This paper presents an exploratory pilot comparing three LLM deployment architectures for financial data governance: (A) platform-managed agents, (B) a custom multi-agent factory pattern, and (C) direct API calls, evaluated on five governance tasks with a single run per approach (n=1 per cell; 15 single-point measurements total). Approaches A and C were executed as real API workloads; Approach B figures come from an agent-based simulation of the DIY factory pattern, not a live execution against the API, and are indicative rather than measured. In this single run, Approach A recorded the highest quality (8.64/10) at 0.105 USD per task, Approach B (simulated) 7.48 at 0.102 USD, and Approach C 6.60 at 0.048 USD; platform-level prompt caching reduced repeated-context costs by roughly 90 percent. No significance testing is applied; all comparisons are descriptive and hypothesis-generating, not confirmatory.Editorial revision. The prose was converted to an impersonal, formal register (single-author paper). No data, numbers, claims, tables, or references were changed from the previous version.
Karl J. Mollan Neyra (Thu,) studied this question.