Ficha is a dataset of SEC 10-K financial filings designed to evaluate how well chunking algorithms handle formal business documents with complex financial terminology, tables, and structured sections.

Dataset Description

• Documents: 88 SEC 10-K filings from major US companies
• Questions: 1,331 question-answer pairs
• Domain: Financial/Business
• Source: SEC EDGAR database

Key Challenges

This dataset tests chunking algorithms on:

• Financial tables and numerical data
• Legal/regulatory language
• Structured sections (Risk Factors, MD&A, Financial Statements)
• Cross-references between sections
• Technical accounting terminology

Dataset Structure

Corpus Config

Questions Config

Usage

from datasets import load_dataset

# Load corpus
corpus = load_dataset("chonkie-ai/ficha", "corpus", split="train")

# Load questions
questions = load_dataset("chonkie-ai/ficha", "questions", split="train")

Part of MTCB

Ficha is part of the Massive Text Chunking Benchmark (MTCB), a comprehensive benchmark for evaluating RAG chunking strategies.

License

CC-BY-4.0