Have you ever been at at that magical point where you have to decide if it's worth putting the effort in to understand the legacy code, or just delete it all and start again?

How could you start again if you don't understand what it is you're replacing?!

Been there quite a few...
Usually I chose the later because it allows to re-structure what you do and especially you do use meaningful names and comments in this case to avoid repeating the first situation....

And to answer the first comment : you don't have to understand broken legacy code to re-implement if you have the requirements cleanly stated

That's how I rewrote from scratch my raytracing engine 4 times and I should really do it again at least once