The CSV Analysis Layer

EditorSession.csvAnalysisSnapshot(scope:) is how data leaves the grid. It commits an active cell edit before reading storage, so the input contains the latest edits rather than only the saved file.

let input = try session.csvAnalysisSnapshot(scope: .visibleRows)
let column = try await Task.detached(priority: .userInitiated) {
    try input.numericColumn(at: 0, missingCodes: ["NA"])
}.value

The snapshot is a Sendable value; calculations belong in separate functions or services and run off the main actor. Taking a snapshot is an explicit action, not work performed while drawing, scrolling, or editing each cell.

• .allRows: all records in file order, regardless of sorting/filtering.
• .visibleRows: all records passing the filters, in displayed sort order. This means the filtered dataset, not only the cells currently inside the viewport.
• table: headers and raw strings, including unsaved edits.
• sourceRows: zero-based original record indices, aligned with snapshot rows.
• columnTypes: the column-header variable-type hints, keyed by column index.
• csvText: serialized captured data with headers, suitable for saving as the input to a manually run R script or other analysis tool.
• fingerprint: a content hash of those exact cells, so a result computed from one capture can be told apart from the table as it stands now.
• inputSummary(dataset:scope:): that fingerprint with the shape and timestamp, for attaching to a result.

Columns are addressed by index because headers can be empty or duplicated. numericColumn(at:missingCodes:) and textColumn(at:missingCodes:) preserve row alignment with [Double?] and [String?]. Blanks, whitespace, absent cells in short records, and any declared missing code are missing; other nonnumeric and non-finite values are invalid. Both categories return their source-row indices. NA is not silently classified as missing — pass it in missingCodes to make it so. Paired calculations must handle missingness jointly across columns rather than compacting each column independently.

Type choices currently describe intent only. They do not coerce cells, drop Skip columns, or validate Integer/date/factor semantics. They live with the open grid and are not encoded into the CSV or persisted across reopening.

The preview header's Data menu provides all-row and filtered-row CSV copy actions built on this API.