Where the exports stop

Delivered reporting queries the same pod that processes invoices and payroll. An unbounded export is a denial of service against your own close, so Oracle bounds them: row limits on screens, row and cell limits on spreadsheet exports, line limits on scheduled deliveries, memory guards on long queries. In a hosted service, administrators cannot raise them; that this is so is practitioner-observed, and the enhancement requests asking for it sit open for years.

The mistake is not the caps. It is teams discovering them one truncated file at a time, then building chunk-by-date-range rituals on a tool that was never the right vehicle. Volume belongs on the extract service and the warehouse; the table below is for knowing where the walls are until you get there.

Two practitioner notes belong beside the table. Figures shift across quarterly releases, which is why each row carries its source class rather than a citation to a moment in time. And a recent release removed the bulk extract logs from the interface, so diagnosing a short file now means reading manifest counts, one more reason the completeness habit below matters.

The unbounded path is the one built for it: the bulk extract service into your own storage, landed raw and modeled in your warehouse, where a query competes with nobody's invoice processing. What that path does and does not carry is its own page, what the extract service does not carry, and the toolchain choice is the reporting tools guide.

Whatever the path, one habit converts every silent ceiling into a loud one: count at the source, count what landed, and fail the run on a difference. Truncation, trimming, and splits all fall to the same check, which is the discipline this catalog applies from payables ties to payroll. Sample figures above are caps, not client data.