BC Finance LLM Golden Set: Regression and Hallucination Test Cases

A collection of 550 test cases for evaluating large language models in financial contexts. It was produced by BC Card and Yonsei University DSL as part of the S2026 LLMOps project. The dataset includes 300 regression test cases, 200 financial edge cases for hallucination detection, and 50 hard negative QA samples.

Use Cases

• Conduct regression testing for LLM updates based on the 300-case regression golden set.
• Detect hallucinations in financial LLM outputs using the 200 edge cases.
• Evaluate model robustness with hard negative samples from the 50-case QA set.
• Implement a deployment gate by setting score thresholds against the provided test suites.

Strengths

• Contains 550 total curated test cases across three specific subsets.
• Includes 200 financial edge cases explicitly designed for hallucination detection.
• Designed for a concrete industrial-academic collaboration (BC Card-Yonsei University S2026 LLMOps project).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for individual files is provided, but total data scale and feature details are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

BC Card and Yonsei University DSL (산학협력)

Collection Method

Created as an output of the S2026 LLMOps project.

Freshness

Last updated 2026-06-29 08:06:05; freshness should be verified.