The Data Eye
Perceptual training for data-quality & pipeline review
Catch the data-quality and pipeline defects AI assistants introduce — silent type coercions, fan-out joins, broken incremental models — before they corrupt a warehouse.
Syllabus
The Data Eye
Why a query can run perfectly and still be wrong — and how to see it.
Joins & Fan-out
The #1 silent bug: a non-unique join key multiplies rows, then SUM lies.
NULL Semantics
NULL is not zero, not empty, and not equal to itself. It eats results.
Grain & Dedup
What is one row? Get the grain wrong and every aggregate is wrong.
Incremental & Idempotency
Will a rerun duplicate rows? The append that isn’t idempotent.
Types & Precision
FLOAT money, integer division, silent casts — the quiet arithmetic of wrong.
Freshness & Tests
Data contracts: the tests that catch corruption before it ships.
AI Data Failure Modes
When the AI rewrites your query — and quietly changes the grain.