Effective: Owner configuration required before launch
Evidence before conclusion
BenchLM prefers benchmark-native leaderboards, official model cards and documentation, papers, and exact public result tables. Published raw benchmark rows are never fabricated. Composite rankings use the disclosed conservative prior when weighted coverage is missing, with source, freshness, coverage, and verification state shown where available.
Rankings and uncertainty
Ranking methods, category weights, eligibility rules, and confidence language are documented in the methodology. Verified and provisional evidence remain visibly distinct. A benchmark score is not presented as proof of universal model quality.
Commercial independence
Advertising, sponsorship, and affiliate relationships do not buy positions, alter weights, suppress unfavorable rows, or determine coverage. Commercial links are labeled and governed by the affiliate disclosure.
Automation and AI assistance
BenchLM uses software and AI-assisted workflows to collect, normalize, check, and draft material. Automated output is not treated as evidence by itself. Data claims must resolve to the source pipeline or a cited public source, and editorial material remains subject to human review before publication.
Corrections
Material factual errors are corrected in the live page and its generated data surfaces. Updated dates describe substantive review, not routine deployment. To report an error, include the page, disputed value, and original source when possible and contact BenchLM.