2026-07-09 –, Singer-Dekkerzaal Language: English
Core Flow is the data platform behind Dutch nature monitoring, designed around one question: what will someone be able to read in fifty years, with no account, no licence, and no us? Everything is stored as open files (Parquet) on S3-compatible object storage and queried directly with DuckDB, so there is no database to outlive and no proprietary layer to replace. Each dataset carries a stable UUID identity, so names and paths can change without breaking references, and data is append-only and historized, so you can ask what was known on any past date. There are no standing servers: scheduled jobs run on the GitLab CI runners we already have. The talk walks through the principles (no vendor lock-in, sovereignty, durability and cost) and how each one shaped the architecture.
Core Flow v1 is the platform we use for Dutch nature data: Natura 2000, vegetation, species and administrative geographies. The talk is aimed at a geo/FOSS audience and focuses on the design choices that let public-interest spatial data survive the decades, not just the next procurement cycle. We treat storage as plain files: Parquet on S3-compatible object storage, queried directly with DuckDB, with nothing proprietary anywhere in the stack you cannot replace. Stable UUID identities sit underneath human-friendly names and paths, which solves a problem every long-lived spatial registry runs into: keeping references intact across renames and reorganisations. Everything is append-only and historized, so the record is reproducible at any past date and corrections supersede rather than erase. And because there are no standing servers (scheduled jobs run on existing GitLab CI runners), the whole thing is cheap and has almost no attack surface. Expect concrete architecture, honest trade-offs, and a clear position: paths are convenience, the UUID is the contract, and open formats are the only durable promise.
