In such a situation I think all tasks should be globally sorted (according to business-determined weighting) by the metrics: - how much do we expect to gain if

In such a situation I think all tasks should be globally sorted (according to business-determined weighting) by the metrics: - how much do we expect to gain if we do it - how much do we expect to lose if we don't - how much (black swan worst case) might we lose if we don't - how easy is it and then programmers should just pick off the top of list, updating the tasks metrics if they subjectively change. That way there are multiple tasks queued but only one in play, and context switching happens when a task changes importance.