Ok I just whatched your demo run. And here's my take.
The business with disallowing pressing A 30 times a second is not arbitrary. Your goals are clearly defined. Beat the game as fast as humanly possible on original hardware. This is a fine goal, but this site is just not about what is humanly possible, so I don't think it could ever be published as a normal run, but as a concept video or something, sure.
Since your goal is to do it as fast as humanly possible, maybe once you master and understand all the techniques you could actually do it on console and record it and send it in to SDA or something, that seems more fitting. Several years ago I have beaten the game in pretty much this same way on console, it took me more like 40 minutes though.
Opinion stuff aside here are some game play related things:
You mentioned skipping the lantern, but I don't think this can be done without UP+DOWN, it won't let you push that thing with zelda if you skip it.
When you get in the 3rd crystal dungeon, it is possible to pull the glitch off much faster. I think you have read some faulty information on how it works (I remember reading something like you have to bounce until the hand comes 7 times or something). But it is really only nessecary to bounce back and forth once. I can't find a link to how it really works, but here is the jist:
If at a certain point in your sword spin you hit the left bumper it caues an offset of 1. If at that point in your sword spin you hit the right bumper it causes an offset of -1. If neither of those happens the offset remains unchanged. If you get 1 then -1 or -1 then 1 it goes back to 0. The glitch works when you leave the room with of an offset of -1 or 1 (can't remember which). If you bump a wall before exiting the room the offset goes back to 0.
Since the glitch can be done faster you won't have to kill any of the mummies.
Anyways best of luck with your run.
About your opinion:
I already answered about SDA and frame-perfection. SDA: They don't want exploration glitches anyway. Frame-perfection: I don't want as fast as humanly possible, I want as fast as possible! As hanzou clarified earlier, the real deal is avoiding left+right/up+down and the floating glitch, I was just not clear enough. But hey, it's my first TAS, you can't expect me to be perfect yet... well, except the submitted video, of course! :P
About the lantern:
Oops, you're right: I can't skip the lantern! Darn! Oh well, it'll be easier! :P
About the bumping:
I haven't checked yet how the bumping glitch technically works. Your info sounds possible, but it's not enough to explain everything: you can trigger two different versions of the bumping glitch, but the wrong one is really really really more easier to get. However it's worth checking out: it could save a lot of time indeed!
About final comment:
Thank you! ;)
About the bumping, my explanation explains that too. The 2 versions of it correspond to the values -1 and 1. Both are glitched but they are offset it different directions only 1 of which happens to work. However this would mean that each has an equal likely chance of happening unless for whatever reason the amount of time allowed to set one of the values is longer during the sword spin than the other.
That being said, I do not really totally understand how it works either, I have never messed with it on emulator. However I am certain it can be done in 1 bounce, and I suspect it does not need to be done bouncing between 2 bumpers and could be done using just one, such as in the 3rd pendant dungeon.
The hardware would check the controller status at one point and ignore anything that happens between two such points.
But you see, I have no reason not to believe that the hardware would interpret both inputs as on during that point.
If you oscillate fast enough, it would seem like you are holding down the buttons to the hardware.
Lets say the hardware samples at 60hz. What if you press the a button at 2000hz. Would the hardware interpret the button being held down? I'm pretty sure it would. Now what if you interleave pressing 2 buttons at 4000hz? Same thing, right? Both buttons should be pressed.
Hmm, after thinking about it for a while, the hardware may capture a noisy signal, especially if your button presses are faster than register setup and hold time...
But even the setup and hold time time increments are quantized, right? You can't work in continuous time when dealing with digital systems; you have to have some overall sampling rate and period to create discrete time. So left+right should be possible, as long as you press buttons faster than your sampling period.
