Don't create disjoint unions?

[Simon]

Hi folks,

this wasn't clear-cut at all. The first post has some good ideas, but jumps to strange conclusions. Read all the responses and interpret yourself.

Original first post follows:



my visits at IchoTolot's always lead to fruitful discussions. We have found many unwritten guidelines for contemporary level design. It's time to present some of them in an orderly fashion. I think Icho will make another tread, he's the master of making interesting large levels.

We want to present guidelines, not rules. If you're a level designer, you can break them at your own discretion -- be sure though that you have a very good reason for breaking them.



Don't create disjoint unions: Don't make levels such that each half is an independent level. For example, don't roll two small levels into one large level, separating the halves by steel or lots of empty space. The same problem occurs when you solve one half, your lems enter the second half, and your actions from the first half have absolutely no effect on the second half.

Modern engines allow players to roll back mistakes, but they don't allow merging/rebasing two separate replays. Whenever you multitask, and fix mistakes in one half, you must rollback and repeat the simultaneous actions in the second half, even if they are completely unrelated.

The problems of disjoint unions tend to outweigh by far the puzzle of dividing the skillset.

Bad examples: Rendezvous at the Mountain, Synchronized Lemming, You take the high road. Also Ground City from Pimolems; it crams three moderately interesting levels into on gigangic disjoint union. Dividing the skillset is easy for all of them. Imagine these union levels with properly separated replacement levels (left half of Rendezvous, then right half of Rendezvous in a separate level). The artificial difficulty of the union level over the separated levels lies purely in the execution.

Somewhat bad example: Lost Your Ground from Icho's pack is a disjoint union of two equal halves. It's not that bad because each half is really small, and you don't have to multitask. But the basic problem remains. There is no extra concept to the solution brought by the mirror image. Icho has chosen to override a guideline for aesthetics here, the level still almost fits on one screen. On the other hand, don't mirror a huge level with no extra gain -- you wouldn't have a good reason to override the guideline!

Good examples Not disjoint unions: Sharing the World from Icho's pack, and Try to Compromise Before You Cue from Pimolems. Both of these look completely symmetrical, but require assymetric solutions. [Edit after Nepster's first comment: Not only assymetric, but a solution where the sides interact. So, these aren't disjoint unions after all, even though they look like one.]

Older threads with design guidelines

Time limits: Give unlimited time wherever possible.
Bridge stretching: Design gaps such that stretched and non-stretched bridges behave the same.

-- Simon

[Proxima]

As you may guess, I disagree almost completely. Much more than I disagreed about time limits -- I'm not dead set against them like Simon, but I do agree that overall the game is improved by the time limit being optional and most levels not using one. For this, though -- several of the levels you've named are among my favourites in their respective packs, and I really enjoy puzzles that are about working out how to divide up the skills -- whether it's between disjoint parts separated by steel (as on my level "Think Inside the Box") or sequential parts of the path for a single group of lemmings (as on "Survival of the Craftiest"). The ONML examples are indeed much easier than these, but not trivial for the level of player ONML is aimed at; I clearly remember from when I first played the game that they stood out as particularly fun levels because of this type of puzzle.

There is, indeed, a difficulty with the replay function, but cautious play can avoid this becoming too much of a problem. For instance, in Havoc 14, you can assign the builder, wait to see whether you placed it correctly, then go back and continue assigning other skills.

(Curiously, like several other recent discussions, this one exactly parallels a debate we've had on the DROD forums. The latest DROD introduces temporal split tokens, which allow you to "record" a series of moves, go back in time to the moment you stepped on the token, and continue play, while a temporal clone of yourself carries out the moves you recorded. The problem is, of course, that you then can't change an individual move of the recording without going back to a checkpoint during recording and then repeating the rest of the recording and however much you played after rewinding. Naturally, there have been some grumbles about this, but there is still a strong consensus that the temporal split token is one of the best new features, and its ability to create interesting puzzles outweighs this disadvantage.)

Finally, I have to say I'm unhappy with the way this thread is presented. The time limits topic has the neutral title "Time limits", so the reader feels invited to come in and debate. This one begins in bold letters "Don't do this." Sure, the small print says "we want to present guidelines, not rules", but that's the small print, after the reader has already been unsettled.

