What portability
actually means
Every data platform claims your data is yours. Very few say precisely what arrives if you leave, which is the only version of the claim worth anything.
Three claims that get confused
Open format. The bytes are in Parquet or Iceberg and anything can read them. Spark, Trino, DuckDB, another warehouse.
Open engine. The software holding the data is open source. It may still use its own storage format, but you can keep running it forever with no vendor involved.
Not locked in. Usually means neither of the above, and is usually asserted rather than specified.
The distinction matters because a vendor can be honest about the second while sounding like the first.
What we actually hand over
- Dagster, with the pipelines
- Airbyte, with the connections
- StarRocks, serving silver and gold
- Iceberg and object storage
- Bronze, silver and gold data
- A manifest and a README
All Apache licensed. No vendor in the loop and no clock running.
- Roles, groups and namespaces
- Row filters and column masks
- Catalog descriptions and owners
- Certification and freshness state
- The lineage graph
- Our deployment configuration
Everything that was metadata for running our software.
Why we publish the right-hand column
Because a promise that hides its exclusions is not a promise, and because the second column is the honest answer to what you were paying for.
The engines are open and yours. The data is open and yours. The governance model, the access policies, the catalog metadata and the semantic layer are software we built, and they stop when the licence stops. Saying that plainly is more useful than "nothing is locked", which nobody can act on.
What that leaves you with
A working substrate. Pipelines that still run, data that is still queryable, engines you can operate indefinitely. What you do not have is the layer that explained what the data meant and who was allowed to see it.
That is the correct outcome commercially, and it is worth understanding on the way in rather than discovering on the way out.
The comparison that matters
With a proprietary warehouse, leaving means an export project: your data is in their format, on their engine, and getting it out is measured in weeks and egress charges. With this architecture, leaving means cancelling a licence while everything keeps running.
That is a real difference and it is worth choosing deliberately. It is just not the same as nothing being locked.