Skip to content
Pawel Works

08 · Insurance · risk data

Cross-source data quality

Consistency rules across sources. Incomplete and aberrant records named before reporting. Parameterised investigation applications. A quality gate in front of the dashboard.

Service · Name the bad row. Then the book.

When this is the job

Three systems, three versions of the same entity. Incomplete rows slip into risk reporting. Aberrant values have no rule. Investigation is a new extract every time. The dashboard inherits a mess that should have died in staging.

How it runs

Write the consistency rules. Detect incomplete and aberrant records against those rules, not against a feeling. Parameterise the investigation cut so the same logic runs tomorrow. Gate activation: nothing reports until the quality plate opens. Document what the source will never give you.

  1. L1

    Rule

    Consistency between sources. Written. Not a vibe.

  2. L2

    Detect

    Incomplete. Aberrant. Boundary. Named.

  3. L3

    Explain

    A flag without a sentence is not a flag.

  4. L4

    Gate

    Nothing activates until the plate opens.

What you leave with

A quality gate the report cannot bypass. Flags with a written explanation. A parameterised investigation application. The book is cleaner because the rows were named.