs, many deployments fail to return consistent results. We introduce the Determinism-Faithfulness Assurance Harness (DFAH), a framework for measuring trajectory determinism, decision determinism, and evidence-conditioned faithfulness in tool-using agents deployed in financial services. Across 4,700+ agentic runs (7 models, 4 providers, 3 financial benchmarks with 50 cases each at T=0.0), we find that decision determinism and task accuracy are not detectably correlated (r = -0.11, 95% CI [-0.49, 0.31], p = 0.63, n = 21 configurations): models can be deterministic without being accurate, and accurate without being deterministic. Because neither metric predicts the other in our sample, both must be measured independently, which is precisely what DFAH provides. Small models (7-20B) achieve near-perfect determinism through rigid pattern matching at the cost of accuracy (20-42%), while frontier models show moderate determinism (50-96%) with variable accuracy. No model achieves both perfect determinism and high accuracy, supporting DFAH's multi-dimensional measurement approach. We provide three financial benchmarks (compliance triage, portfolio constraints, and DataOps exceptions; 50 cases each) together with an open-source stress-test harness. Across these benchmarks and DFAH evaluation settings, Tier 1 models with schema-first architectures achieved determinism levels consistent with audit replay requirements.