{
  "$defs": {
    "AssemblySource": {
      "additionalProperties": false,
      "description": "One upload of a multi-file dataset version (M4) and where it goes.\n\nThe upload is ingested with its own ``mapping`` into node ``node``; a file that is\nitself a tree mounts its nodes *under* ``node`` (file group ``/psf`` at node\n``/instrument`` becomes ``/instrument/psf``). Sources of the same node are combined by\n``combine`` (see :data:`CombineRule`).",
      "examples": [
        {
          "combine": "concat",
          "concat_dim": "year",
          "concat_label": 2024,
          "mapping": {
            "format": "netcdf"
          },
          "node": "/",
          "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p0"
        }
      ],
      "properties": {
        "combine": {
          "default": "node",
          "enum": [
            "node",
            "concat",
            "merge"
          ],
          "title": "Combine",
          "type": "string"
        },
        "concat_dim": {
          "anyOf": [
            {
              "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "combine 'concat': the dim to stack on",
          "title": "Concat Dim"
        },
        "concat_label": {
          "anyOf": [
            {
              "type": "integer"
            },
            {
              "type": "number"
            },
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "combine 'concat': this file's label along concat_dim",
          "title": "Concat Label"
        },
        "join": {
          "default": "exact",
          "description": "How coords align when sources are combined",
          "enum": [
            "exact",
            "inner",
            "outer"
          ],
          "title": "Join",
          "type": "string"
        },
        "mapping": {
          "$ref": "#/$defs/IngestMapping"
        },
        "node": {
          "default": "/",
          "pattern": "^/(?:[A-Za-z_][A-Za-z0-9_.\\-]{0,63}(?:/[A-Za-z_][A-Za-z0-9_.\\-]{0,63}){0,7})?$",
          "title": "Node",
          "type": "string"
        },
        "upload_id": {
          "examples": [
            "run_01jaf3q2x8m9v7k5h4g3f2e1d0"
          ],
          "pattern": "^[A-Za-z0-9][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Upload Id",
          "type": "string"
        }
      },
      "required": [
        "upload_id",
        "mapping"
      ],
      "title": "AssemblySource",
      "type": "object"
    },
    "CollectionOptions": {
      "additionalProperties": false,
      "description": "How a client manifest and image parts become a collection version.\n\nThe manifest is a CSV, TSV or Parquet table with one row per image and at least the\nsha256 of each image's bytes. Its other columns become the id, label, split and path of\neach image, or are kept as coords.",
      "properties": {
        "columns": {
          "anyOf": [
            {
              "items": {
                "maxLength": 256,
                "minLength": 1,
                "type": "string"
              },
              "maxItems": 64,
              "type": "array"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Other columns kept as coords (default: all others)",
          "title": "Columns"
        },
        "dim": {
          "default": "image",
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Dim",
          "type": "string"
        },
        "id_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Unique ids of the image dim (default: the path, else the sha256)",
          "title": "Id Column"
        },
        "label_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Class labels: the text variable 'label(image)', role target",
          "title": "Label Column"
        },
        "max_images": {
          "default": 100000,
          "maximum": 1000000,
          "minimum": 1,
          "title": "Max Images",
          "type": "integer"
        },
        "mode": {
          "default": "auto",
          "description": "auto: rgb when an image has colour, else gray",
          "enum": [
            "auto",
            "gray",
            "rgb"
          ],
          "title": "Mode",
          "type": "string"
        },
        "path_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": "path",
          "description": "Kept as the coord 'path' (ignored when the default is missing)",
          "title": "Path Column"
        },
        "resize": {
          "default": "stretch",
          "enum": [
            "stretch",
            "crop",
            "pad"
          ],
          "title": "Resize",
          "type": "string"
        },
        "sha256_column": {
          "default": "sha256",
          "description": "Lowercase hex sha256 of each image",
          "maxLength": 256,
          "minLength": 1,
          "title": "Sha256 Column",
          "type": "string"
        },
        "size": {
          "anyOf": [
            {
              "maxItems": 2,
              "minItems": 2,
              "prefixItems": [
                {
                  "type": "integer"
                },
                {
                  "type": "integer"
                }
              ],
              "type": "array"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Canonical (height, width) for samples and model defaults; each 1..4096",
          "title": "Size"
        },
        "split_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Train/test assignment: the text coord 'split(image)'",
          "title": "Split Column"
        },
        "var": {
          "default": "pixels",
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Var",
