Builds the closed selector-safe public identity for an acquired provider.
The declaration projection is operator-authored safe behavior identity. The acquisition projection contains only bounded facts captured from the active provider. Unattested selectors, endpoints, commands, paths, credentials, OAuth authority, and identifiers derived from secret-bearing values are never admitted. An adapter may explicitly attest its exact model target as public. Audited executable and launcher content digests may be included as non-secret build identity without exposing their paths.
llm_identity/1 extracts model attribution from a complete current LLM
snapshot, including declaration config that carries max_calls. It also
accepts the previous three-key config so already-captured traces stay
attributable. Malformed or private snapshots return no attribution and remain
loadable by the canonical trace boundary.
Summary
Functions
Extracts {alias, installation revision, resolved model} from one valid
current LLM provider snapshot.
Functions
@spec llm_identity(term()) :: {:ok, %{alias: binary(), installation_revision: binary(), resolved_model: binary()}} | :error
Extracts {alias, installation revision, resolved model} from one valid
current LLM provider snapshot.
Hash verification detects inconsistent or edited trace data; the hashes are
not signatures. Invalid and private snapshots return :error. A current
snapshot whose declaration config omits max_calls remains attributable.