Independence Statement & Hub Selection

BenchAudit accepts no sponsorships, paid listings, consulting retainers, or commercial affiliations with any model lab or evaluation hub. Every benchmark in the index is scored strictly from verifiable primary-source documentation using the same four criteria: Privacy, Originality, Freshness, and Grading.

Why certain evaluation hubs appear frequently in Tier S: Organizations such as Artificial Analysis, Epoch AI, Scale Labs, and ARC Prize operate some of the largest portfolios of actively maintained, held-out private evaluations with standardized API harnesses. Their share of Tier S reflects those verifiable technical properties (private test holdouts, newly authored items, and isolated third-party execution), not any partnership or preference. Benchmarks from any independent research group or lab that meet the same primary-source criteria receive identical scores.

Corrections & Submission Policy

Benchmark contamination resistance changes when test sets are leaked, open-sourced, refreshed, or updated with new grading sandboxes. If you are a benchmark maintainer or researcher and spot an outdated fact, broken citation, or newly published evaluation that should be audited:

  • Send the primary-source URL (official paper, GitHub repository, leaderboard page, or documentation) and the specific quoted passage to admin@benchaudit.org.
  • Every update requires verifiable primary-source text; private verbal assurances without documented technical controls are not counted as private holdouts.
  • Confirmed corrections are updated in the underlying fact records, reflected across the web index, entity pages, Markdown mirrors, and periodic CC BY 4.0 Hugging Face snapshots.

Contact

BenchAudit is maintained independently. For benchmark submissions, primary-source corrections, or dataset inquiries, reach us by email.