Skip to content
Pawel Works

05 · Banking

Credit-risk controls and recon

Independent scoring and dataset validation, AML classification at volume, accounting reconciliation. Python and SAS controls. A grain you can recompute two ways.

Service · Validate the book. Name the exception.

When this is the job

The scorecard and the book do not agree. Outliers sit in the tail. An AML flag is a person in a queue with no written criterion. Finance cannot tie the provision. Nobody escalates because nobody named the exception.

How it runs

Start at the grain. Reproduce the figure two ways. Isolate scoring drift from data-quality failure from policy mismatch. Write the exception. Escalate against a criterion, not a feeling. Document what remains unverified.

  1. L1

    Book

    Exposure summed two ways. Integer. If they differ, stop.

  2. L2

    Score

    Independent of the engine that produced it. Outliers named.

  3. L3

    Flag

    AML against a written rule. Volume, not a story.

  4. L4

    Escalate

    A criterion. A queue. A name on the gate.

What you leave with

A book that ties. Exceptions named. AML flags without identity theatre. A handover that says what the core will not give you.