The constraint
Margin was leaking somewhere in payment processing and nobody could point at where. The fallback path had been configured once, correctly, and had drifted out of correctness as the provider changed its pricing.
What we changed
We instrumented the routing decision itself, then rebuilt the flow so the cheapest viable path is selected at request time rather than assumed at configuration time.
Where it landed
All identified leakage recovered, and — more usefully — a cost regression test that fails the build if the routing logic ever silently changes again.