          "type": "string"
        }
      },
      "title": "CollectionOptions",
      "type": "object"
    },
    "CollectionSource": {
      "additionalProperties": false,
      "description": "``DatasetVersionCreate.collection``: a manifest upload plus the parts it needs.\n\n``base_version``: ``None`` uses the latest READY collection version of the dataset\n(when there is one) as the source of images the parts do not hold; ``0`` uses no base.",
      "properties": {
        "base_version": {
          "anyOf": [
            {
              "minimum": 0,
              "type": "integer"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Base Version"
        },
        "manifest_upload_id": {
          "examples": [
            "run_01jaf3q2x8m9v7k5h4g3f2e1d0"
          ],
          "pattern": "^[A-Za-z0-9][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Manifest Upload Id",
          "type": "string"
        },
        "options": {
          "$ref": "#/$defs/CollectionOptions"
        },
        "part_upload_ids": {
          "items": {
            "examples": [
              "run_01jaf3q2x8m9v7k5h4g3f2e1d0"
            ],
            "pattern": "^[A-Za-z0-9][A-Za-z0-9_.\\-]{0,127}$",
            "type": "string"
          },
          "maxItems": 64,
          "title": "Part Upload Ids",
          "type": "array"
        }
      },
      "required": [
        "manifest_upload_id"
      ],
      "title": "CollectionSource",
      "type": "object"
    },
    "DimMapping": {
      "additionalProperties": false,
      "description": "A dim of the dataset, read from one column of a tabular upload.",
      "examples": [
        {
          "column": "date",
          "spec": {
            "freq": "D",
            "kind": "time",
            "policy": "extend"
          }
        }
      ],
      "properties": {
        "column": {
          "maxLength": 256,
          "minLength": 1,
          "title": "Column",
          "type": "string"
        },
        "spec": {
          "$ref": "#/$defs/DimSpec"
        }
      },
      "required": [
        "column",
        "spec"
      ],
      "title": "DimMapping",
      "type": "object"
    },
    "DimPolicy": {
      "description": "What may happen to a dim's coordinates at score time.",
      "enum": [
        "fixed",
        "extend",
        "new_levels"
      ],
      "title": "DimPolicy",
      "type": "string"
    },
    "DimSpec": {
      "additionalProperties": false,
      "description": "A dimension and its score-time policy.",
      "properties": {
        "description": {
          "anyOf": [
            {
              "maxLength": 2000,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Description"
        },
        "freq": {
          "anyOf": [
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Pandas-style frequency, e.g. 'W-SUN'",
          "title": "Freq"
        },
        "kind": {
          "enum": [
            "categorical",
            "time",
            "integer",
            "float",
            "index"
          ],
          "title": "Kind",
          "type": "string"
        },
        "max_extension": {
          "anyOf": [
            {
              "minimum": 0,
              "type": "integer"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Periods allowed past the fit",
          "title": "Max Extension"
        },
        "new_level_strategy": {
          "default": "error",
          "enum": [
            "error",
            "hyperprior",
            "pooled",
            "zero"
          ],
          "title": "New Level Strategy",
          "type": "string"
        },
        "ordered": {
          "default": false,
          "title": "Ordered",
          "type": "boolean"
        },
        "policy": {
          "$ref": "#/$defs/DimPolicy",
          "default": "fixed"
        }
      },
      "required": [
        "kind"
      ],
      "title": "DimSpec",
      "type": "object"
    },
    "Hdf5Coord": {
      "additionalProperties": false,
      "description": "The coordinate of dim ``dim`` in node ``node`` of a raw HDF5 upload (M4).\n\nExactly one source: ``path`` (a 1-D HDF5 dataset holding the labels), or ``start`` with\nan optional ``step`` (a range: numbers, or a date and a pandas frequency such as\n``\"2024-01-01\"`` + ``\"W-MON\"``), or neither (an integer index ``0..n-1``). ``spec``\noverrides the inferred :class:`~mapi_spec.schema.DimSpec`.",
      "examples": [
        {
          "dim": "row",
          "spec": {
            "kind": "float",
            "ordered": true
          },
          "start": 0.0,
          "step": 0.05
        },
        {
          "dim": "date",
          "path": "/axes/date",
          "spec": {
            "kind": "time",
            "policy": "extend"
          }
        }
      ],
      "properties": {
        "dim": {
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Dim",
          "type": "string"
        },
        "node": {
          "default": "/",
          "pattern": "^/(?:[A-Za-z_][A-Za-z0-9_.\\-]{0,63}(?:/[A-Za-z_][A-Za-z0-9_.\\-]{0,63}){0,7})?$",
          "title": "Node",
          "type": "string"
        },
        "path": {
          "anyOf": [
            {
              "pattern": "^/[^\\x00]{1,1024}$",
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Path"
        },
        "spec": {
          "anyOf": [
            {
              "$ref": "#/$defs/DimSpec"
            },
            {
              "type": "null"
            }
          ],
          "default": null
        },
        "start": {
          "anyOf": [
            {
              "type": "integer"
