How the files were split across concurrent groups — one entry per group, never empty. groups.length is the concurrency the run actually used; 1 for watch and single-file runs.

Worth having because the split is not reproducible from the outside: it is recomputed each run from recorded timings and the core count. When a test only fails alongside a particular neighbour, this is what identifies the neighbour — and run({ inputs: result.groups[2].files }) re-runs that bundle on its own.

There is deliberately no URL here. A group's server is closed by the time this result exists, so an address would be a link that never resolves; RunGroup.output is the durable equivalent, still on disk, and it is what --open points a browser at.