metadata
license: cc-by-nc-sa-4.0
task_categories:
- question-answering
- text-generation
language:
- en
pretty_name: CCA Numerical Reasoning
size_categories:
- 1K<n<10K
tags:
- numerical-reasoning
- finance
- credit-cards
- program-of-thought
- arithmetic
- financial-literacy
CCA Numerical Reasoning
CCA Numerical Reasoning is an 800-question dataset of credit-card-agreement questions designed to evaluate numerical reasoning over consumer finance terms. Each example contains a natural-language question, a ground-truth answer, output units, input-unit metadata, executable-style reasoning steps, a program trace, question tense metadata, financial-term tags, and the source credit card agreement identifier.
The dataset contains 800 numerical reasoning questions derived from 27 credit card agreements.
Dataset Files
creditqa.json: 800 numerical reasoning questions.
Dataset Structure
Each row in creditqa.json has the following fields:
{
"question": "Natural-language question",
"ground_truth": "Expected answer",
"output_units": "$ | % | binary | month | day | year | week | transaction | hour | categorical",
"input_units": {
"from_doc": {
"%": 0,
"month": 0,
"day": 0,
"year": 0,
"$": 0,
"binary": 0,
"categorical": 0
},
"from_user": {
"%": 0,
"month": 0,
"day": 0,
"year": 0,
"$": 0,
"binary": 0,
"categorical": 0
},
"total": {
"%": 0,
"month": 0,
"day": 0,
"year": 0,
"$": 0,
"binary": 0,
"categorical": 0
}
},
"steps": [
{
"operator": "multiply",
"arg_1": "1000",
"arg_2": "0.02",
"resp": 20
}
],
"program": [
"multiply(1000, 0.02)"
],
"question_tense": "FIRST | THIRD",
"financial_terms": [
"PURCHASE APR",
"INTEREST ACCRUAL"
],
"credit_card": "source_agreement_filename.md"
}