Quick topic while I put off studying for my finals. I would actually be surprised if this thread doesn't already exist, but I couldn't find anything. Feel free to merge this with an existing thread, as appropriate.
Very simple question: What can make a game easy to speedrun (unassisted, of course) while hard to TAS? What games have these qualities? Obviously, the TAS is always going to have a faster time and more polished quality, but I imagine there are good examples of games where the unassisted speedrun was practiced and produced in a matter of months while the tool-assisted version took years of studying, decompiling, and redoing major segments just to meet the high standards we set.
The one aspect that came to my mind was luck manipulation. The unassisted runner would accept the law of averages and not challenge their results too much while the tool-assisted runner would spend days just trying to manipulate the result he wants. I think a few RPGs could fit this profile (Fire Emblem, perhaps?).
(Another possibility that just crossed my mind is most 3-D games. For a TASer, the addition of a third dimension increases the difficulty of optimizing movements by tenfold, while a conventional speedrunner just does the best he can. I think Mario 64 runs might fall under the "easy to speedrun, hard to TAS" category.)
I don't know how good an example it is, but FatRatKnight and I are working on a Final Fantasy Legend TAS. While work has been off and on, we've devoted a lot of time and energy to decoding the RNG, going so far as to make a bot that will save and reset every frame to see how we can manipulate the RNG. Meanwhile, over at SDA, Poxnor has been able to produce a run that (from what I understand) he began to seriously practice in February (about two-and-a-half months' work). The same RNG that is giving FatRatKnight and me so much trouble has a period of just 256 and begins from a fixed state, so Poxnor just accepted the luck he was presented with and structured his run around making the most of it.
What other games are like this?
Additionally, what makes a game particularly easy to TAS but brutally difficult to speedrun unassisted? Here's a simple but diabolical scenario that comes to my mind:
• Have a trick in the first level that the player has a 1 in 2 chance of pulling off and saves one second.
• Have a trick in the second level that the player has a 1 in 4 chance of pulling off and saves two seconds.
• Have a trick in the third level that the player has a 1 in 8 chance of pulling off and saves four seconds.
...
• Have a trick in the nth level that the player has a 1 in 2^n chance of pulling off and saves 2^(n-1) seconds.
This is known as the St. Petersburg lottery (Edit: Okay, not exactly, but the spirit is similar). While the TAS should be straightforward, it would be frustrating to an unassisted runner because they would be forced to keep sub-optimal runs in the desperate hope that they nail one of the later time-savers. Are there any games like this? What other qualities can make a run especially easy to TAS but cruelly difficult to speedrun conventionally?
You nearly fully answered your question.
Bobo the King wrote:
-studying, decompiling, and redoing major segments just to meet the high standards we set
-luck manipulation
-third dimension increases the difficulty of optimizing movements by tenfold
-lot of time and energy to decoding the RNG
-make a bot that will save and reset every frame to see how we can manipulate the RNG
Notice that games are determinable, and it's easier to find out what action does what in the memory while you are tasing instead of unassisted speedrunning. Also add the factor that you have full access to possible input combinations (U+D, L+R, L+R+U etc) and you can view everything the game has (hitboxes, layers, hardware values (NMI)).
I can't see any logical connection between St. Petersburg paradox and the thing you mentioned. Easy to TAS: Tetris, you can simply manipulate the next blocks (of course the unassisted could be easy if they use the MAME version (always the same block series after reset)). Hard to TAS: Mega Man (DelayStageClear).
PhD in TASing 🎓 speedrun enthusiast ❤🚷🔥 white hat hacker ▓ black box tester ░ censorships and rules...
Desyncs and variable framerate. Lag is a problem too, but that's a problem in normal speedruns too, if you by lag mean that the framerate drops so that the game doesn't accept input 60 frames per second, but goes more like 20 fps, which is common in N64 games.
How about a game with the following:
-The optimal speedrun, using all tricks that can be performed in real time, is almost trivial.
-However, there is an ultra obscure series of glitches with vanishingly small probability to pull off, that a TAS will be expected to do to be any faster than the RTS.
