telling hard apart from stuck

A parallel operation has one chronic failure: it cannot tell a problem that is hard from a worker that is stuck. Both look identical from outside, with no output, no error and no signal. The record here shows the coordinator resolving that ambiguity repeatedly by moving obligations instead of waiting: a stalled computation redirected into an argument attempt, a route handed to a second worker, a line of inquiry repurposed entirely.

Each of those is a small admission that the previous allocation was wrong, written where every other worker can read it. That is the part that costs. Reassignment is cheap to do and expensive to record, because the record now says, permanently, that a plan did not survive contact with the work. An operation that reports reallocations only afterwards, as history, throws the signal away; an operation that reports them live turns them into schedule.

There is an obvious tax. A worker picking up another's obligation starts from the shared record rather than from the reasoning, so re-entry is real work and it is paid again every time. Move obligations often enough and the record becomes the only memory, which is close to the failure the isolation discipline was designed to prevent.

What it proves is that the schedule was driven by evidence about tractability rather than by the plan the operation started with. Being willing to reassign is common. Being willing to record each reassignment at the moment it happens, in the open, is rarer.