Google released a tool that manufactures realistic logistics planning problems. What's moving this week isn't how to solve them — it's who supplies the material a solver gets tested against.
Google Research published MilleMiglia, a generator for middle-mile logistics problems — the leg between warehouses, where freight moves in bulk before it reaches local delivery. Real operational data is commercially sensitive and rarely leaves the companies that hold it, so researchers have long had to fall back on artificial cases they built themselves. The same week brought work using large language models to hunt for the conditions under which a cyber-physical system breaks. That one also targets the test material rather than the solver: let a machine produce the cases worth trying.
How good an algorithm looks depends on what you measured it against. When every group brings its own problem set, there is no way to tell whether an improvement was real. A shared generator moves that foundation out of each lab and into the open. And logistics is a domain where solution quality lands on fuel bills and delivery windows, so small gains carry money with them. Whether the generated instances capture what actually makes real routing hard, though, only shows up once people put them to work.
Whether other groups re-measure their own methods on this generator, and whether those numbers get pooled into a shared leaderboard. A tool nobody adopts leaves the comparison problem exactly where it was.