ump_intake

package ump-intake

ump_intake: shared data intake for the UMP POC (spec §6, datasets, extraction and lineage).

Three pure stages plus the content-addressed blob write:

parsing: CSV or XLSX bytes, an inline series or connector rows to a ParsedTable;
mapping: each column to a shared name with a confidence, the reason and alternatives;
panel: apply the mapping, roll platforms up to channels, tie out to printed totals (G-DATA-02) and render the canonical weekly CSV.

Used by apps/onboarding (wizard intake) and apps/data (catalogue upload, connected datasets).

Read the ump-intake guide first for the main entry points.

Modules

Module	What it holds
`ump_intake.blobs`	Content-addressed blob writes and reads through core’s `POST /v1/blobs` (the GCS stand-in), so every write is principal-checked and emits `blob.stored`. The client is any object with `upload_blob` and `get_blob` coroutines (`ump_sdk.AsyncCoreClient`).
`ump_intake.mapping`	Mapping suggestions (stage 2): each column to a shared name (date, KPI id, media id from the registry media tree, control, or ignore), with a confidence, the reason and alternatives.
`ump_intake.panel`	Panel build (stage 3): apply the confirmed mapping, aggregate columns that map to the same media id, tie the extracted totals to the printed ones (G-DATA-02), and render the canonical weekly CSV.
`ump_intake.parsing`	Parsing (stage 1): CSV or XLSX bytes, an inline series, or connector rows to a `ParsedTable`.

Re-exports

Documented on the page of the module that defines them.

Name	Kind	Summary
`CONTROL_HINTS`	constant
`DATE_NAMES`	constant
`DEFAULT_TIME_ID`	constant	The time registry entry used when the caller names none (POC default, proposed).
`EXPOSURE_TOKENS`	constant
`EXTRACTION_RATINGS`	constant
`KPI_ALIASES`	constant
`SPEND_TOKENS`	constant
`TOTALS_TOLERANCE_REL`	constant	§15 G-DATA-02 tie-out tolerance (relative), proposed. Callers may pass their own `tolerance_rel`.
`TOTAL_RE`	constant
`WEEK_ENDING_NAMES`	constant	Headers (after `_norm`) that label a weekly row by its last day (spec §4: convert to week start).
`BlobClient`	class
`ParsedTable`	class
`build_panel`	function	Apply `{column: target}`. Returns the canonical table, tie-out, aggregation notes and fit hints.
`channel_of`	function	The channel-level ancestor of a media id (a platform such as `social.meta` rolls up to `social`).
`convert_week_ending`	function	Spec §4: relabel a week-ending date column by the date each week starts (six days earlier), as a logged step. Returns a new table (the input is left as it was) and the step record. Unreadable dates are kept as they were and counted; a start date that is not the entry’s `week_start` weekday is converted anyway and labelled in the issues, so the caller sees the weeks do not line up with the registry.
`default_rating`	function
`header_index`	function	First row with at least two non-empty, non-numeric cells.
`kpi_ids`	function
`media_tree`	function
`panel_rows`	function	Read a canonical panel CSV back to `(columns, rows)` with numbers as floats and dates as ISO text.
`parse_csv_bytes`	function
`parse_inline`	function	`{dates, kpi, spend: {ch: [...]}, controls: {...}, kpi_name?}` (the adapter SeriesData shape).
`parse_rows`	function	Rows read through a data connection (a list of lists, or of dicts read in `columns` order) to a table, so mapping suggestions and `build_panel` work on connected data exactly as on a file. Blank and printed total rows are separated the same way. `raw_bytes` is the CSV rendering, held in memory only: callers decide whether the copy policy lets it be stored.
`parse_upload`	function
`parse_xlsx_bytes`	function
`preview`	function
`put_blob`	function	Write-once content-addressed file. Returns (hex, `blob://sha256/<hex>`).
`read_blob`	function
`slug`	function
`split_rows`	function	Separate data rows from blank and printed total rows.
`suggest_column`	function
`suggest_mapping`	function
`tie_out`	function	G-DATA-02: extracted totals against printed totals, per source column. The headline pair is the column with the largest relative gap, so the gate reflects the worst column.
`time_entries`	function	The spec §4 time registry: `{id, tz, week_start, grain, fiscal, valid}` entries.
`time_entry`	function	One time registry entry (`DEFAULT_TIME_ID` when `time_id` is None); 422 `SCHEMA_INVALID` if unknown.
`to_csv`	function	CSV text with `\n` line ends; an optional printed `Total` row at the bottom.
`to_date`	function
`to_number`	function