Verify-or-Trust: Benchmark Data for LLM Orchestration of a Biology Foundation Model

A benchmark dataset for evaluating whether a Large Language Model correctly allocates verification when orchestrating a fallible biology foundation model. The dataset, created by jang1563, includes a substrate table from the GEARS/Norman experiment. It was last updated on June 17, 2026.

Use Cases

• Benchmarking LLM verification strategies based on predicted perturbation effects from a foundation model.
• Evaluating trust allocation in multi-model AI systems based on the GEARS/Norman substrate data.
• Training or testing AI agents that orchestrate biology models based on the described benchmark harness.

Strengths

• Dataset is part of a defined benchmark with an open-source harness (Apache-2.0) available on GitHub.
• Last updated on 2026-06-17, suggesting recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

jang1563 on Hugging Face.

Collection Method

Data artifacts for the Verify-or-Trust benchmark; likely generated from the GEARS/Norman experiment.

Freshness

Last updated 2026-06-17 04:15:13; freshness should be verified.