            },
            {
              "type": "number"
            },
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Start"
        },
        "step": {
          "anyOf": [
            {
              "type": "integer"
            },
            {
              "type": "number"
            },
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Step"
        }
      },
      "required": [
        "dim"
      ],
      "title": "Hdf5Coord",
      "type": "object"
    },
    "Hdf5Mapping": {
      "additionalProperties": false,
      "description": "How a raw HDF5 upload becomes a (tree) dataset (M4): which arrays become which\nvariables of which node, with which dims and coords. Arrays that are not mapped are\nreported in one warning, never read.",
      "examples": [
        {
          "coords": [
            {
              "dim": "row",
              "spec": {
                "kind": "float",
                "ordered": true
              },
              "start": 0.0,
              "step": 0.05
            }
          ],
          "vars": [
            {
              "dims": [
                "row",
                "col"
              ],
              "name": "image",
              "node": "/",
              "path": "/observations/image",
              "units": "counts"
            },
            {
              "dims": [
                "prow",
                "pcol"
              ],
              "name": "psf",
              "node": "/psf",
              "path": "/psf/kernel"
            }
          ]
        }
      ],
      "properties": {
        "coords": {
          "items": {
            "$ref": "#/$defs/Hdf5Coord"
          },
          "maxItems": 64,
          "title": "Coords",
          "type": "array"
        },
        "vars": {
          "items": {
            "$ref": "#/$defs/Hdf5Var"
          },
          "maxItems": 1024,
          "minItems": 1,
          "title": "Vars",
          "type": "array"
        }
      },
      "required": [
        "vars"
      ],
      "title": "Hdf5Mapping",
      "type": "object"
    },
    "Hdf5Var": {
      "additionalProperties": false,
      "description": "A data variable read from one HDF5 dataset of a raw HDF5 upload (M4).\n\n``dims`` names the array's axes in order (its length must equal the array's ``ndim``;\nthe sandbox checks it). The variable is stored as ``name`` in tree node ``node``.",
      "examples": [
        {
          "dims": [
            "row",
            "col"
          ],
          "name": "image",
          "node": "/",
          "path": "/observations/image",
          "units": "counts"
        }
      ],
      "properties": {
        "description": {
          "anyOf": [
            {
              "maxLength": 2000,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Description"
        },
        "dims": {
          "items": {
            "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
            "type": "string"
          },
          "maxItems": 32,
          "title": "Dims",
          "type": "array"
        },
        "missing": {
          "default": "forbid",
          "enum": [
            "forbid",
            "allow",
            "mask"
          ],
          "title": "Missing",
          "type": "string"
        },
        "name": {
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Name",
          "type": "string"
        },
        "node": {
          "default": "/",
          "pattern": "^/(?:[A-Za-z_][A-Za-z0-9_.\\-]{0,63}(?:/[A-Za-z_][A-Za-z0-9_.\\-]{0,63}){0,7})?$",
          "title": "Node",
          "type": "string"
        },
        "path": {
          "pattern": "^/[^\\x00]{1,1024}$",
          "title": "Path",
          "type": "string"
        },
        "role": {
          "$ref": "#/$defs/Role",
          "default": "covariate"
        },
        "units": {
          "anyOf": [
            {
              "maxLength": 64,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Units"
        }
      },
      "required": [
        "path",
        "name"
      ],
      "title": "Hdf5Var",
      "type": "object"
    },
    "ImageFolderOptions": {
      "additionalProperties": false,
      "description": "How a zip of images and a manifest table becomes ``pixels(image, y, x, channel)``.\n\nThe manifest has one row per image: the image path in the zip and, optionally, an id, a\nclass label and a train/test split. The images must have 8 bits per channel and one\nsize, unless ``size`` resizes them. Every byte is decoded inside the sandboxed ingest run.",
      "examples": [
        {
          "label_column": "class",
          "manifest": "manifest.csv",
          "path_column": "file",
          "split_column": "split"
        },
        {
          "id_column": "id",
          "label_column": "label",
          "manifest": "labels.parquet",
          "max_images": 5000,
          "mode": "rgb",
          "resize": "crop",
          "size": [
            32,
            32
          ]
        }
      ],
      "properties": {
        "columns": {
          "anyOf": [
            {
              "items": {
                "maxLength": 256,
                "minLength": 1,
                "type": "string"
              },
              "maxItems": 64,
              "type": "array"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Other columns kept as coords (default: all others)",
          "title": "Columns"
        },
        "dim": {
          "default": "image",
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Dim",
          "type": "string"
        },
