Export a dashboard (or one page) as one workbook, a sheet per data panel.

Exports every data-bound panel of the dashboard, or of one page (pageId), as ONE XLSX workbook with one sheet per panel (DVT-4175), or as a zip of one CSV per panel (format: csv). Panel queries are read from the STORED spec; the client sends ids and filter params, NEVER SQL (unlike POST /v1/data/export). Each distinct data source is authorized and resolved exactly as POST /v1/data/query does it (tenant check, role mapping, per-user OAuth grants, uploaded-file URLs, decrypt audit), and every panel query consumes the same per-org query rate limits. Results are never cached, the request is recorded as a data.export audit entry (not written when every panel failed), and it requires data:query AND data:export plus dashboard:read on this dashboard. Render-session tokens cannot reach it.

Scope: with pageId omitted the export covers the root panels plus every non-hidden page; with pageId set, that page. Sheets follow page order then panel order; a sheet is named after the panel title (falling back to its id), made Excel-legal, de-duplicated with " (2)", " (3)" and cut to 31 characters. Panels that are not data-bound (text, filter, action-button, container ...) are skipped silently. Each sheet gets the same formatting a single-panel export gets (labels, number/date formats, bold frozen header, autofilter, widths; tables also carry their conditional formatting and theme colors).

params carries the active filter / drill state and is merged into every panel's query exactly as POST /v1/dashboards/{id}/email does: it overrides the keys a panel itself declares in data.params, and a key no panel in scope declares is a 422 that names it. include narrows the export to the listed panel ids (still in spec order); every id must be a data-bound panel in scope.

A panel that cannot be loaded does NOT fail the export: its sheet (or CSV) holds one cell, "This panel could not be loaded: <reason>", with a fixed-vocabulary reason (never raw warehouse or resolver text; e.g. a source needing the viewer's account connection reads "connect your account for this data source"), and its id is listed in X-DVT-Failed-Panels. A panel cut at the engine's row cap carries a comment on A1 of its sheet and is listed in X-DVT-Truncated-Panels.

Bounds: at most 25 panels; a per-export size limit on data cells (rows x columns) across the whole workbook (50,000 on Gallery and Enterprise, 200,000 in the Snowflake Native App). A panel that would overshoot the remaining cell budget is cut to it and listed as truncated; once the budget is spent the remaining panels are NOT queried, hold the text "Skipped: this export reached its size limit..." and are listed in X-DVT-Skipped-Panels. Panels are queried sequentially; no query starts more than 18 s after the request arrived and the whole loop ends at 20 s, so a panel not reached in time is reported as failed ("timed out before this panel ran"). Only one dashboard export runs per organization at a time, plus a small per-server cap on concurrent exports (a refused one gets 429 with Retry-After: 5). If every exported panel failed the response is 502 and no file is produced.