[ccexplore]

Finally, I have to say I'm unhappy with the way this thread is presented. The time limits topic has the neutral title "Time limits", so the reader feels invited to come in and debate. This one begins in bold letters "Don't do this." Sure, the small print says "we want to present guidelines, not rules", but that's the small print, after the reader has already been unsettled.

Okay, I can fix that part.   Though to be fair, it's possible someone might not even bother to read it if the words "don't..." aren't there.  Unlike recognizable terms like "time limits", that one probably sounded more like some off-topic math discussion to the layman.  Having the "don't" part might at least pique the curiosity of some potential readers to actually read the topic.

============

Anyway, I feel like this is directly a consequence of not having a better replay scheme for handling multitasking.  I also recognize it is probably tricky to design a good replay scheme that works well for multitasking.

I think it's also worth mentioning that since multitasking does not necessarily require a disjoint union design, the multitasking replay problem can still apply even without a disjoint union design.

Maybe a better, more generic way to look at this is whether some multitasking you are requiring in the level (whether introduced by a disjoint union design, or some other means) adds anything to the level besides execution difficulty.  For example, Proxima's point about working out how to divide up the skills, could be something where the multitasking adds to the level.

============

As for specific examples from the original games that were mentioned:

- Rendezvous at the Mountain: AFAIK multitasking is not required due to the generous time limit, and it is fairly easy to set up either side to become "safe" while you work on the other side first.  So ultimately, the main complain I can see is that the level is too large for some's tastes.  The skillset given appears generous enough that there really isn't much of a puzzle element either in having both sides.  Ultimately the multitasking disjoint union does not add but also does not also detract much from the level.  Of course, it has to mentioned that there is no "rendezvous" conceptually if the level is not allowed two entrances and one exit, for what that's worth.

- Synchronized Lemming:  this one is definitely about multitasking especially given the tight time limit.  There also is some small amount of working out skills distribution due to the less generous skillset.  It is also obviously not interesting if you are forced to break the level down into its individual 4 parts.  The level is reasonably straightforward and short enough that I don't really see a huge pain point around the replay deficiencies with multitasking.  Ultimately, I'd say it adds maybe a tiny bit to the level and does not detract too much.  Given that it's not like the game is blasting the player constantly with these types of levels throughout the entire rating, I think this is an acceptable occasional change of pace.  It would definitely be less okay if the level consists of four much larger disjoint areas with much more complex and error-prone paths to execute.

- You take the high road:  this one does feel like it might encroach on annoying multitasking.  With RR 99, 100% required and a 2-minute time limit, multitasking does seem required (though I suspect it might not be strictly necessary) and there does appear to be a few spots where fatal mistakes are quite possible.  But having played it, I believe once you know exactly how you are going to blaze the path, execution mistakes are usually almost immediately apparent, and the moves from each half don't interweave too much temporally, so it seems you are rarely if ever forced to roll back the replay substantially to address mistakes.  The skillset gives you more than you actually need, and each half is relatively simple, so I'd say the multitasking did not really add much puzzle-wise here.

The biggest complain I can see may be that by virtue of the forced multitasking, the player solving the level likely needs to initially focus on each half separately, then restart the level at least once to execute the two solutions put together.  Of course, that complain can probably be applied to pretty much all multitasking levels.  At least this one is relatively short.

Frankly for me, there just isn't much to the level with or without multitasking.  If the level isn't that great to start with, does it matter whether it has a disjoint union design?

[Simon]

Now I'm unhappy with the topic title. The initial post is clearly an appeal against them, and so should be labelled appropriately. "Time Limits" is a bad topic title for the same reason, too indescriptive. I'm fine with "Don't create disjoint unions?". There is no need to make topic titles neutral, but there is a severe need to make them descriptive.

I'm honestly and thoroughly surprised by the like for the existing disjoint unions. If it's personal taste, fine, there's nothing to disagree then. If the levels are liked for the novelty, fine. I also agree how there could be appeal in figuring out the skill allocation. As mentioned already, skill allocation is not a puzzling issue with any L1/ONML disjoint union.

geoo has suggested on IRC how disjoint union is a design principle that's not good or bad on its own. I'm sure you can do interesting things with it, and that's fine. You can override guidelines if you see fit. I want to point out that the existing disjoint unions, IMHO, suck. It's super tricky to get it right, so it's appropriate to be wary of them as a default.