        "id_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Unique labels of the image dim (default: the row number)",
          "title": "Id Column"
        },
        "label_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Class labels: the text variable 'label(image)', role target",
          "title": "Label Column"
        },
        "manifest": {
          "default": "manifest.csv",
          "description": "Path of the manifest in the zip",
          "maxLength": 256,
          "minLength": 1,
          "title": "Manifest",
          "type": "string"
        },
        "max_images": {
          "default": 20000,
          "maximum": 100000,
          "minimum": 1,
          "title": "Max Images",
          "type": "integer"
        },
        "mode": {
          "default": "auto",
          "description": "auto: rgb when an image has colour, else gray",
          "enum": [
            "auto",
            "gray",
            "rgb"
          ],
          "title": "Mode",
          "type": "string"
        },
        "path_column": {
          "default": "path",
          "description": "Image paths relative to the zip root",
          "maxLength": 256,
          "minLength": 1,
          "title": "Path Column",
          "type": "string"
        },
        "resize": {
          "default": "stretch",
          "description": "How images are resized (only with size)",
          "enum": [
            "stretch",
            "crop",
            "pad"
          ],
          "title": "Resize",
          "type": "string"
        },
        "size": {
          "anyOf": [
            {
              "maxItems": 2,
              "minItems": 2,
              "prefixItems": [
                {
                  "type": "integer"
                },
                {
                  "type": "integer"
                }
              ],
              "type": "array"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "(height, width) to resize every image to; each 1..4096",
          "title": "Size"
        },
        "split_column": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Train/test assignment: the text coord 'split(image)'",
          "title": "Split Column"
        },
        "var": {
          "default": "pixels",
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Var",
          "type": "string"
        }
      },
      "title": "ImageFolderOptions",
      "type": "object"
    },
    "IngestAssembly": {
      "additionalProperties": false,
      "description": "A dataset version built from several uploads (M4): every source is ingested in the\nsame sandboxed run (inputs ``upload.<i>``, ``i`` = the source's index), then combined\nper node. The result is a tree dataset unless every source lands in the root.",
      "examples": [
        {
          "sources": [
            {
              "combine": "concat",
              "concat_dim": "year",
              "concat_label": 2024,
              "mapping": {
                "format": "netcdf"
              },
              "node": "/",
              "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p0"
            },
            {
              "combine": "concat",
              "concat_dim": "year",
              "concat_label": 2025,
              "mapping": {
                "format": "netcdf"
              },
              "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p1"
            },
            {
              "mapping": {
                "format": "netcdf"
              },
              "node": "/instrument",
              "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p2"
            }
          ]
        }
      ],
      "properties": {
        "sources": {
          "items": {
            "$ref": "#/$defs/AssemblySource"
          },
          "maxItems": 64,
          "minItems": 1,
          "title": "Sources",
          "type": "array"
        }
      },
      "required": [
        "sources"
      ],
      "title": "IngestAssembly",
      "type": "object"
    },
    "IngestMapping": {
      "additionalProperties": false,
      "description": "How an upload becomes a labelled dataset (Meridian ``CoordToColumns`` style).\n\nTabular formats need a mapping; ``zarr_zip``/``netcdf`` are self-describing and may omit\n``dims``/``vars`` (with groups, they rename/select per node by name). A raw ``hdf5``\nfile (M4) needs ``hdf5`` (:class:`Hdf5Mapping`) instead of ``dims``/``vars``. Parsing\nhappens only inside a sandboxed ingest run.\n\nWith the ``mff`` preset the long table is first pivoted to one column\nper ``VariableName`` (plus ``Period`` and the key columns); ``vars`` and ``wide`` then\npick and rename those variables (``column`` = a ``VariableName`` value) and ``dims`` may\nrefine the dims the preset derives (``column`` = an MFF column). Without ``vars`` and\n``wide`` every variable is kept under its own name.",
      "examples": [
        {
          "dims": {
            "date": {
              "column": "date",
              "spec": {
                "freq": "D",
                "kind": "time",
                "policy": "extend"
              }
            },
            "market": {
              "column": "market",
              "spec": {
                "kind": "categorical"
              }
            }
          },
          "format": "csv",
          "vars": {
            "revenue_local": {
              "column": "revenue",
              "role": "target",
              "units": "local"
            },
            "spend_local": {
              "column": "spend",
              "role": "spend",
              "units": "local"
            }
          }
        },
        {
