Where to begin. I think I'm just going to start by writing general thoughts, and hopefully they'll somehow mold themselves in to a coherency.
Game physics.
One difference in this game from the original is that going off track slows you down. Not too much though. In RR1 as long as you kept the accelerator down, you maintained speed, it just prevented you from accelerating. However, like everything, this can be used to your advantage.
Rear-ending a car instantly slows you down to that car's speed. If the speed difference is great enough, it will actually wipe you out. I was happy to see that didn't rear-end any cars in your run, as was *very* common and usually completely unnecessary in the published RR1 run. But that's just one thing.
Side swiping a car doesn't slow you down at all, but does push you quite a ways away from the other car. This can be used and abused while going around corners, which you did, but not nearly as much as possible. I'll get to that later.
Hard braking slows you down a lot (makes sense, eh?). Usually (I'm hesitant to say always), it can and should be avoided.
Other cars are triggered when you reach a certain point in the track. Many of them change lanes, sometimes to a set pattern. Other cars (I deem them 'enemies') will deliberately change lanes in front of you and/or attempt to run you off the track. Many times their position can be manipulated right before a turn so that you can side-swipe them without having to slow down.
Hit boxes are rather small. It is possible to pass a car in the outside lane *on the outside* even if there are poles along the road. This can help set you up for an oncoming curve. It's even possible to pass in between two cars right next to each other. Not while traveling on a straightaway, but while going around a curve. What happens is you end up side swiping the outside car, which throws you into the adjacent car, which throws you into the outside car, etc. If you time it right, the last collision will be with the outside car, which will place you high on the inside of the curve. If you time it wrong, you'll end up off the track... but this shouldn't be a problem for a TAS. This process doesn't slow you down at all.
The runs.
While watching Saturn's RR1 run, I wasn't convinced at all that it was optimal. In fact, there were many tracks which I was quite sure were not. This run doesn't fare much better.
In general, it is a much better strategy to go off the track a little bit before a sharp turn than to have to brake in the middle of it. Sometimes this isn't possible due to poles, but often it is. The curve right around frame 46000 is a prime example. Open road, no cars, no poles, and then a sharp curve that you end up holding the break down almost entirely through. Up to this point I had only observed minor mistakes, but this one made me throw my hands up in the air. Not much later, around 50000, you did it again. I really have a hard time understanding this, because in the previous track, I saw you go of the road before a turn when it was unnecessary.
Another great alternative to breaking is to let off the accelerator a bit. Around frame 71000 you bounce of a car to maintain speed. But then you immediately brake afterwards, because it was a very sharp, very long turn. A much better way to have handled this, would have been to let off the accelerator after the first side swipe, so that you were travelling the same speed as the other car, and then continually side swipe it for the entire curve, or as many times as necessary.
Something you noticed very late in the run is that at the finish line, running into one of the flags is a lot faster than braking down. I'm not sure if this can only be done on tracks which end on a curve, but if this can be done for all tracks, it definitely should be (side swipe, perhaps?).
I don't mean to discourage you at all. This run isn't poorly done, but I think the low re-record count shows. I think a well done Rad Racer II run would be a good replacement for the existing RR1 run. However, I feel that this run could be improved quite a bit, which is why I'm voting no.
print reduce(lambda x,p:p/2*x/p+2*10**1000,range(6643,1,-2))
