NY wrote:

If you do any of those functions (crop, resize, rotate, color
adjustment, borders) with any package, you will always have to re-encode
because you are actually changing the data. Cutting out frames is
essentially selectively copying from source to destination, with a bit
of re-encoding around each join.

It's the "cutting out frames" case which can be done without
re-encoding. You have to cut on GOP boundaries to make it work.

And at least one video standard, there's a software that
allows cutting any frame, and just the current GOP needs
to be re-worked to meld the ends properly.

You can use FFMPEG to join videos on GOP boundaries.
I've done that with Cinepak encoded content. Being
careful to make each segment a multiple of the GOP
size. (I.e. a 12 frame GOP, a 12000 frame segment,
so one movie chunk has 1000 GOPs in it.)

https://en.wikipedia.org/wiki/Group_of_pictures

"I picture or I frame (intra coded picture) –
a picture that is coded independently of all other pictures."

Those take up more space than the other frames, but
they're also the secret to lossless editing (on roughly
half-second boundaries). This isn't sufficiently
accurate editing for most people - you'd need some
generous fade-to-blacks to work that way.

I've studied the data rate of videos, using ffprobe,
and having it dump packets to an output file. And you
can post-process the file and get size info, to show
the size of an I-frame, versus the others. You'll easily
be able to see the cadence. (Media info tools will tell
you the GOP value, without doing all of that. But this is
part of the fun of studying stuff.)

The largest GOP value is around 600 frames. And that's
not a popular value to try either. Using that is for
bar bets (ugliest, least responsive video).

Paul