          "dims": {
            "market": {
              "column": "market",
              "spec": {
                "kind": "categorical"
              }
            },
            "week": {
              "column": "week",
              "spec": {
                "freq": "W-SUN",
                "kind": "time",
                "policy": "extend"
              }
            }
          },
          "format": "csv",
          "vars": {
            "revenue": {
              "column": "revenue",
              "role": "target",
              "units": "EUR"
            }
          },
          "wide": {
            "spend": {
              "columns": {
                "search": "search_spend",
                "social": "social_spend",
                "tv": "tv_spend"
              },
              "dim": "channel",
              "role": "spend",
              "units": "EUR"
            }
          }
        },
        {
          "format": "mff",
          "mff": {
            "freq": "W-MON"
          },
          "preset": "mff"
        },
        {
          "format": "parquet",
          "meridian": {
            "kpi": "revenue",
            "media": [
              "tv_impressions"
            ],
            "media_channels": [
              "tv"
            ],
            "media_spend": [
              "tv_spend"
            ]
          },
          "preset": "meridian"
        },
        {
          "format": "hdf5",
          "hdf5": {
            "coords": [
              {
                "dim": "row",
                "spec": {
                  "kind": "float",
                  "ordered": true
                },
                "start": 0.0,
                "step": 0.05
              }
            ],
            "vars": [
              {
                "dims": [
                  "row",
                  "col"
                ],
                "name": "image",
                "node": "/",
                "path": "/observations/image",
                "units": "counts"
              },
              {
                "dims": [
                  "prow",
                  "pcol"
                ],
                "name": "psf",
                "node": "/psf",
                "path": "/psf/kernel"
              }
            ]
          }
        },
        {
          "format": "image_folder",
          "images": {
            "label_column": "class",
            "manifest": "manifest.csv",
            "path_column": "file",
            "split_column": "split"
          }
        },
        {
          "format": "safetensors",
          "weights": {
            "max_parameters": 50000000
          }
        }
      ],
      "properties": {
        "dims": {
          "additionalProperties": {
            "$ref": "#/$defs/DimMapping"
          },
          "maxProperties": 32,
          "title": "Dims",
          "type": "object"
        },
        "format": {
          "enum": [
            "csv",
            "parquet",
            "mff",
            "zarr_zip",
            "netcdf",
            "hdf5",
            "image_folder",
            "safetensors"
          ],
          "title": "Format",
          "type": "string"
        },
        "hdf5": {
          "anyOf": [
            {
              "$ref": "#/$defs/Hdf5Mapping"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "M4: the arrays, dims and coords of a raw HDF5 upload (format 'hdf5')"
        },
        "images": {
          "anyOf": [
            {
              "$ref": "#/$defs/ImageFolderOptions"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "v0.14: options of an 'image_folder' upload (defaults when left out)"
        },
        "layout": {
          "default": "dense",
          "enum": [
            "dense",
            "long"
          ],
          "title": "Layout",
          "type": "string"
        },
        "meridian": {
          "anyOf": [
            {
              "$ref": "#/$defs/MeridianColumns"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "The meridian preset's columns"
        },
        "mff": {
          "anyOf": [
            {
              "$ref": "#/$defs/MffOptions"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Options of the mff preset"
        },
        "preset": {
          "$ref": "#/$defs/IngestPreset",
          "default": "none",
          "description": "M3: a file contract instead of (or refined by) the column mapping"
        },
        "vars": {
          "additionalProperties": {
            "$ref": "#/$defs/VarMapping"
          },
          "maxProperties": 1024,
          "title": "Vars",
          "type": "object"
        },
        "weights": {
          "anyOf": [
            {
              "$ref": "#/$defs/WeightsOptions"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "v0.15: options of a 'safetensors' upload (defaults when left out)"
        },
        "wide": {
          "additionalProperties": {
            "$ref": "#/$defs/WideVarMapping"
          },
          "description": "M3: variables read from one column per label of a dim (wide tables)",
          "maxProperties": 64,
          "title": "Wide",
          "type": "object"
        }
      },
      "required": [
        "format"
      ],
      "title": "IngestMapping",
      "type": "object"
    },
    "IngestPreset": {
      "enum": [
        "none",
        "mff",
        "meridian"
      ],
      "title": "IngestPreset",
      "type": "string"
    },
    "JsonValue": {},
    "MeridianColumns": {
      "additionalProperties": false,
