Skill Ceiling

People talk about the depth of games a lot, but it’s tricky to figure out exactly what that means. I’ve been thinking about a new way to measure the depth of games. It’s the Skill Ceiling.

The skill ceiling is basically the answer to the question:

How much can a player theoretically improve their skill at this game before there is no way to get better?


How good can a hypothetical perfect player be at this game? How much better is this than average performance?

It’s easy to measure games on this scale.

Modern Warfare 2 has an extremely high skill ceiling – so high that no human being will ever come close to it. The game runs at 60fps, game verbs generally have little latency, and the game state can change very quickly. It is possible, with excellent tactics and aim, to eliminate entire enemy teams in moments. A theoretical perfect player in that game could singlehandedly defeat an entire team of very skilled human players.

In the middle of the spectrum, Assassin’s Creed 2 has a medium skill ceiling. A perfect player could do significantly better than an normal player, but would not be so astronomically beyond him as in MW2. Many game actions initiate momentary losses of player control while Ezio swings a sword or grabs another handhold, during which the normal player can mentally catch up with the perfect player. And the gamestate cannot change nearly as fast as in MW2. Even with perfect swordplay, for example, it would take at least 50 seconds or so to kill 10 enemies, since the animations to kill enemies take about 5 seconds. A normal player could probably get quite close to this level of performance with practice, since the optimal strategy does not require inhuman reflexes. Ezio’s performance is limited by the game instead of being limited only by player skill.

At the bottom of the spectrum are games that quickly break down into degenerate strategies, like Tic Tac Toe. Game designers who want their games to play out exactly like the envision them often end up with games like this, since there are so few strategies. This is the problem with using quicktime events as skill challenges. Since there is a binary outcome which can be secured by an average player, there is no difference between an average player and a theoretically perfect player.

Toylikes like The Sims can’t be measured by skill ceiling per se since there is no traditional goal or competition, so this concept breaks down when applied to them.

You’ll note a sharp downward trend in terms of replayability as the skill ceiling falls. The reason for this is that replayability ends as soon as a player consumes all content and hits the skill ceiling. MW2 multiplayer is endlessly replayable since there are always ways to get better, even for players who are inhumanly skilled.

There are two ways to build skill ceiling that I’ve thought of:

The first is the SHMUP method, which is to simply demand such accurate moment-to-moment input that no player could ever do it perfectly. The game simply runs so fast that no human mind can keep up. Racing games, SHMUPS, and fighting games depend on this method to a high degree. Pretty much any game sped up to a high speed will start to do this. Imagine playing Tic-Tac-Toe, but you get a third of a second to make your move. As a mental exercise, you can strip out this element of any game by slowing it down. Imagine MW2 multiplayer played at 10% normal speed.

The second is the Go method. This has nothing to do with twitch skills and all to do with managing complex strategic and tactical information. The game simply presents so many options and variables that nobody can easily see where it is going, even if they can stare at is as long as they want. This type of complexity is harder to do meaningfully than simple high frequency input.

The best games use both these methods. Modern Warfare 2 multiplayer, for example, uses both. Starcraft does as well. Both games are very different, but both have vibrant online communities.

Even in Modern Warfare 2, note the replayability difference between online multiplayer and Spec Ops, which really only uses the SHMUP method of building skill ceiling since the AIs are so simple and predictable.

The above theory is descriptive, but not predictive. I’ll be thinking and writing more about how to design games for high skill ceiling in the future.

10 thoughts on “Skill Ceiling

  1. Andrew

    Okay then, so good points all :D neat stuff to put down.

    I take issue that you need to have a mixture of modes to have the best games (your implication being all of the best competitive games have some micro?). However, there are compelling examples for a mix…however, twitch-gaming (micromanagement and the games you listed there) are possibly not the best way of having ceilings and you start to get ceiling only pushed higher by people who literally learn all the muscle memory needed to play…(I take less issue with games having more combinations since many allow specialisation for a purpose; team based play, or similar).

    Then again it’s a tough call. I’m young enough to have a reasonable amount of success in fast games :) yet I deplore the thinking that’s a necessary component to make it the best type of game.

    Accuracy can also come in more forms then just picking the best action in the shortest amount of time after all. Micromanagement might be in fact an detriment in some games.

    For instance Civilization 4 or the Total War series are perhaps good examples.

    Civ4 is definately “Go” style, but the time limitations are based around turns and not real time combat. It has “micromanagement” down to the picking of individual cities worker’s task assignments…It also has apart from too many combinations of strategy to ever have a perfect game, randomisation in the process too (you can’t learn the map, thus the ceiling is higher for learning more about combinations and real estate then choke points etc.). Something added to the mix that Starcraft and CoD (actually, 95% of RTS and FPS games) have.

    Total War adds in a second game mode – you can perhaps excel at the real time combat, but not the turn based parts, etc. The turn based part is much less dense then Civ4 and therefore is probably purely “go” with little reaction/action planning any more then moving large armies around. The teal time part is not micromanagement based I think (large formations need more cohesion then micro control) but succeeds more in the area of time-limited thinking and “accuracy”. It’s more interesting how it merges, in many ways in great ways, RTS and TBS.

    In any case, this is as you said only with competitive games or goal-based ones. There are a large amount of games that can’t fit into that definition well (apart from The Sims :) ). Or ones which generally are great in one or the other category. Some people even do like Quicktime Events purely because they have a low ceiling (or as far as I can see) – but if the intention is to have replayability (as the examples all tend towards) this is a fruitless area.

    I need to play MW2 if you say it is as “go like” as you say it is. However the fact one of the successful strategies actually is using a knife, I do wonder if they just added too many combinations to make it worth trying to get to the higher ceiling levels. Only one way to tell; play it myself, once it is cheaper.

