For SaaS teams shipping customer data to warehouses and buckets
Export ContractTreat your export's schema like the API it is.
Describe the export and paste its schema from the last batch and this one. Your browser diffs them against the add-only contract - renamed, removed, retyped and tightened columns break pipelines - ranks the native share against Parquet drops and the cadence, and checks partition keys, small files, a named freshness posture, per-recipient credentials, who pays egress and residency by region. Then the design, and the contract your customers read.
The export and its schema free
Delivery, access, cost and regions free
The checks free
What the page flags
The gates
The camp and the cadence
This batch against the last
Residency by region
Area by area
Your schema section free
This batch's columns with their types and nullability, every change since the last batch with breaking ones marked, and the freshness posture.
Schema (Markdown)
The export design metered
Pricing as soon as the session is ready...
Sign inThe reply could not be read. Here it is as it came back:
Caveats
Caveats for your team
The page's check of this reply
Saved exports
How it works - and what is sourced
The rankings follow the skill. A native warehouse or lake share leads when target accounts concentrate on one shareable platform (the 30-40% line is self-set): nothing is copied, nobody pays egress, no load job introduces drift. Partitioned Parquet on object storage is the universal fallback and the second surface for bucket-only consumers; with no warehouse-resident accounts the share is deleted. Streaming push is only for event-shaped data at sub-minute freshness, with a no-replay warning. The cadence starts at a scheduled full dump and moves up only when a real customer freshness need or the dump's cost forces it; CDC brings delete tombstones and reconciliation with it. One named freshness posture - fixed, best-effort or SLA - is published either way.
The schema diff holds this batch to the add-only contract: a removed, renamed, retyped or newly non-nullable column breaks every consumer reading by name, so it ships as a new column with the old one kept through a dated deprecation window; new columns are appended at the end and nullable. Partitions follow what consumers filter on, files land at roughly 256-512 MB with compaction, recipients read storage directly through per-recipient credentials, someone is named to pay storage, compute and egress, and every regulated region is pinned, papered or not offered. None of this is legal advice.