      "description": "Google Meridian's ``CoordToColumns`` contract for a wide CSV/Parquet table.\n\n``media``/``media_spend`` (and ``reach``/``frequency``/``rf_spend``) are parallel lists:\nthe *i*-th column of each belongs to the *i*-th entry of ``media_channels``\n(``rf_channels``). ``geo`` is omitted for a national model; ``kpi_type`` says whether\nthe KPI is revenue (ROI in currency) or a count (``revenue_per_kpi`` converts it).",
      "examples": [
        {
          "controls": [
            "competitor_sales",
            "sentiment"
          ],
          "geo": "geo",
          "kpi": "conversions",
          "kpi_type": "non_revenue",
          "media": [
            "tv_impressions",
            "search_impressions"
          ],
          "media_channels": [
            "tv",
            "search"
          ],
          "media_spend": [
            "tv_spend",
            "search_spend"
          ],
          "population": "population",
          "revenue_per_kpi": "revenue_per_conversion",
          "time": "time"
        }
      ],
      "properties": {
        "controls": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Controls",
          "type": "array"
        },
        "freq": {
          "anyOf": [
            {
              "maxLength": 16,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": "W",
          "description": "Time frequency (pandas alias)",
          "title": "Freq"
        },
        "frequency": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Frequency",
          "type": "array"
        },
        "geo": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": "geo",
          "title": "Geo"
        },
        "kpi": {
          "default": "kpi",
          "maxLength": 256,
          "minLength": 1,
          "title": "Kpi",
          "type": "string"
        },
        "kpi_type": {
          "default": "revenue",
          "enum": [
            "revenue",
            "non_revenue"
          ],
          "title": "Kpi Type",
          "type": "string"
        },
        "media": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Media",
          "type": "array"
        },
        "media_channels": {
          "items": {
            "type": "string"
          },
          "maxItems": 128,
          "title": "Media Channels",
          "type": "array"
        },
        "media_spend": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Media Spend",
          "type": "array"
        },
        "non_media_treatments": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Non Media Treatments",
          "type": "array"
        },
        "organic_media": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Organic Media",
          "type": "array"
        },
        "population": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Population"
        },
        "reach": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Reach",
          "type": "array"
        },
        "revenue_per_kpi": {
          "anyOf": [
            {
              "maxLength": 256,
              "minLength": 1,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Revenue Per Kpi"
        },
        "rf_channels": {
          "items": {
            "type": "string"
          },
          "maxItems": 128,
          "title": "Rf Channels",
          "type": "array"
        },
        "rf_spend": {
          "items": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxItems": 128,
          "title": "Rf Spend",
          "type": "array"
        },
        "time": {
          "default": "time",
          "maxLength": 256,
          "minLength": 1,
          "title": "Time",
          "type": "string"
        }
      },
      "title": "MeridianColumns",
      "type": "object"
    },
    "MffOptions": {
      "additionalProperties": false,
      "description": "The MFF (\"model feature format\") long-table preset.\n\nOne row per ``(period, geography, product, campaign, outlet, creative, variable)`` with\nthe value in ``value_column``. ``Period`` becomes the time dim; ``VariableName`` values\nbecome data variables. The rest follows mmm-framework's MFF semantics (its\n``MFFLoader``), so both read a file the same way:\n\n* ``Geography`` and ``Product`` become dims when they hold more than one level. A blank\n  label (or ``National``, ``Total``, ``All``) is the coarser level: a variable given\n  only there (national TV) is spread across the levels - equal weights, or the KPI's\n  share of each level (``allocation``; mmm-framework's defaults: geographies equally,\n  products by sales). A variable may not mix the two levels.\n* ``Campaign``, ``Outlet`` and ``Creative`` are breakdowns: rows that differ only there\n  are summed, unless the column is listed in ``split`` (then it becomes a dim).\n\n``combine`` joins several key columns into one dim (``{\"geo\": [\"Geography\", \"Product\"]}``\ngives labels such as ``US|Watches``) - a model with one panel dim can then fit a\ngeography x product panel. A combined dim is created even when it has a single label,\nso ``{\"geo\": [\"Geography\"]}`` keeps a national table's geo dim.",
      "properties": {
        "allocation": {
          "additionalProperties": {
            "enum": [
              "equal",
              "kpi"
            ],
            "type": "string"
          },
          "description": "v0.5: how a variable given only at the coarser level (a blank or 'National' label) is spread across the levels of a key column: equally, or by the KPI's share of each level (the variable with role 'target', or 'kpi')",
          "title": "Allocation",
          "type": "object"
        },
        "combine": {
          "description": "M3: dim -> key columns whose labels are joined ('|') into that one dim",
