The €908 that decomposed to exactly zero

The €100 Lakehouse — 8/12

May 2026: the new dashboard was €908 short of the legacy one. I refused to call it "2% drift".

So I decomposed it, athlete by athlete, through a real identity bridge — 1,377 rows, not fingerprinting — until the delta had names:

→ −844.00: 12 payments from 12 athletes that don't exist in the new platform at all
→ −64.00: one athlete whose money IS there, consolidated into a €128 payment dated April
→ 0.00: the legacy split a €228 transfer into 3×76; the platform stores one row
→ 0.00: a demo fixture the legacy never had — +80.00 in gold, rightly excluded by the served mart

Residual: exactly 0.00. Even the coincidence got explained — two unrelated €80 items cancelling each other out.

The report also states what I don't know: residual 0.00 in 7 of 13 months, never above €110 — and ~€12,869 before 202501 declared NOT KNOWN. The explained verdict stops at 202501.

One more thing: remove the ghost-bridge athletes and 5 of 13 months snap to exactly 0.00. A year of "systematic 1–2% noise", one mechanism.

It wasn't noise — it was a pointer.

Next week: who the €844 belonged to.

What's your tolerance threshold for "close enough" in financial data — and have you ever decomposed a small delta all the way to zero?

Databricks · dbt · Airflow · Terraform · AWS

P.S. New tech post every Wednesday.

#100EuroLakehouse #DataQuality #AnalyticsEngineering