About both of the ONML disjoint unions, when I solved them 12 years ago, I recall finding them easy, but wondering why I got stuck on them nonetheless. It was all execution.

I feel High Road is the worst offender, with the need for scrolling. You could easily mirror one of the halves in High Road, and get the same disjoint union level without the scrolling. I believe the need for scrolling in High Road is deliberate, the L1 devs liked to make hectic levels. The result are unnecessary errors on the player's side. This isn't a dexterity game; you can do all planning before play, you aren't forced to cope with unexpected situations during play.

Rendezvous: The title can't be a saving grace for a bad level. Either the level is interesting or it's not. The level was likely chosen because it looks monumental. Icho's pack finishes with a monumental level too, which may be too big for my taste, but at least it's not 80 % builder chaining.

Fast Food Kitchen: Obvious disjoint union, probably even better of an example, and a horribly trivial level. The Tricky-ranked copy is of the same difficulty as the Mayhem version.

Replay function shortcomings: Agree, that could help tremendously.

Bonus: 4 new lovely level picture links in the original post!

-- Simon

[ccexplore]

Haha, cute, love the use of the disingenuous question mark headline device to appease the neutralists.

I'm honestly and thoroughly surprised by the like for the existing disjoint unions.

"Like" may be too strong a word.  I'd say I'm more or less aligned with geoo's stance on this.  I also explained that maybe the discussion should center more around the multitasking aspect given that seems to be where the primary objection is around.  Using levels from the original games as examples tend to have the weakness that most of them also tend to violate other design principles we've gravitated towards.  For example, Rendezvous seems to be too big and builder-heavy, something which is hardly unique to that level in the game and certainly does not require disjoint union design to achieve.

I'd say it is definitely the case that in the very few instances in the original games where multitasking is featured, it was definitely more about upping execution difficulty than actually improving the puzzle aspects of the level.  Keep in mind this is also a game that features things like timed bombers.  My experience is that the timed bomber levels in the original games are by far much more annoying than the multitasking levels, and I seem to recall there're more timed bomber levels than multitasking levels as well.  Disjoint union I would say is just not as widespread a technique being used as other more problematic design elements.

The need to switch back and forth between two distinct areas is definitely an annoyance even if we solve the multitasking replay issue.  In that regards it is perhaps a good thing that Lix does not feature a minimap, which hopefully level designers should take as a hint that they should avoid gratuitously forcing the player into that situation.

[Nepster]

Let me add just my 2ct to this discussion.

The "Good examples" in the first post are bad examples.
They are indeed very good levels, but they do not fit the topic: They are not disjoint unions! In (at least my) solutions both sides interact, so they share the mirror design seen as well in Fast Food Kitchen, but this is as far as the similarities go.

Synchronized Lemmings is a good level for ONML, but not for NeoLemmix.
In L1 and ONML the difficulty in executing multitasking is a deliberate part of the game. We may consider this good or bad, but that is a topic that is independent from disjoint design. Once you accept that L1/ONML was meant to have levels, where the main goal was to succeed in executing an easily found solution, then levels like Synchronized Lemmings become indeed very good levels, in the sense that they feature this part of the game in an interesting way.
On the other hand, NeoLemmix with all its features removes this part of the Lemmings game (again as a deliberate design choice). Thus it removes the sole reason, why levels like Synchronized Lemmings were included in the first place. So it is no wonder that the level that remains is seen as a bad level.

On the other hand, I totally agree that levels like Rendezvous at the Mountain are bad levels even in L1. And I agree as well, that You take the high road would have been a much better level for ONML, if one half would have been mirrored.

Disjoint unions are not the real problem here.
Coming back to the topic at hand, I feel that disjoint unions are not the real problem. The real problem is, when level designers feel that an uninteresting level can be made better (or more difficult), by prolonging the level or adding further hatches. This does not help at all! On the contrary: If makes the level worse, because the new level will not be more difficult, but simply more annoying.
Of course, frequently such prolonged levels will coincide with disjoint union levels, but that does not mean they are the same. There are some disjoint unions that are good puzzles (e.g. by having to figure out the skill distribution), while there are some bad prolonged levels without a disjoint union design.

