Filament is very finicky and a bit difficult to manage. The way it is looping down from the extruder and back up to the winder looks like a drooping length of string, but it doesn’t act that way. It is flexible for the first couple of inches while it is hot, and then sets into a shape which may be a straight length or a bend, depending on what it was doing when it cooled. Then it runs along a path like string, but being semi-rigid it also supports itself along the way.
For the filament to feed consistently, it needs to maintain the same shape as much as possible by keeping the same orientation as it hardens. If the shape of the bend changes near the extruder, then that bend will push the filament around when it gets to the puller, creating another change in the bend at the extruder which pushes the filament when it in turn reaches the puller in a never ending cycle. You can see it in action in this timelapse-
The key here is that the filament pushed back behind the nozzle opening, creating a tighter bend in one spot. When that bend got near the puller it pushed the filament behind the nozzle again. No amount of tweaking on the speed could compensate for that. It only ran about an hour, and then developed a kink that was not going to make it through the puller, so I stopped it.
I also took the fan off halfway through. With the fan, any extra movement of the filament near the nozzle gets quickly hardened into a permanent bend. If it stays soft longer, there is a chance for it to straighten out again before hardening. Cooling would be beneficial if it could freeze the filament right as it exits the nozzle, before it has a chance to move around much,.
Here is another attempt at the loop, but this time I moved the extruder or winder back and forth to make sure the filament kept the same angle coming out of the nozzle no matter how it was pushed around. Eventually the inconsistencies created while guiding it to the puller worked themselves out and it ran steadily for 3 hours until it was time to stop it for the night.