Datasets:
Tasks:
Question Answering
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
finance
structured-finance
asset-backed-securities
sec-edgar
long-context
document-understanding
License:
Download eval.yaml from ritik12/WaterfallBench: direct link, hf CLI and curl.
- Browser
- Download file 1.04 kB
-
https://huggingface.co/datasets/ritik12/WaterfallBench/resolve/main/eval.yaml
- Command line
-
hf download hf://datasets/ritik12/WaterfallBench/eval.yaml
-
curl -L -o eval.yaml https://huggingface.co/datasets/ritik12/WaterfallBench/resolve/main/eval.yaml
1.04 kB
| name: WaterfallBench | |
| description: > | |
| WaterfallBench tests whether a model can execute an asset-backed securities waterfall to the cent from the deal | |
| documents alone. Each of the 20 tasks is a real auto-loan or lease securitization at its first distribution date: | |
| the final prospectus, the closing agreements and that month's servicer report with every waterfall result removed | |
| (7-11 SEC EDGAR filings, 500K-740K tokens). The model returns the servicing fee, each note class's interest, | |
| principal and ending balance, the reserve balance and the residual in integer cents, graded exactly against what | |
| the trustee paid. Tasks run as Harbor tasks; answer keys are private. Scores are percentages (0-100). | |
| evaluation_framework: harbor | |
| leaderboard: | |
| base_models_only: false | |
| tasks: | |
| # Primary metric: share of tasks where every graded field is exact to the cent. | |
| - id: fully_resolved | |
| config: default | |
| split: test | |
| # Share of non-trivial graded fields exactly right. | |
| - id: field_accuracy | |
| config: default | |
| split: test | |