Southwest lost $1.2 billion because its crew scheduling software kept optimizing an airline that no longer existed. The part most people miss: the solver didn't crash. It worked exactly as designed. Legacy crew schedulers run on batch cycles. They snapshot the network, freeze time, and compute the cheapest valid schedule. Fine 350 days a year. But during a cascade, the network changes every few minutes — crews move, connections break — and the "optimal" plan is invalid before anyone sees it. Then the data black hole opens. Crews stranded at outstations call in to report where they are; hold times stretch into hours. The solver needs certainty — "Captain Smith is at Gate B7" — but during a meltdown, certainty doesn't exist. So it optimizes a phantom airline. That's Southwest in December 2022, and Spirit in July 2024, when its system created conflicting assignments for 43% of available crews — a $50–100M day. Since October 2024 it's worse: DOT's rule turns every 3-hour delay into an automatic cash refund. For a 300-flight carrier, one bad day — 50 flights past 3 hours, 150 passengers each — is $2.1M in mandatory refunds before you've paid a single hotel voucher. Our take: you don't have to rip out Jeppesen or IBS. You need an ML layer that reasons about probable crew positions instead of demanding a certainty that isn't there. If you've worked an ops control center during a cascade — what broke first, the software or the crew tracking? #AirlineOperations #IROPS