Package ai.pipestream.module.pipelineprobe.moduletest.scaffold

package ai.pipestream.module.pipelineprobe.moduletest.scaffold
  • Class
    Description
    ChunkerScaffold
    Chunker scaffold: the raw chunker input (echo output) carries no vector-set directives, so the chunker no-ops ("nothing to chunk").
    OpenSearchSinkScaffold
    opensearch-sink scaffold, e2e-faithful: provisions exactly what a pipeline crawl provisions before the sink sees a doc — the full 2×2 vector-set registration (sentence-10-3 + sentences_internal × minilm + paraphrase-minilm) via EmbedderFixturesRegistrar, matching what the captured corpus docs actually carry; an IndexPlan with a PINNED id and the caller-chosen IndexingStrategy (NESTED / SEPARATE_INDICES / CHUNK_COMBINED — the same three the e2e exercises; default CHUNK_COMBINED, the e2e default), gated on READY like the crawl gates dispatch; the semantic centroid configs assigned to the index, so the sink can write the document/section/paragraph centroid SPRs the corpus carries ("KNN field not provisioned at bind time" otherwise). The plan id + instance ride StreamMetadata.context_params — what OpenSearchSinkOptionsParser reads.
    ParserScaffold
    Parser scaffold: the parser has no captured <step>-input-court dataset — its input is the raw source file, which exists BEFORE any capture tap (and the court corpus is JDBC, which skips the parser entirely).
    PassthroughScaffold
    No-setup scaffold: the corpus doc already carries everything the step needs.
    StepScaffold
    Per-step integration scaffolding for module perf / isolation tests.
    StepScaffolds
    Factory for per-step StepScaffolds — forStep("embedder") etc.