How I audited Domo mortgage dataflows with Claude Code
At Domopalooza, I used Claude Code and the Domo API to audit mortgage dataflows that still ran successfully. The live fix log recorded 25 fixes across five ETLs and zero errors applying the changes.
How a successful run produced wrong numbers
In a 2018–2020 Home Mortgage Disclosure Act dataflow, eleven branches of a purchaser_type rule checked the same value. Only the first branch could run. The mapping labeled every row “Not applicable,” including rows that should have identified a mortgage purchaser.
Another rule used Prinicial residence where the data used Principal residence. Downstream filters returned zero matching rows without raising an error. Eight other CASE statements omitted an ELSE branch, so unmatched values became NULL.
What changed during the keynote?
Using the Domo API, I saved corrections to the purchaser codes and residence spelling, added fallback branches, aligned join keys, and guarded three divisions with NULLIF. The on-screen log counted changes applied during the demonstration.
Watch the Domo CEO keynote and AI demonstration · Explore the keynote project on GitHub
What would I look for in another pipeline?
I’d start with rules that can quietly change the meaning of the output. Repeated conditions, filters that match no records, and missing fallback values all deserve inspection even when the job finishes successfully.