v4.296.0 committed BoardReminder alone. `Hook::workMovedOn` — the method it
calls — stayed in the working tree, so every stop under a profile fataled with
an undefined method and the routine silently stopped printing. The hook is
non-blocking, so nothing failed loudly: a discipline just vanished.

The add that was meant to carry both named a path that does not exist, and git
refuses the whole pathspec when one member is unmatched. The error saying so
was suppressed by a `2>/dev/null` on the same line.

Also in here, from a review of the same thread:

The stop question now pays its debt by RECORDING rather than by the clock
running on. It paced against a monotonic count, so it fired every stretch
however much had been written down — the metronome this thread exists to
remove, twenty lines from the sibling that already did it correctly.

The routine spends its mark LAST, once it knows it has a routine to say. It
marked first and then discovered there was no profile, burning the first
reading — which is owed unconditionally — on a stop that said nothing.

`Hook::workMovedOn` takes the work count as an argument. Reaching into the
journal from the hook base tied every hook to one concrete source and closed a
namespace cycle; the caller knows what work means for its own signal.

And `commandments upgrade`: move the package, sync, and re-prepare every lane
the board does not show as live. `--check` names both the installed and the
available version, and exits non-zero when availability could not be measured,
because "nobody found out" must not reach a script as "up to date".