Marble Madness - August 27, 2019

Edit history:

SDAVerification: 2019-08-27 12:55:46 pm

2019-08-11 01:04:01 am

Game Page: http://speeddemosarchive.com/MarbleMadness.html

Marble Madness (ntscus) (nes) [Any %] [Single Segment]

Decision: Accept

Congratulations to 'yelsraek'!

Thread title:

SDAVerification

Edit history:

SDAVerification: 2019-08-14 09:00:29 am

SDAVerification: 2019-08-14 08:59:38 am

SDAVerification: 2019-08-14 08:59:33 am

2019-08-11 01:04:01 am

Run Information

Marble Madness (ntscus) (nes) [Any %] [Single Segment]

Verification Files

http://v.speeddemosarchive.com/MarbleMadness-20190809/Marbles_240883_HQ_4x3_aspect.mp4

Please refer to the Verification Guidelines before posting.

Please post your opinions about the run and be certain to conclude your post with a verdict (Accept/Reject). If you wish to remain anonymous, you can also send a pm with your reply to 'sdaverification' (please state clearly in that case which run you have verified). This is not a contest where the majority wins - Each verification will be judged on its content.

SDAVerification

2019-08-14 09:02:37 am

Video file replaced (incorrect aspect ratio in the previous file).

puwexil

2019-08-14 10:23:36 am

Professional Second Banana

A/V quality fine for an NES game.

Enjoyed watching yelsraek's recent SGDQ run of this and have admired his dedication to squeezing every last possible frame out of this short & sweet run. Easy accept from me.

ktwo

2019-08-17 06:42:11 am

A/V: Very good (I wish I had a capture setup as good as this...)

Gameplay: This game has been optimized to a very high level and any improvement is by definition absurdly good. There are no mistakes in the run and the only known time savers are the occasional tighter cornering and if tricks currently considered TAS only start to be used. The well-written run comments explain this better than I can though, so I leave it at that.

Speaking of the run comments, I have two remarks.
The first is about the frame rule. It's mentioned in several places, but it's never explained how big it is. It would be interesting to include that information, even though I'm sure it's documented elsewhere.
The second remark is about the shortcut in the Silly race. It says that it's "pixel-perfect frame-perfect". Does that mean that not only does the marble need to be on the right pixel, but the shortcut is also only open on specific frames?

This run is absolutely top notch. A big congrats to the runner for the amazing achievement. The accepts don't get much easier than this.

ACCEPT!

A/V: Very good (I wish I had a capture setup as good as this...)

Gameplay: This game has been optimized to a very high level and any improvement is by definition absurdly good. There are no mistakes in the run and the only known time savers are the occasional tighter cornering and if tricks currently considered TAS only start to be used. The well-written run comments explain this better than I can though, so I leave it at that.

Speaking of the run comments, I have two remarks. 
The first is about the frame rule. It's mentioned in several places, but it's never explained how big it is. It would be interesting to include that information, even though I'm sure it's documented elsewhere.
The second remark is about the shortcut in the Silly race. It says that it's "pixel-perfect frame-perfect". Does that mean that not only does the marble need to be on the right pixel, but the shortcut is also only open on specific frames?

This run is absolutely top notch. A big congrats to the runner for the amazing achievement. The accepts don't get much easier than this.

ACCEPT!

yelsraek

2019-08-19 08:13:59 pm

Thank you for the kind words, Puwexil and ktwo!

ktwo, regarding your remarks:
- excellent catch! I talk about frame rules so much these days, that I forget to explain it sometimes! The following has been added to an edited comments file, before the level commentary:

"Frame Rule explanation: In Marble Madness, the game will only advance from one race to the next every 16 frames. At 60 frames per second, this equates to 0.266 seconds. For the first five races of the level commentary, time saved or lost will only be in these intervals. For the sixth and final race, the platform cycles rotate every two frame rules, which is every 32 frames or 0.533 seconds. Timing ends on an exact frame rather than a frame rule, because loss of control occurs on the frame that the 6000 appears over the marble, which is the accepted timing method."

- I reached out to elipsis to discuss this part further. I believe the shortcut to be "always open" if the marble is in the right spot, and as such have removed the "frame-perfect" reference to the edited comments file.

Thanks for the notes!