I suspect that in general any situation in which very careful optimization performed throughout the entire run can save a small handful of frames is one that meets these requirements. The realtime speedrunner isn't going to be able to do these optimizations reliably, and the gains aren't big enough to make it worth their time to keep trying until they manage it through persistence. The TASer of course can do the optimizations, but that doesn't necessarily make them any easier to manage. Lag control and luck manipulation are two obvious sources of this kind of problem. Complex movement patterns probably also qualify -- subpixel manipulation, for example.
Pyrel - an open-source rewrite of the Angband roguelike game in Python.
Desyncs and variable framerate. Lag is a problem too, but that's a problem in normal speedruns too, if you by lag mean that the framerate drops so that the game doesn't accept input 60 frames per second, but goes more like 20 fps, which is common in N64 games.
Desyncs? Isn't that just caused by bugs in the emulator?
Variable framerate seems to be related to lag. Which is actually what you also say.
I don't know how good an example it is, but FatRatKnight and I are working on a Final Fantasy Legend TAS. While work has been off and on, we've devoted a lot of time and energy to decoding the RNG, going so far as to make a bot that will save and reset every frame to see how we can manipulate the RNG. Meanwhile, over at SDA, Poxnor has been able to produce a run that (from what I understand) he began to seriously practice in February (about two-and-a-half months' work). The same RNG that is giving FatRatKnight and me so much trouble has a period of just 256 and begins from a fixed state, so Poxnor just accepted the luck he was presented with and structured his run around making the most of it.
This is a perfect example for my reason. Nothing 'required' you to put that much effort into your TAS. However, you wanted to have the best possible TAS, so you put the effort in. On the other hand, Poxnor could have made multiple attempts, and not just accepted the luck he was given, and possibly made a better RTS.
Super Mario 64 is actually another example. The TAS itself didn't require the frame precision that it has (remember, a ~30 second longer TAS was published, and well loved). The people who made the new TASes WANTED to expend the effort. Additionally, a RTS could spend their lives attempting to maximize their speed, at their own discretion.
In short, a TAS will never be harder to make than an RTS, as a TAS can just be an RTS done in frame-advance. However, it comes down to the individual, and how much time they want to put into achieving perfection, on both sides of the RTS/TAS world.
Sage advice from a friend of Jim: So put your tinfoil hat back in the closet, open your eyes to the truth, and realize that the government is in fact causing austismal cancer with it's 9/11 fluoride vaccinations of your water supply.
Surely the biggest reason of all would be a game that doesn't emulate properly?
More seriously, for a game to be harder to make an interesting TAS than to make an interesting speedrun, would be a game where the two routes were very different. (Imagine a series of jumps that nobody can make in realtime, that are easy to do in frame advance, and lead to a faster but massively more difficult route through the level. It's hard to see how that could happen in the absence of some really weird glitches, though.)
Zombies Ate My Neighbors 100% (rescue all neighbors), first level you must get a bazooka or monster potion effect from the few random drops on the level. An speedrunner will usually stay two hours on average waiting for this drop to occur. While this is trivial for a TAS.
Once the run is started the speedrun becomes easier than the TAS though but this is because the runner will disregard anything but game completion and final time once he went through hell for that first drop :), a TAS would keep manipulating drops to get an extreme advantage on final time.
I think that what the original poster meant was that in both unassisted and assisted playing the goal is to complete the game as fast as is possible, using realistic expectations of what is possible unassisted or tool-assisted. The goal of a TAS is not to beat the unassisted run (in which case it would certainly be much easier), but to beat the game as fast as possible using all the tricks available, which can indeed make creating the run much more difficult, as many of those tricks are not even expected to be used in an unassisted run (eg. luck manipulation giving you 100% success rate).
An experienced speedrunner can make a 1-hour speedrun of a game in... well, literally 1 hour (assuming a bit of luck so that he doesn't need to restart). However, making a competent TAS of the same game could require months, if not even years. That could be taken as a measurement of how much more difficult it is.
An experienced speedrunner can make a 1-hour speedrun of a game in... well, literally 1 hour (assuming a bit of luck so that he doesn't need to restart). However, making a competent TAS of the same game could require months, if not even years.