    In any case, I love your definitions/explanations of terms, more, more!

  2. tom

    Im really interested in this area and I agree with much of your conclusions.
    I find certain games like Ikaruga very interesting since they seem to demmand both sorts of gameplay. The scoring structure for ikargua is very strategic and requires puzzle solving type skills. There are other more puzzle based games that also mix the twitch with the cerebral, they generally typify the ‘easy to learn, hard to master’ style of design, which is really interesting to me.

  3. Alexx Kay

    “MW2 multiplayer is endlessly replayable since there are always ways to get better, even for players who are inhumanly skilled.”

    On a purely technical level, I agree with this statement.

    On the other hand, the *speed* of improvement is, IMHO, equally as important as the amount of room at the top. I fairly early reached a point in MW2 where my skill growth over time flattened significantly. I continued to improve, but the degree of improvement per unit time was very small, and not very gratifying.

    Another interesting measure is an individual’s degree of improvement *relative to the player pool*. I can improve my raw performance, but unless I put in a greater-than-average amount of time, my average opponent is improving faster than I am.

  4. Tynan Sylvester Post author

    Thanks for commenting everybody.

    “On the other hand, the *speed* of improvement is, IMHO, equally as important as the amount of room at the top.”

    I agree. I was thinking about this recently. I think it would be very hard or impossible to make a game in which all types of players improve at the same rate. Some people improve at different games at different speeds simply due to varying inherent aptitudes.

    However, you could partially alleviate this by offering different play styles that people can choose to improve on. For example, I know you have trouble with rapid targeting in MW2. If the game had a strong engineer role (TF2 style) which was about building/managing strategic assets instead of shooting, it might allow some people to spend their time improving aiming and movement, while others improve their strategic building skills, but both are still playing the game. You might be able to hold onto a wider variety of players, as long as both the strategic and shooting skill paths were as deep as MW2 is now.

    It would be a very different game, of course, and probably harder to design due to complexity.

    IW obviously decided that they had enough players that they didn’t want to risk alienating them by broadening/unfocusing the experience. In their position, they were right (see sales numbers).

  5. Alexx Kay

    I think there was a relatively low-risk method to alleviate this that they left on the floor. As shipped, the Leaderboard ticker in the MP lobby always reports Kills rankings and Score rankings, never anything else. Yet they *track* dozens of stats. They could have attempted to dynamically find one or more stats for each player in which that player is showing steady improvement, or just really good performance. Instead of continually reminding me how much I suck at the base measures, they could have at least occasionally encouraged me about my accuracy. The “Accolades” system is a step in that direction, but it’s a baby step.

  6. Tynan Sylvester Post author

    True…. I’m really surprised at how rudimentary their stat system is, considering their position in the market, the competitive nature of the game, and the time they’d had to put one together.

  7. Nick Meng

    Some random thoughts:

    The Decision Tree: Wide trees = more decisions, deep trees = more situations. Complexity is proportional to the size of your tree, and in most cases, the skill ceiling should be proportional to complexity.

    Fundamental Limits: First, no matter how big your trees, there are clearly a set of moves that are the “best” moves. Your game is only as complex as the subtree of viable actions. Second, you can only offer so many decisions before it becomes overwhelming for a player; big trees = analysis paralysis.

    SHMUPs and Go in the context of trees: SHMUPS = wide trees, clearly optimal actions, the good actions are physically hard to do. Go = deep tree, with an equally large viable tree.

    Stretching the Complexity Budget:Ultimately, there’s a limit on your complexity based on the kind of game you’re making. How do we build the most interesting tree for the player? Some ideas:
    Bad Choices: You need bad choices in a game (otherwise, how can you have better players). Make the most out of the bad choices added:
    – bad choices should also be good choices in other situations
    – the consequence of bad choices show up deeper in the tree.
    – bad choices appear to be locally optimal
    – bad choices should be rewarding (i.e. taunt kills)
    – Randomizing the situation: forces players to both correctly identify their situation and understand the viable response. Can be employed both in the initial phases as well as during the game (L4D)
    – Randomizing results: Obscures viable decisions, as they still may lead to a non-viable result. Increases the types of situations that will occur.
    Situation Dynamics:
    – Counters: The presence of counters increase the value of information. Counters encourage people who can analyze an opponents play. Counters will make your viable tree smaller. Counters force cycles of counters
    – Hidden Information: Allows for deception and opponent reading
    – Team Coordination: Will increase the frustration level
    – Prisoner’s Dilemma: If cooperation is a choice, betrayal should be viable.

  8. Miroslav

    The skill ceiling does not work in single-player video games.

    People play single-player games until:

    1. they feel they is nothing else to master or that they can’t improve anymore
    2. they feel that further improvement is boring

    But even in multiplayer games, skill ceiling might not be enough. Sprinting, for example, isn’t consider deep (or even fun) even though there is a lot to master.

  9. Miroslav

    Point being that competitive games have a tendency to go from fun to serious simply due to competitive motive. Single-player games almost always don’t, and so people don’t really bother with tests that aren’t fun.

  10. aaa

    “The first is the SHMUP method, which is to simply demand such accurate moment-to-moment input that no player could ever do it perfectly.”

    Shmups aren’t all this demanding and there are many out there, even ones in the bullet hell genre, that are very forgiving. Shield mechanics, small hitboxes, bullet clearing bombs… mistakes are heavily penalized, but input is so responsive and precise in a good shmup, that deaths are entirely a matter of player error.

    See: ChoRenSha 68K

Comments are closed.