[Simon]

"Like" may be too strong a word.

Yes, I got that your stance was pretty neutral. I was surprised about Proxima's almost-complete disagreement.

The "Good examples" in the first post are bad examples.

Correct. Annotated the OP accordingly.

I'd like to see some recent disjoint unions, along with the reasons why they're deemed good levels. I know Proxima has a disjoint union in the Lix community pack, which I found very irritating to play due to torus scrolling, and therefore haven't attempted. But geoo likes this level.

Once you accept that L1/ONML was meant to have levels, where the main goal was to succeed in executing an easily found solution, then levels like Synchronized Lemmings become indeed very good levels, in the sense that they feature this part of the game in an interesting way.

Right, then it fits with DMA's design ideas well.

I am more concerned with disjoint unions in contemporary level design, and chose the DMA levels as readily understandable examples.

Disjoint unions are not the real problem here.
Coming back to the topic at hand, I feel that disjoint unions are not the real problem. The real problem is, when level designers feel that an uninteresting level can be made better (or more difficult), by prolonging the level or adding further hatches. This does not help at all! On the contrary: If makes the level worse, because the new level will not be more difficult, but simply more annoying.

Excellent insight. This can well be a guideline on its own. There is a sizable intersection of applicable levels with disjoint-union levels, and it points sharply to problems in some non-disjoint-union levels.

On the other hand, there are disjoint unions that this fails to describe, because they seem to have been designed as disjoint unions from the get-go -- Fast Food Kitchen and Rendezvous.

So, while I don't deem this the real problem behind disjoint unions, I feel it's a much more important guideline on its own. Perfect.

-- Simon

[Nepster]

I'd like to see some recent disjoint unions, along with the reasons why they're deemed good levels.

Not really recent, but here some good disjoint unions:
1) Oh, the possibilities... in shveg03.dat
The level is split into four disjoint parts, but the skillset is very restricted. So you have to decide between saving all lemmings from one hatch (which one?) or saving some lemmings from multiple hatches (again which ones?). And only very few of your choices will actually work...
(The actual solution is not that interesting, but that is due to the design of the pieces, not because it is a disjoint union.)

2) Everyone But The Bomber in twbestof.dat
Again we have four disjoint parts. Each part in itself can be solved in many ways. But the challenge comes in distributing the skills and choosing the correct solution for each part.

3) Multiple Choice in twbestof.dat
Similar to the level in shveg03.dat: Four disjoint parts and one has to choose now three out of four to save. The time limit is annoying, but otherwise...

4) The Final Countdown in GARJEN06.dat
Mirror design, but different solutions for both sides required (due to skill restrictions).

On the other hand, there are disjoint unions that this fails to describe, because they seem to have been designed as disjoint unions from the get-go -- Fast Food Kitchen and Rendezvous.

I am not so sure with Rendezvous. It looks very similar to some early levels of mine (which fall squarely in the category of bad disjoint unions), and they were typically created in the following way: After one side was done and playtested I realized, that there is some more space available. So I filled it with the second half of the level .

[Simon]

This is very exciting, because many of these introduce a new concept: You can forego a sublevel if you perform well in other sublevels of the disjoint union.

Even ONML has an instance of this (Oh no, it's the 4th dimension; I should find the picture later and link it). You can ignore the hatch in the lower left. I did not remember this level when writing my original post. It's a more interesting disjoint union, and I didn't recall it to be as annoying as the others for me.

Actions in one sublevel affect what you have to do in a different one, because the save requirement is low enough. It's more complex than merely allocating skills. At the same time, it takes off the drudgery of doing everything. I feel it's plenty to warrant a disjoint union design every once in a while. Thanks for the examples!

-- Simon

[ccexplore]

Even ONML has an instance of this (Oh no, it's the 4th dimension; I should find the picture later and link it). You can ignore the hatch in the lower left. I did not remember this level when writing my original post. It's a more interesting disjoint union, and I didn't recall it to be as annoying as the others for me.

I'd say that level is not strictly speaking a disjoint union, even though it is close to one.  If you want to save the upper-right group, IIRC you are meant to get someone from one of the other groups to bash through the one-way wall, as the wall's top is close enough to the ceiling (ie. top boundary of level area) to prevent any climbers from the upper-right group from getting over the wall themselves.  Although I guess technically you could probably climb-bomb your way through it instead.

