Datasets:
Tasks:
Question Answering
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
finance
structured-finance
asset-backed-securities
sec-edgar
long-context
document-understanding
License:
Request: add WaterfallBench to the Benchmark allow-list
#1
by ritik12 - opened
We've added a root eval.yaml (evaluation_framework: harbor) with two tasks: fully_resolved (primary) and field_accuracy, both percentages. WaterfallBench has 20 real ABS securitizations from SEC EDGAR, graded to the cent against what the trustee actually paid. There is no LLM judge, and the answer keys are private. The dataset card documents the tasks, metrics and current results. Could you add it to the allow-list, and tell us whether the eval.yaml needs any changes?