groundtruth-dynamic-benchmarking/CITATION.cff at main · EigenformAI/groundtruth-dynamic-benchmarking · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
cff-version: 1.2.0
title: "Groundtruth Dynamic Benchmarking"
message: "If you use this benchmark, its harness, or its rubrics, please cite it as below."
type: software
authors:
  - name: "Eigenform"
repository-code: "https://github.com/EigenformAI/groundtruth-dynamic-benchmarking"
url: "https://eigenform.ai/benchmark"
abstract: "Groundtruth Dynamic Benchmarking is an Earth Sciences AI benchmark: it evaluates how well large language models reason over real geoscience source material such as mineral-exploration reports, government geological archives, and technical filings. A candidate model answers a fixed question set drawn from primary records; an independent LLM judge grades every answer 0-10 against a calibrated, source-anchored rubric whose every claim carries a verbatim evidence locator. Geology is the first edition; the harness runs any schema-2.0 rubric. A public leaderboard hosts a model track and a harness track over three 50-question districts."
keywords:
  - "earth sciences AI"
  - "geoscience"
  - "LLM benchmark"
  - "LLM evaluation"
  - "LLM-as-judge"
  - "rubric grading"
  - "mineral exploration"
  - "question answering"
  - "retrieval-augmented generation"
  - "fine-tuning"
version: "1.1.0"
date-released: "2026-09-02"
license: MIT