That was an excellent observation. I like to explain the phenomenon you observed using queuing theory - As time went on, each of you (and you yourself) were abl

That was an excellent observation. I like to explain the phenomenon you observed using queuing theory - As time went on, each of you (and you yourself) were able to estimate tasks very, very accurately. I'd like to think that this was a period of low attrition.. but even if it was not, you were able to model the behavior of a 100 employees (a large enough set) to come up with good estimates. An agile/scrum process is designed to work with tasks of inaccurate estimates - the whole business of story points is designed around that fact. Since your underlying phenomenon changed, the process was no longer optimal. I'm not sure if you experimented with your own "lean" inventions beyond agile - single queues, multiple queues, queues with an artificial stop signal to reduce variance, etc. - but it would have been interesting.