Member Junction
    Preparing search index...
    VENDOR_TYPE: {
        InferenceProvider: "5B043EC3-1FF2-4730-B5D2-7CFDA50979B3";
        ModelDeveloper: "10DB468E-F2CE-475D-9F39-2DF2DE75D257";
    } = ...

    Realistic AI-metadata catalog fixtures shared across AI-stack unit tests.

    These mirror the SHAPE and representative VALUES of the production seed data (metadata/ai-vendors, ai-models, ai-model-types, ai-vendor-type-definitions, ai-configurations) so model-selection / failover / execution tests exercise believable catalogs: real driver-class strings (AnthropicLLM / OpenAILLM / GeminiLLM / GroqLLM…), real API names, real PowerRanks, and the multi-vendor / multi-priority layout (a model offered by its developer + several inference providers at different priorities) that selection logic actually has to navigate.

    The IDs match the real hardcoded seed UUIDs where they are fixed (vendor type definitions, vendors, model types, configurations); model/prompt-model IDs are stable test UUIDs.

    Pure data only — no vitest/mock coupling. Tests wire these into whatever AIEngine mock their package uses (see the AIPromptRunner suites in packages/AI/Prompts for the reference wiring). Originally proven in packages/AI/Prompts/src/tests/fixtures/ai-metadata.fixtures.ts and promoted here so every AI package can build a models+vendors+prompts catalog without copying it.

    Type Declaration

    • ReadonlyInferenceProvider: "5B043EC3-1FF2-4730-B5D2-7CFDA50979B3"
    • ReadonlyModelDeveloper: "10DB468E-F2CE-475D-9F39-2DF2DE75D257"