Hi Martin,

  *   How large was the whole dataset? As Schematron is usually implemented by
transpiling to XSLT and conventional XSLT (be it 1 or 2 or 3) load the whole
dataset into memory, did you run into memory problems?
Good question.
In this case, the datasets were relatively modest in size (on the order of a
few hundred to a few thousand records per dataset), so I did not run into
memory constraints.
Youbre absolutely right that typical Schematron implementations (via XSLT)
load the entire document into memory, so this approach would need to be
adapted for larger datasets. In a production setting, I would likely:

  *   partition the data and validate in chunks,
  *   apply the same validation rules across those subsets, or
  *   use streaming or iterative approaches where feasible
For this task, the goal was to validate transformation correctness rather than
build a scalable validation pipeline, so the in-memory approach was
sufficient.
That said, I do think the core ideabexpressing business rules declaratively
and validating them against the full datasetbremains applicable even as the
implementation strategy evolves for larger volumes.
Best regards,
Roger