Thank you for the kind words, Puwexil and ktwo!

ktwo, regarding your remarks:
- excellent catch!  I talk about frame rules so much these days, that I forget to explain it sometimes!  The following has been added to an edited comments file, before the level commentary:

"Frame Rule explanation:  In Marble Madness, the game will only advance from one race to the next every 16 frames.  At 60 frames per second, this equates to 0.266 seconds.  For the first five races of the level commentary, time saved or lost will only be in these intervals.  For the sixth and final race, the platform cycles rotate every two frame rules, which is every 32 frames or 0.533 seconds.  Timing ends on an exact frame rather than a frame rule, because loss of control occurs on the frame that the 6000 appears over the marble, which is the accepted timing method."

- I reached out to elipsis to discuss this part further.  I believe the shortcut to be "always open" if the marble is in the right spot, and as such have removed the "frame-perfect" reference to the edited comments file.

Thanks for the notes!

ktwo

2019-08-20 03:19:12 pm

I got a bit curious about the Silly race skip, so I did a few quick tests. There is indeed a magic pixel that you need to pass through. The marble position is stored in $390 (x) and $3A8 (y). The labeling of x and y is of course arbitrary due to the isometric perspective.

The pixel you pass through the wall is in (16,-56) in decimal values. As you already wrote, the frame you do this on is not important. However, it's still a bit more to it than that. First of all, your movement speed is 1.5 pixels/frame when pressing a direction and A, which means that it's possible to overshoot the opening as a result of what's essentially outside player control. The other thing is that you need to arc in so that the y-speed/x-speed ratio isn't bigger than 1 (to be confirmed if that's the criteria). You can't just hug the left wall (not that the game allows you to do this to begin with, but whatever...) and just press left at the correct y-position.

One idea to improve the success rate based on the above observations is to let go of A just before pressing left for arcing in towards the magic pixel. Releasing A immediately slows down the speed to 1 pixel/frame. Due to some quirky game mechanics, once you press left and A, you will accelerate to above 1 pixel/frame in the y-direction though, which is unwanted. However, if you just press left (and not A), you will move 1 pixel/frame in both the x- and y-directions. That should avoid the possibility of overshooting by a pixel and reduce it to "only" a frame-perfect trick. It's of course slower to not press A, but with muscle memory it should be possible to reduce such a manoeuver to only a few frames.

I got a bit curious about the Silly race skip, so I did a few quick tests. There is indeed a magic pixel that you need to pass through. The marble position is stored in $390 (x) and $3A8 (y). The labeling of x and y is of course arbitrary due to the isometric perspective.

The pixel you pass through the wall is in (16,-56) in decimal values. As you already wrote, the frame you do this on is not important. However, it's still a bit more to it than that. First of all, your movement speed is 1.5 pixels/frame when pressing a direction and A, which means that it's possible to overshoot the opening as a result of what's essentially outside player control. The other thing is that you need to arc in so that the y-speed/x-speed ratio isn't bigger than 1 (to be confirmed if that's the criteria). You can't just hug the left wall (not that the game allows you to do this to begin with, but whatever...) and just press left at the correct y-position.

One idea to improve the success rate based on the above observations is to let go of A just before pressing left for arcing in towards the magic pixel. Releasing A immediately slows down the speed to 1 pixel/frame. Due to some quirky game mechanics, once you press left and A, you will accelerate to above 1 pixel/frame in the y-direction though, which is unwanted. However, if you just press left (and not A), you will move 1 pixel/frame in both the x- and y-directions. That should avoid the possibility of overshooting by a pixel and reduce it to "only" a frame-perfect trick. It's of course slower to not press A, but with muscle memory it should be possible to reduce such a manoeuver to only a few frames.

yelsraek

2019-08-21 01:50:34 pm

That is fascinating information about the way acceleration works in this game!

The strategy you mention about releasing A to arc into the magic pixel reminds me of how I approached the skip when I was first learning it. I couldn't quite figure out the fast wall bounces that I do now, so I opted to ease into it, and it seemed to work. That said, the fast wall bounces certainly feel quicker than a slower arc, but I won't deny the potential for a faster strat.

A little bit of extra information about the skip that might interest you:
- You mentioned the game doesn't let you just rub along the left wall. Technically, you can manufacture that rub while doing the fast bounces. If your angle is too low to bounce into the skip, and you just hold left, you should find yourself motionless, rubbing against the left wall. At this point, light releases and re-presses of left will slowly work you up the wall. I imagine if I put this into a TAS editor and released and re-pressed left over and over on alternating frames, it might LOOK like I rub against the left wall slowly and eventually get through the skip, albeit very slowly.

