Iceberg v3 is table stakes now; the lock-in moved upstairs

June's summit season closed an argument we have refereed in every platform selection since 2023. Apache Iceberg v3 reached general availability on Snowflake, with deletion vectors, row lineage, a variant type and default values, and Databricks committed to the same format family across its catalog. The table format question is answered. Open column, open format, every major engine reads it. Whether your team is ready to operate open tables is a separate question, and five questions still decide it.

That changes migration scoring in one specific way: stop paying a premium for format insurance. Two years ago, a defensible fear was choosing the losing side of a format war and paying an exit tax later. That fear is now retired, and any vendor argument still built on it is dated inventory.

The switching costs did not disappear. They moved up the stack. The Open Semantic Interchange standard now counts more than 50 vendors, which tells you exactly where the vendors believe the next lock-in surface sits: semantic models, catalogs and the agent layers reading them. Formats converge when differentiation stops paying; the differentiation budget went upstairs.

What we now put in a selection scorecard, in place of format risk: where semantic definitions live, in what format they export, and what leaves with you on the day you go. Ask each vendor for the exit story in writing. A platform confident in its product will describe the exit plainly; a platform confident in your switching costs will describe the roadmap instead.

The deliverable version: one page, three columns. What is open (the tables), what is portable with work (the semantic layer, if you keep it in the interchange format), what is hostage (whatever has no export path). Score the third column, not the first. The first one the industry just settled for you.