Data pipeline surface not built yet
The Data Pipeline is the batch companion to Case Studio. It imports source data (CSV, JSON, JSONL, FHIR R4 bundles, paper extracts, free-text logs), generates synthetic vignettes against user-authored clinical plausibility constraints and diversity targets, and emits typed BenchmarkItem JSONL the runner consumes. A dormant identifier heuristic can add advisory provenance; unlabeled items hand off to the Annotation flow.
The architecture-of-record lives at
docs/data-pipeline.md, mirroring
docs/case-studio.md. The three new skills
(data-source-intake,
data-synthesis,
data-item-shape) and the planned Cloud
Function callables
(pipeline_profile_upload,
pipeline_synthesize_batch,
pipeline_shape_items,
pipeline_merge_labels) ship together once
the surface is wired.