Iceberg, Hudi or Delta Lake
Six questions about write patterns, query engines, platform gravity and operational capacity produce a weighted recommendation — and show which answer contributed each point. The reasoning is on the page, so you can disagree with a weight rather than with a verdict.
Answer for your actual situation; the result updates as you go.
Why weights and not a flowchart
Most three-format comparisons end in a decision tree whose first branch is "are you on Databricks?" That is a funnel, not a decision aid. Real projects have four or five factors pulling at once — write pattern, primary engine, existing platform, freshness requirement, how much operational tuning the team can absorb, and tolerance for lock-in — and they routinely disagree. Hence a weighted sum, with every contributing reason shown.
The weights concentrate on three genuine differences. Record-level upserts and CDC favour Hudi: merge-on-read tables plus a record index give the lowest write amplification, and incremental queries are first-class. Engine neutrality and catalog choice favour Iceberg: first-class support in Trino, Athena and Flink, a REST catalog spec that keeps engines swappable, and cloud-managed paths that default to it. Platform gravity favours Delta: on Databricks, Unity Catalog, Photon and DLT are built around it, and another format means giving up platform features.
Two things it deliberately does not decide for you. Cross-format interoperability — Iceberg/Delta conversion, Unity Catalog's Iceberg support — moves fast enough to be worth re-checking the week you decide. And migration cost: once a few hundred tables are in production, "the better format" rarely beats "the format already running".