FinCARDS: Card-Based Analyst Reranking for Financial Document Question Answering

TL;DR

FinCARDS introduces a structured, constraint-based reranking framework for financial document question answering, improving stability and transparency without model fine-tuning.

Contribution

It reframes evidence selection as constraint satisfaction using schema-aligned fields, enabling deterministic matching and auditable decision traces.

Findings

Substantially improves early-rank retrieval in financial QA benchmarks.

Reduces ranking variance compared to baseline methods.

Does not require model fine-tuning or variable inference budgets.

Abstract

Financial question answering (QA) over long corporate filings requires evidence to satisfy strict constraints on entities, financial metrics, fiscal periods, and numeric values. However, existing LLM-based rerankers primarily optimize semantic relevance, leading to unstable rankings and opaque decisions on long documents. We propose FinCards, a structured reranking framework that reframes financial evidence selection as constraint satisfaction under a finance-aware schema. FinCards represents filing chunks and questions using aligned schema fields (entities, metrics, periods, and numeric spans), enabling deterministic field-level matching. Evidence is selected via a multi-stage tournament reranking with stability-aware aggregation, producing auditable decision traces. Across two corporate filing QA benchmarks, FinCards substantially improves early-rank retrieval over both lexical and LLM-based reranking baselines, while reducing ranking variance, without requiring model fine-tuning or unpredictable inference budgets. Our code is available at https://github.com/XanderZhou2022/FINCARDS.