Editorial and data standards
Our aim is to make evaluation evidence easier to inspect. We separate reported observations, source descriptions, and our explanatory writing so readers can tell what each claim is based on.
Evidence before rankings
Catalog records preserve their source identity. Score observations retain the available source, unit, protocol, and date precision. A reported model score is not an independent endorsement or a reproduction by this website.
We do not combine unrelated benchmarks into a universal model ranking. Matching metric names alone do not prove comparable conditions. When evaluation context is missing, the page should say so.
How guides are prepared
A guide explains the task, evaluation method, useful applications, and limitations of a named benchmark. References should support the nearby claims. Version-specific details stay tied to that version.
A review date indicates when the cited material was checked for that guide. It is not the benchmark release date, proof of a fresh experiment, or a claim of external peer review.
Automation and AI assistance
Automated collection helps maintain the catalog, and AI tools assist research, drafting, and translation. Those tools can make mistakes. Explanatory pages cite public evidence and distinguish interpretation from reported facts.
Neither fluent writing nor an automated check establishes that a scientific claim is correct. Readers should consult the original methods and results before making consequential decisions.
Corrections and updates
A useful correction identifies the page, the disputed statement or result, and a primary source that supports the change. We distinguish a source changing its report from this website making a transcription or interpretation error.
Snapshot dates describe saved collections. They do not promise uninterrupted daily coverage. A material change to an explanatory page should update its review date; an unchanged page should not receive a fresh date merely to appear current.
Research context and commercial separation
Cybersecurity pages describe evaluation environments and reported measurements for research. They are not instructions or authorization to attack real systems. A benchmark’s capability claim should not be mistaken for a deployment safety assessment.
There are currently no advertisements or paid ranking placements. If commercial relationships are introduced, relevant pages will disclose them; payment will not be presented as scientific evidence.