Comparison
Data diffvsGolden run
Data diff
before merging, you run both versions of the model and see that exactly nine hundred rows differ, all in one region.
Comparing the output of a changed transformation against the current one, row by row. It converts 'I think this refactor is safe' into a list you can walk through, and it catches the accidental cross join that a row-count check would miss. It is expensive on large tables, which is why it is usually run on a sampled or single-partition basis, and that sampling has to be stated when the result is presented as evidence.
Full entry →Golden run
before merging the refactor you run the new model over last month and diff its output against what the old one produced.
A retained known-good output used as the comparison baseline when changing transformation logic. It is how you refactor a two-hundred-line model with any confidence, because the test is not 'does it run' but 'is every row the same as before'. Where rows do differ, the useful discipline is that each difference must be explained rather than accepted in bulk.
Full entry →