Skip to content
SynthWorldDeterministic identity benchmarks
Esc
navigateopen⌘Jpreview

Benchmark publication

Benchmark publication is an allowlisted transition, not a copy operation. The validated benchmark registry records lifecycle and sensitivity separately from capability maturity.

Publication requires approved artifacts, checksums, schemas, integrity tests, packaging checks, documentation, and review. Public/evaluator separation still applies when both sides are published. No generated world, evaluator key, private source import, or local run receipt is published automatically.

See the benchmark reference. Existing approved frozen tables are already published on Hugging Face; publishing additional families is a later controlled phase and is not implied by package availability.

Was this page helpful?