What it measures
Take every content word used after a stop. For each use,
find where in the session that word first appeared. If it appeared after
the stop, the kept part brought it in itself. If it appeared before, the
kept part is leaning on something you are about to replace.
01≡
Read the session
The same messages the Rewind picker lists, straight from the file on
your disk.
02▁▂▃▄▅▆▇
Score every stop
One number per possible stop: the share of the kept part's vocabulary
that it did not borrow.
03▁▁▇▇▇
Grade the steps
Each rise is measured against the wiggle of its own curve, so a real
break and a sub-topic are told apart.
04↓
Name the position
A number to scroll to, the message quoted, and a note to paste that
says what to keep exact.
▅▄▃▂▁▇▇▇▇▂▁A break, then a fall
The curve steps up, holds flat, then falls away. The flat run is the
safe zone. Stop anywhere inside it.
▄▃▂▁▇▇▇▇▇▇▇A break, and it holds
It steps up and stays up. The break happened, and nothing since has
needed the past. Anything newer is as safe and saves more.
▇▆▆▅▅▄▄▃▃▂▁One long topic
No step is clear, but the sub-topics are real. It grades them and lets
you choose, rather than pretending there is a clean answer.