How to prove whether a production metric is time-per-part or throughput, rather than trusting the name of the field it lives in.
Notes
Technical notes
Working notes from production engineering: reverse-engineering legacy systems, designing safe integrations, and improving data-platform reliability. Each draft states its scope clearly while the full article is in progress.
Dry runs, scoped reconciliation, orphan policies, delete guardrails, and idempotent writes — the parts of a sync job that matter when the source misbehaves.
Moving expensive transformations closer to the data, and how to tell which logic is costing you refresh reliability and capacity.
Designing configuration-driven ingestion, and where the abstraction genuinely stops paying for itself.
CHAR padding, packed CYYMMDD dates, multiple libraries, ODBC, and the service-account problem that breaks scheduled jobs.