Apple added a form to its dispute flow. Disputes that were started but never finished silently vanished — submitted, never routed, never investigated. The CFPB called it $89 million. Here's the part every bank should sit with: it wasn't a freak bug. It was a reachable state in the workflow with no exit. A dispute landed in (Form A submitted, Form B pending) and no path existed from there to "investigation opened." Under Reg Z §1026.13, many of those messages were valid billing error notices — the kind a creditor must acknowledge in 30 days and resolve within two billing cycles. They just sat in a dead state instead. Apple and Goldman had one integration point between two systems. Most large card issuers have 10–15 systems touching a single dispute: the network portal, case management, ledger, letter generation, bureau reporting. Every API change, every partner integration opens a new path through that maze — and any path can hide a dead state. Chargebacks are projected to hit 337 million a year by 2026 — more paths, more places to lose one. This is the gap nobody automates around: testing checks the paths you thought of. A formal model checker explores every reachable state and proves the invariant holds — every submitted dispute reaches investigation inside 30 days — or hands you the exact path that breaks it. And it compounds: one dual-network dispute can trigger Reg Z, Reg E, Visa VCR (30/70/100 days) and Mastercard timelines at once. Manual design can't guarantee all of them hold on every path. A proof can. The real shift is from "we tested it and it worked" to "we proved it cannot fail." You can pass your last exam and still not be able to prove to the board that the process can't fail — and that's the gap a proof closes. Save this for the next time someone pitches dispute "automation," then ask the question that matters: can you prove it's compliant, or just process it faster? #FormalVerification #BankingCompliance #RegTech #DisputeResolution #CFPB