Research & evaluations

Research policy

Standards for evidence, editorial independence and corrections.

Evidence and publication status

A pilot is exploratory. A documentary assessment evaluates what sources describe. A hands-on evaluation requires recorded task execution. A reviewed release requires an actual reviewer record; the current catalog does not claim that status.

Commercial independence

Vendors cannot purchase scores or benchmark positions. Material funding and commercial relationships relevant to a release must be disclosed. Evidence supplied by a vendor must be identified and evaluated under the same criteria as other evidence.

Corrections

Submit the benchmark URL, release, disputed result, proposed correction and supporting source to hello@yute.io. Confirmed corrections require a dated change entry. Results that cannot be supported should be qualified, superseded or withdrawn, with the reason retained.

Submit evidence

Data access

Each release states what can be downloaded and what requires an access request. Publishing a link or excerpt does not grant rights to redistribute third-party material. Private client data and credentials must not enter public artifacts.

Versions and review

Changing questions, scoring criteria or weights creates a new protocol version. A rerun records model and collection context. Results from incompatible conditions must not be displayed as a continuous trend without explanation.