Metadata & Provenance: The Run Manifest

The problem — Without a manifest, a folder of TIFFs is just pixels. Skip metadata capture and you silently lose the context that makes results comparable: which compound and dose sat in which well, what scanner and exposure produced the field, which wells were controls. Months later you cannot reproduce a run, and you cannot tell a real biological signal from a swapped plate map.

What it is / how it works — A run manifest is the structured record of what was acquired and how. Huisman et al. split this into two complementary kinds: provenance metadata (the instrument and acquisition story — MPM) and quality metadata (calibration and error — MQM). The community has converged on a tiered, FAIR-aligned model, 4DN-BINA-OME, and tools like micro meta app make capturing it tractable — validated across 16 core facilities to produce more uniform metadata. In Fovea pipelines the manifest is written alongside the harmonized ome zarr store using ome ngff conventions, so plate map, perturbations, acquisition geometry, and controls travel with the pixels rather than in a lab notebook.

Where it breaks — Metadata captured once, by hand, at deposition time is metadata that drifts from the data. The fix is to make the manifest a build artifact: generated at ingest, validated against a schema, versioned, and carried downstream so every figure traces back to its acquisition. A manifest that cannot answer "which well, which control, which scanner" is not provenance — it is decoration. This is the upstream contract that makes Why Microscopy Needs Verified Outputs enforceable and feeds the The QC-Aware Report.