          "maxProperties": 4,
          "patternProperties": {
            "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$": {
              "items": {
                "type": "string"
              },
              "type": "array"
            }
          },
          "title": "Combine",
          "type": "object"
        },
        "freq": {
          "anyOf": [
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Pandas offset alias; inferred when omitted",
          "title": "Freq"
        },
        "key_columns": {
          "additionalProperties": {
            "type": "string"
          },
          "description": "Key column -> dim name",
          "title": "Key Columns",
          "type": "object"
        },
        "kpi": {
          "anyOf": [
            {
              "maxLength": 256,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "v0.5: the KPI variable of 'kpi' allocation",
          "title": "Kpi"
        },
        "missing": {
          "default": "allow",
          "enum": [
            "forbid",
            "allow",
            "mask"
          ],
          "title": "Missing",
          "type": "string"
        },
        "period_column": {
          "default": "Period",
          "title": "Period Column",
          "type": "string"
        },
        "roles": {
          "additionalProperties": {
            "type": "string"
          },
          "description": "Variable -> role",
          "title": "Roles",
          "type": "object"
        },
        "split": {
          "description": "v0.5: breakdown key columns (Campaign, Outlet, Creative) kept as dims instead of summed",
          "items": {
            "type": "string"
          },
          "maxItems": 3,
          "title": "Split",
          "type": "array"
        },
        "time_dim": {
          "default": "date",
          "title": "Time Dim",
          "type": "string"
        },
        "units": {
          "additionalProperties": {
            "type": "string"
          },
          "description": "Variable -> units",
          "title": "Units",
          "type": "object"
        },
        "value_column": {
          "default": "VariableValue",
          "title": "Value Column",
          "type": "string"
        },
        "variable_column": {
          "default": "VariableName",
          "title": "Variable Column",
          "type": "string"
        }
      },
      "title": "MffOptions",
      "type": "object"
    },
    "Role": {
      "description": "What a variable means to a model.",
      "enum": [
        "target",
        "covariate",
        "media",
        "spend",
        "control",
        "index",
        "mask",
        "constant",
        "weight",
        "exposure"
      ],
      "title": "Role",
      "type": "string"
    },
    "VarMapping": {
      "additionalProperties": false,
      "description": "A data variable read from one column (dtype is inferred during ingest).",
      "examples": [
        {
          "column": "revenue",
          "role": "target",
          "units": "EUR"
        }
      ],
      "properties": {
        "column": {
          "maxLength": 256,
          "minLength": 1,
          "title": "Column",
          "type": "string"
        },
        "description": {
          "anyOf": [
            {
              "maxLength": 2000,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Description"
        },
        "missing": {
          "default": "forbid",
          "enum": [
            "forbid",
            "allow",
            "mask"
          ],
          "title": "Missing",
          "type": "string"
        },
        "role": {
          "$ref": "#/$defs/Role",
          "default": "covariate"
        },
        "units": {
          "anyOf": [
            {
              "maxLength": 64,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Units"
        }
      },
      "required": [
        "column"
      ],
      "title": "VarMapping",
      "type": "object"
    },
    "WeightsOptions": {
      "additionalProperties": false,
      "description": "A safetensors file of pretrained network weights (read in the sandboxed ingest run).\n\nThe file becomes a weights dataset: ``weights(param)`` float32 plus the tensor table in\nthe ``mapi_weights`` attr. Models and steps read it as a side table.",
      "examples": [
        {
          "max_parameters": 50000000
        },
        {
          "arch": {
            "blocks": [
              1,
              1,
              1
            ],
            "channels": 3,
            "kind": "resnet",
            "widths": [
              16,
              32,
              64
            ]
          }
        }
      ],
      "properties": {
        "arch": {
          "anyOf": [
            {
              "additionalProperties": {
                "$ref": "#/$defs/JsonValue"
              },
              "type": "object"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "description": "Architecture JSON, when the file's __metadata__ has no 'mapi_arch'",
          "title": "Arch"
        },
        "max_parameters": {
          "default": 200000000,
          "maximum": 500000000,
          "minimum": 1,
          "title": "Max Parameters",
          "type": "integer"
        }
      },
      "title": "WeightsOptions",
      "type": "object"
    },
    "WideVarMapping": {
      "additionalProperties": false,
      "description": "One variable read from several columns, one per label of ``dim`` (a wide table).\n\n``{\"dim\": \"channel\", \"columns\": {\"tv\": \"tv_spend\", \"search\": \"search_spend\"}}`` gives a\nvariable along ``channel`` (plus the table's other dims) with coords ``tv`` and\n``search``. Several wide variables may share a dim; their labels must then agree.",
