Show HN: FinTrace – point-in-time guardrails for financial LLM agents
Category: ai-ml
Tags: financial-nlp, llm-agents, point-in-time-data, benchmark, sec-edgar
Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)
FinTrace is a point-in-time financial research agent framework and benchmark that prevents LLM look-ahead bias by enforcing as-of date gating on SEC EDGAR data. It includes a leakage-controlled benchmark (PITfall) showing that structural time-gating reduces future information leakage from 82% to 0%, with ablations proving the gate is more effective than prompt discipline alone.
Target audience: financial ML researchers, quantitative analysts, AI safety engineers
Repository: https://github.com/FredXing1/FinTrace · Python · Apache-2.0
View on Hacker News