Tier 2 includes CC-BY-4.0, OpenMDW-1.1 and every gated model even if its licence is MIT. Select these models explicitly by catalog ID. They are never defaults, mirrored weights or offline model packages. They load pinned publisher files with SHA-256 verification and may convert into the user's cache; converted tier-2 artifacts must not be redistributed. ModelInfo preserves licence, source, revision, file digests and exact attribution.
A publisher gate requires your own account's acceptance and an authorized read token. The library does not accept terms for you. Supply the explicit token parameter on ModelStore.PullAsync, or inject ONESPEECH_MODEL_TOKEN through your host's secret manager. Do not put credentials in command-line arguments, source, JSON transcripts, shared shell history or logs. The CLI reads the environment variable. Gate authorization is required even for cached files and custom sources.
HTTP authorization is sent only on the original exact publisher request, never to redirects or mirrors. Token material is not persisted by the downloader. Distinguish ModelTokenRequiredException, ModelTokenRejectedException and ModelGateNotAcceptedException; fix account access rather than retrying indefinitely. Network failures and digest mismatches have separate typed errors. Never bypass a gate with an unofficial mirror.
The segmentation slot supports authorized local segmentation-3.0 snapshots and overlap-aware turns with our clustering. Its network entry point still refuses inference, and community-1 has no native pipeline adapter. Successful pre-fetch alone does not imply that a family has a working inference adapter. Nemotron 3 Diarization and Parakeet are ungated tier-2 examples with published inference support. See model loading for cache precedence, trust boundaries and custom ModelSource hooks.