Engineers have a ritual for changes that matter: someone else reads the diff before it ships. Nobody does that for a closed AML case, which is odd, because a closed case becomes a reference for every similar one that follows.
Let’s run the ritual anyway. The “PR” is case-4821. The commit message reads: “Closed: consistent with profile.”
The review comments
Why? The message states a verdict, not a reason. What changed my mind from “unusual” to “fine”?
Evidence? Which documents or earlier cases supported it? I can’t see links.
Scope? Does “consistent” cover this counterparty only, or the whole pattern?
Expiry? Is this still true if the business model changes?
Related PRs? Did this reference earlier cases, and are they linked?
A commit-message standard for closures
Treat it like conventional commits. A title with the verdict. A body with the evidence and what you ruled out. A footer with scope, date, review-by and links to related cases.
That’s the shape of a good AML case memory entry, and it’s the data model behind retrieval you can trust, as we explain in RAG-powered AI that generates evidence. For where this sits in the stack, see the missing layer of AML technology.
A diagram of when to reuse a past decision: Diagram. A walk-through of how a reviewer samples cases: Blogspot walk through.
Over to you
What would your commit message for “Closed: consistent with profile” say?
Should a closure need a second reader, like a PR does, or is that overkill?
Hot take: a closure with no stated scope is a bug. Agree, or fight me?
Brought to you by TechnoXander.
Top comments (1)
My own rule: no closure without a “valid until” line. Do you review closures the way you review code?