Proxion published notes on testing the grader using seeded error injection to evaluate its detection rate for defect classes.

Proxion turns institutional financial judgment into structured training and evaluation data for AI systems. Most finance content explains the facts. It does not capture the judgment behind real decisions: how practitioners underwrite risk, pressure test assumptions, build valuation logic, and reason through complex financial situations. Proxion captures that judgment. We work with credentialed finance professionals across investment banking, private equity, and asset management to produce training datasets, evaluation benchmarks, reasoning traces, rubrics, and preference data for financial AI systems. Our work helps AI systems move beyond financial retrieval toward financial reasoning. We support frontier AI labs, fintech platforms, and financial institutions building reliable, finance-grade AI. Based in New York.