I'm also not sure I would really count the ability to completely disregard one group in that level as that much of a good thing.  That group also happens to be arguably the one that takes the most work to create a path to the exit.  In comparison the other three groups' paths to exit are almost trivial.  The ability to abandon that lower-left group therefore feels more like a cop-out, a concession to perhaps making the level more Wicked than Havoc.  (Though it does arguably keep out a lot of somewhat heavy building from being required.)  The level has a low enough RR that it hardly counts as multitasking even in the very few parts where it exists.  The upper groups are safe by default and the lower-right group only takes one builder to reach a safe configuration, so even if it were required to save all groups, it doesn't take much to quickly get to a point where you can focus solely on dealing with the lower-left group first.

One other thing worth mentioning is that with multi-entrance levels, being very generous with the save requirement may end up leading the level designer into situations where one group can be completely ignored even if that may not have been the intention, since the game doesn't provide a way to enforce, for example, minimum save requirements per entrance.

I'll grant you that more generically speaking, making the decisions of choosing how many lemmings from which group to save, in addition to the obvious distribution of skill usages, is an interesting puzzle element to consider with multi-entrance levels.

[Simon]

6 years later.

The disjoint union can focus on skill distribution, as Proxima said, or on selection of how many to solve from which area, as Nepster and ccx said, etc., and all these ideas can certainly lead to great puzzles that are best realized as a disjoint union.

The big problem is the tooling of the games: To solve a disjoint union, the game UI should ideally offer several replays. Maybe the map is partitioned into several areas, and there is a separate replay for each area. When you assign or click the air, you drop future assignments only from the replay of the area where the mouse cursor is.

I'm 70 % sure that, 6 years ago, my original distaste of disjoint unions came from the hassle of solving each sublevel while avoiding the nasty unwanted replay modifications in other sublevels.

NL comes closest to this tooling idea with its replay-insertion mode that doesn't drop future assignments on clicking, and you have to manually erase them.

-- Simon

[WillLem]

We have found many unwritten guidelines for contemporary level design
---
We want to present guidelines, not rules. If you're a level designer, you can break them at your own discretion -- be sure though that you have a very good reason for breaking them

To be honest, even before reading the remainder of this post I'm already at odds with the attitude presented here; i.e. that there are any "guidelines" or "rules" that a level designer must be prepared to justify to a self-appointed quality control expert.

Ideally, packs should be made up of a variety of levels - some good, some bad, some with time limits, some without, some disjointed, some unified. If every level ends up basically the same because it conforms to an impossible set of standards preferred by a small group of players/designers, then level pack will very quickly become predictable and boring, even if not solvable by the average player.

So, I say: create whatever levels you want to create, and don't justify them to anyone. They will either be played and enjoyed by people, or not.

In addition, regarding your more recent post:

The big problem is the tooling of the games: To solve a disjoint union, the game UI should ideally offer several replays

Should it? I instinctively dislike the idea that any game engine or UI should force a designer to create certain types of level by rendering other types of level pointless, obsolete or easier than they should be. Sure, I love using NL's replay editor, but it comes at a price.

There have been many times when I've had to scrap levels because they simply wouldn't work on one of the modern interfaces. Some of them have been bad ideas, maybe, but others would have been perfectly playable and enjoyable levels. What it means is that we're ultimately left with a certain type of level being preferable on that engine (be it NL, Lix or SL). Whilst this is not necessarily a bad thing, it does mean that level packs ultimately become less varied.

A disjoint union level played on an engine which allows editing of each part of the level effectively reduces the level to two (or more) separate levels (which you've stated is your preference for dealing with this type of level), rather than keeping it as the multi-tasking challenge it's meant to be.

I've created this topic as a counterpoint to this discussion.

[IchoTolot]

Ideally, packs should be made up of a variety of levels - some good, some bad, some with time limits, some without, some disjointed, some unified.

That is the thing I would highly disagree with.

Not the variety part, but the examples are poorly chosen: Why make intentionally bad levels? Just because every level comes out good doesn't mean the pack misses out on variety.

Time attack levels I would add to variety, but on the other side: Why do you specifically need to have timers or disjointed levels?

