Manifest

Every download writes a manifest.json into the request directory. It is collekt’s machine-readable record of what was requested, fetched, reused, and skipped, and it is the contract downstream consumers rely on.

Structure

{
  "manifest_schema_version": "0.1",
  "request": {
    "region": {"west": -6.0, "east": 20.0, "south": 35.0, "north": 45.0},
    "start": "2023-06-15T00:00:00Z",
    "end": "2023-06-15T23:59:59Z",
    "metadata": {}
  },
  "summary": {"planned": 0, "downloaded": 1, "reused": 0, "skipped": 0, "failed": 0},
  "sources": { "cmems_glorys_my": { "kind": "cmems", "variables": ["uo", "vo"], "...": "..." } },
  "files":    [ { "source": "cmems_glorys_my", "status": "downloaded", "path": "cmems/global/glorys/…nc", "format": "netcdf", "inspection": {"size_bytes": 12345, "data_variables": ["uo", "vo"]} } ],
  "planned":  [],
  "warnings": [],
  "failures": []
}

• Paths are relative to the request directory, so a collection directory or zip archive is relocatable.
• Downloaded/reused files carry an inspection block: variables/sizes for NetCDF, columns/row count for Parquet, otherwise the file size.
• warnings holds skipped sources; failures holds failed ones. Both are non-blocking unless the run is --strict.

Draft schema

A permissive draft JSON Schema ships at collekt/resources/schema/manifest.schema.json. During alpha the manifest stays flexible and the schema is provided for reference only — it is not enforced. Core fields are required; provider-specific metadata under details may evolve. manifest_schema_version is bumped when the shape changes materially; a strict, validated schema is a 1.0 goal.