      "examples": [
        {
          "columns": {
            "search": "search_spend",
            "social": "social_spend",
            "tv": "tv_spend"
          },
          "dim": "channel",
          "role": "spend",
          "units": "EUR"
        }
      ],
      "properties": {
        "columns": {
          "additionalProperties": {
            "maxLength": 256,
            "minLength": 1,
            "type": "string"
          },
          "maxProperties": 256,
          "minProperties": 1,
          "title": "Columns",
          "type": "object"
        },
        "description": {
          "anyOf": [
            {
              "maxLength": 2000,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Description"
        },
        "dim": {
          "pattern": "^[A-Za-z_][A-Za-z0-9_.\\-]{0,127}$",
          "title": "Dim",
          "type": "string"
        },
        "dim_spec": {
          "$ref": "#/$defs/DimSpec"
        },
        "missing": {
          "default": "forbid",
          "enum": [
            "forbid",
            "allow",
            "mask"
          ],
          "title": "Missing",
          "type": "string"
        },
        "role": {
          "$ref": "#/$defs/Role",
          "default": "covariate"
        },
        "units": {
          "anyOf": [
            {
              "maxLength": 64,
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "default": null,
          "title": "Units"
        }
      },
      "required": [
        "dim",
        "columns"
      ],
      "title": "WideVarMapping",
      "type": "object"
    }
  },
  "$id": "https://schemas.mapi.dev/v1/api-DatasetVersionCreate.json",
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "additionalProperties": false,
  "description": "Turn finished uploads into a new version via a sandboxed ingest run (202).\n\nExactly one of: ``upload_id`` + ``mapping`` (one file), ``assembly`` (M4: several\nfiles combined per node into one version, see :class:`IngestAssembly`), or\n``collection`` (v0.15: an image collection from a manifest and image parts, see\n:class:`~mapi_spec.collection.CollectionSource`).",
  "examples": [
    {
      "mapping": {
        "dims": {
          "date": {
            "column": "date",
            "spec": {
              "freq": "D",
              "kind": "time",
              "policy": "extend"
            }
          },
          "market": {
            "column": "market",
            "spec": {
              "kind": "categorical"
            }
          }
        },
        "format": "csv",
        "vars": {
          "revenue_local": {
            "column": "revenue",
            "role": "target",
            "units": "local"
          },
          "spend_local": {
            "column": "spend",
            "role": "spend",
            "units": "local"
          }
        }
      },
      "message": "Q3 refresh",
      "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p0"
    },
    {
      "assembly": {
        "sources": [
          {
            "combine": "concat",
            "concat_dim": "year",
            "concat_label": 2024,
            "mapping": {
              "format": "netcdf"
            },
            "node": "/",
            "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p0"
          },
          {
            "combine": "concat",
            "concat_dim": "year",
            "concat_label": 2025,
            "mapping": {
              "format": "netcdf"
            },
            "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p1"
          },
          {
            "mapping": {
              "format": "netcdf"
            },
            "node": "/instrument",
            "upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p2"
          }
        ]
      },
      "message": "2024 and 2025 cubes plus the instrument"
    },
    {
      "collection": {
        "manifest_upload_id": "upl_01jaf3d1e2f3g4h5j6k7m8n9p0",
        "options": {
          "label_column": "label",
          "size": [
            64,
            64
          ],
          "split_column": "split"
        },
        "part_upload_ids": [
          "upl_01jaf3d1e2f3g4h5j6k7m8n9p1"
        ]
      },
      "message": "Product photos, October"
    }
  ],
  "properties": {
    "assembly": {
      "anyOf": [
        {
          "$ref": "#/$defs/IngestAssembly"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "description": "M4: several uploads combined into one (tree) version"
    },
    "collection": {
      "anyOf": [
        {
          "$ref": "#/$defs/CollectionSource"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "description": "v0.15: an image collection (a manifest upload and image part uploads)"
    },
    "mapping": {
      "anyOf": [
        {
          "$ref": "#/$defs/IngestMapping"
        },
        {
          "type": "null"
        }
      ],
      "default": null
    },
    "message": {
      "anyOf": [
        {
          "maxLength": 2000,
          "type": "string"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "title": "Message"
    },
    "upload_id": {
      "anyOf": [
        {
          "examples": [
            "run_01jaf3q2x8m9v7k5h4g3f2e1d0"
          ],
          "pattern": "^[A-Za-z0-9][A-Za-z0-9_.\\-]{0,127}$",
          "type": "string"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "title": "Upload Id"
    }
  },
  "title": "DatasetVersionCreate",
  "type": "object",
  "x-mapi-contract-version": "0.2"
}