LemRunner for example only has locked exit levels and I wouldn't call out that pack for a lack of variety there.

The solution itself is the variety in Lemmings and the possibilities there are always endless! It doesn't matter if you leave out or entirely focus on timed, locked exit,....etc levels. It is always endless.

To be honest, even before reading the remainder of this post I'm already at odds with the attitude presented here; i.e. that there are any "guidelines" or "rules" that a level designer must be prepared to justify to a self-appointed quality control expert.

In a long running community you will run into this no matter the case or the topic.

Over the time a meta/codex/guideline/culture..etc will evolve. There is no stopping it and it is inevitable.

Example: Everybody build their hidden trap/stuff level at first. We had a laugh, but let's say if you hear the joke for 500 times it becomes very annoying and you don't want to waste your time with it again. That's why new content creators are usually guided to avoid those things.

The mindset shifted to puzzle centric from execution centric as execution held solutions and ideas back. So it got eliminated. Older engines still offer that regardless, so the execution centric approach is still available. It is there on the other execution centric engines that things like disjoint unions fit in. NL and Lix chose another path though.

Catering to both on the same engine will result in a mess. You need to do a thing right and commit to it and not do two things unfocussed. This problem can be very serious in software developing!

[Proxima]

A disjoint union level played on an engine which allows editing of each part of the level effectively reduces the level to two (or more) separate levels (which you've stated is your preference for dealing with this type of level), rather than keeping it as the multi-tasking challenge it's meant to be.

Speaking as someone who loves disjoint union levels: when I design one, it is all about the puzzle of allocating your resources between the different sections, and absolutely not meant to be a multi-tasking challenge. You can challenge yourself to play it pause-free if you like, as you can with any level, but it wasn't designed with that in mind and it might be impossible or unnecessarily frustrating.

In response to the rest of your post: I think you have misunderstood what Simon's attitude actually is. We are not trying to gatekeep the community; we are trying to help new members find their feet. That's what Simon meant by "guidelines, not rules". NeoLemmix has evolved a long way from original Lemmings, but because of the surface similarity, new players often don't realise this. Some types of level have been rendered obsolete because this was an absolutely necessary sacrifice so we could have the features that support NeoLemmix gameplay. Clear physics mode allows designers to make really gorgeous-looking tilesets without having to worry that the player will be frustrated by not knowing what's terrain, what's steel and what's background. That's such an important feature for my enjoyment of the game, it's hard to care very much that we can't have hidden-exit levels any more.

No, this doesn't mean that packs have become less varied. Quite the opposite. Some ideas, like the hidden exits and the Floater Frenzies, don't work and get thrown out. In their place, we have a ton of new tricks up our sleeve that the designers of the original games never dreamed of. As you will know from my reviews, new possibilities with the original skills were being discovered as early as MazuLems; and NeoLemmix has more than twice as many skills as original Lemmings, and a correspondingly huge increase in the amount of potential for interesting tricks and interactions.

[∫tan x dx]

I think the distinction between 'good' and 'bad' disjoint union can thought of in terms of 'global vs local' effects. By a 'local' effect, I refer to the results of assigning skills to one of the sub-levels. What takes place within one sub-level has, by definition, no effect on any other.
For example, suppose I dig in one sub-level. The actual outcome of that assignment, the point in time it was assigned, and its location within the sub-level do not affect any other sub-level.
If a level contains sub-levels with only local challenges, then I would this categorise this as a true disjoint union and therefore would agree that it could be split off into its own level.

This contrasts with 'global' effects, such as skill assignments. If one sub-level uses a builder, then another sub-level cannot use that builder. Thus, an overabundance of skills is the issue in this regard. If resources are precious then that could result in an interesting puzzle, in which one sub-level must forgo an obvious solution in order that another sub-level can be solved at all.

To put it simply: local bad, global good.

Now, expanding on the concept of global effects, consider the use of buttons in NeoLemmix. Buttons that open locked doors are global; if three sub-levels each have their own buttons and locked exits, then each sub-level depends on each other level to press all buttons.
So, there are interesting puzzles that can be constructed around this mechanic.
For example, suppose a level has some number of disjoint sub-levels, where the main puzzle is synchronisation: the buttons need to be pressed according to the flow of each sub-level, so that the exits are all opened, and the crowds can all exit safely.