If there's a ton of work done at the various checkpoints, the pace is slow but maybe the length of the line isn't very far?
Also a bit of an outdated mental model. Had the privilege of touring the tesla factory recently... one neat tidbit that stuck with me is the host described "the line" more like a river with many tributaries joining in all along the watershed.
So if you wanted to have a non-softball engineering discussion about "the line", you would ask questions like which tributaries are the bottlenecks, how much parallelism is there, are improvements marginal or do you wholesale upgrade entire sublines at once, what's the caching strategy, etc.