- Holding left up against the wall also appears to move your marble DOWN the wall by at least one pixel in certain circumstances. I'm aware of this phenomenon because if you do the fast bounce techniques and miss the magic pixel by one pixel too high, if you continue to hold left and rub against the wall, the skip will eventually "let you in" for lack of a better phrase.

That is fascinating information about the way acceleration works in this game!

The strategy you mention about releasing A to arc into the magic pixel reminds me of how I approached the skip when I was first learning it.  I couldn't quite figure out the fast wall bounces that I do now, so I opted to ease into it, and it seemed to work.  That said, the fast wall bounces certainly feel quicker than a slower arc, but I won't deny the potential for a faster strat.

A little bit of extra information about the skip that might interest you:
- You mentioned the game doesn't let you just rub along the left wall.  Technically, you can manufacture that rub while doing the fast bounces.  If your angle is too low to bounce into the skip, and you just hold left, you should find yourself motionless, rubbing against the left wall.  At this point, light releases and re-presses of left will slowly work you up the wall.  I imagine if I put this into a TAS editor and released and re-pressed left over and over on alternating frames, it might LOOK like I rub against the left wall slowly and eventually get through the skip, albeit very slowly.

- Holding left up against the wall also appears to move your marble DOWN the wall by at least one pixel in certain circumstances.  I'm aware of this phenomenon because if you do the fast bounce techniques and miss the magic pixel by one pixel too high, if you continue to hold left and rub against the wall, the skip will eventually "let you in" for lack of a better phrase.

ktwo

2019-08-23 08:02:58 am

The TAS arcs in towards the wall without bouncing. Is there anything special you do to create additional speed when bouncing? When just bumping into the left wall normally, I agree that it looks like the marble shoots off, but that's more of an optical illusion based on diagonal speed-up. The y-speed doesn't go beyond 1.5 pixels/frame for me.

Unless there is a technique I'm missing, the double-bounce you do is slower than the arc method of the TAS. Frame 8445 in your run and frame 7560 in the TAS have almost the same positions (give or take a frame). Frame 8492 in your run and frame 7588 in the TAS have the same position (the marble makes a very distinct jump when clipping through the wall, so it's a good reference point). Your path took 47 frames, while the TAS only spent 28 for reaching the same end point.

Just to clarify about my suggested technique in the previous post. It wasn't focusing on the arc path to speed things up, although it now looks like the arc path is faster. The focus was to make the clip more consistent for a minimal time loss. Right now, a lot of the attempts are going to go down the drain because it's impossible to hit the correct pixel if you move 2 pixels every other frame. Of course, the trick is so precise that you're not going to know when a missed skip was because of bad luck and when it was because you just missed the input by a frame. I believe you're going to see the success rate increase if you let go of A and go at default speed though.

The TAS arcs in towards the wall without bouncing. Is there anything special you do to create additional speed when bouncing? When just bumping into the left wall normally, I agree that it looks like the marble shoots off, but that's more of an optical illusion based on diagonal speed-up. The y-speed doesn't go beyond 1.5 pixels/frame for me.

Unless there is a technique I'm missing, the double-bounce you do is slower than the arc method of the TAS. Frame 8445 in your run and frame 7560 in the TAS have almost the same positions (give or take a frame). Frame 8492 in your run and frame 7588 in the TAS have the same position (the marble makes a very distinct jump when clipping through the wall, so it's a good reference point). Your path took 47 frames, while the TAS only spent 28 for reaching the same end point.

Just to clarify about my suggested technique in the previous post. It wasn't focusing on the arc path to speed things up, although it now looks like the arc path is faster. The focus was to make the clip more consistent for a minimal time loss. Right now, a lot of the attempts are going to go down the drain because it's impossible to hit the correct pixel if you move 2 pixels every other frame. Of course, the trick is so precise that you're not going to know when a missed skip was because of bad luck and when it was because you just missed the input by a frame. I believe you're going to see the success rate increase if you let go of A and go at default speed though.

yelsraek

2019-08-23 07:49:52 pm

I think I understand what you mean more clearly now. Thank you so much!

SDAVerification

2019-08-27 12:55:47 pm

Decision posted.

Elipsis

2019-09-16 12:17:43 am

Wow, belatedly found this thread. Obviously congrats to yelstraek! ktwo, thanks for all the research you did on this stupid trick - I think it's definitely worth experimenting with. Thank you for confirming several things we've suspected for years.