Speedruns can take months and years too, it all depends on how much time you're going to put into it, and how perfect of a run you want to make. For example, my brother put like... I don't know, maybe a month to a 5 second speedrun (the record had been 6 seconds for 9 years), while I only put 5 hours of TASing into the same level. So yeah... maybe it actually took 9 years to get 5 seconds on Defection?
Saying that something took months or years can be quite inaccurate though... it's better to say how many hours of work there was. I could say that it took Wyster and me 4 years to make the PD run, but it's not like we were working on it 10 hours a day for 4 years, Wyster made most of the levels in a few months, then there was a pause of maybe 3 years, until I made the rest.
A game that uses a true RNG would be impossible to TAS but no harder to speedrun than a game that uses a PRNG. However, I don't know of any games that do this.
I don't follow. TASing by definition implies full control over all inputs to the system, which would have to include the source of randomness a "true RNG" implies. If your game system includes a radioactive source and a Geiger counter, for example, then a TAS-capable emulator of that system would have, as one of its inputs, the particle count for the frame. Or alternately, the emulator would include a simulated (not truly random) radiation source, which converts the randomness problem into standard PRNG manipulation.
Pyrel - an open-source rewrite of the Angband roguelike game in Python.
I'd say that in general games with hard to manipulate RNGs and many ultra-rare events that can speed up the run are a million times harder to TAS than to speedrun them. An RTS runner wouln't be expected to have perfect luck, but a TASer would. You'd have to do careful planning, maybe run brute-forcing scripts over weeks and months, it'd be hell.
Joined: 11/22/2004
Posts: 1468
Location: Rotterdam, The Netherlands
There are also a lot of things that make a game much harder to speedrun. Such as events with a very low probability. In Quake speedrunning, in level E3M1 there's a slope jump that can be replaced with a bunny jump (I'll spare you the details of how exactly that works) but it's so incredibly unlikely to succeed that it's not used in even Quake done Quickest. I think it's done in the easy version of the speedrun, but not the nightmare version.
Any event involving a high degree of luck, regardless of how much skill is required, will be a massive obstacle in a speedrun but of no consequence in a TAS.
I made a single-player SMB1 clone that runs at 84000 fps a few years ago.
<klmz> it reminds me of that people used to keep quoting adelikat's IRC statements in the old good days
<adelikat> no doubt
<adelikat> klmz, they still do
I think that what the original poster meant was that in both unassisted and assisted playing the goal is to complete the game as fast as is possible, using realistic expectations of what is possible unassisted or tool-assisted. The goal of a TAS is not to beat the unassisted run (in which case it would certainly be much easier), but to beat the game as fast as possible using all the tricks available, which can indeed make creating the run much more difficult, as many of those tricks are not even expected to be used in an unassisted run (eg. luck manipulation giving you 100% success rate).
I was one day watching Torje's stream on twitch, where he was speedrunning LoZ:OoT, and the conversation veered towards the OoT TAS and why nobody seems willing to do one, using modern routes and tricks. He mentioned something that might be the perfect example of what the original post (and the thread title) is asking.
Not that OoT any% is easy to speedrun, of course (getting even close to the time of the current any% record requires an enormous amount of practice and work), but it's of course perfectly possible. According to him, however, an optimal TAS would require such an astonishing amount of work to get certain tricks working that it's unyielding.
He specifically mentioned the floor clip trick inside the Deku tree. It's possible to do even in real time, as he demonstrated in that very stream, but requires a very specific setup that requires quite some time. A TAS would obviously need to do the trick without any setup at all, instead going right to it. And, apparently, it requires such an accurate positioning / running direction / timing or something, that it would be an almost insurmountable amount of work, even with TAS tools.
(I'm not sure I fully understood the reason, but he knows infinitely more about the internals of the game than I do, so I'm not questioning his expertise and knowledge.)
Also, movement can be optimized to a ridiculous level everywhere in OoT, with all the twisty sidehops you're expected to do, so even nominally boring stuff becomes tedious.
