Problem-Driven Game Design

In my experience, a problem-driven approach is the only reliable generator of high-quality game designs. The more time I spend focusing on specific, well-articulated problems and questions, the more likely it is that my final design will be novel, deep, and valuable.

This approach pervades every stage of my design process, including the initial choice of what to work on. I won’t even start a project if I don’t have at least one clearly-stated problem or question in mind. Maybe I dislike something about an existing game or genre. Maybe I have a question about some game system that I’ve never seen in action before. It doesn’t need to be an earth-shattering question. It just needs to be well-formed, interesting, and unresolved.

The desire to get rich is unlikely to generate a high-quality design. The desire to create a new game in a genre that I truly love is equally unlikely to generate a high-quality design. The issue is not one of ethics or aesthetics. The issue is that the design space is mind-bogglingly vast and only sparsely populated with high-quality games. I’m standing in a desert, and I need to know which way to go. Neither love nor money helps me in this situation.

Problems provide direction. They point. They’re features of the local environment that say, “There may be water over there.” Success isn’t guaranteed, but this methodology succeeds more often than random chance would dictate, which is not true of any other approach I’ve tried.

An Example

Werewolf is a great game, but there are things I don’t like about it:

  • It requires a moderator and a logistically awkward eye-closing phase.
  • It doesn’t play well with fewer than seven players, and is often long.
  • Being a werewolf is unpleasantly exhausting.
  • Players are often eliminated early and must either leave the room or sit in silence.

Not everyone is bothered by these problems. That’s fine. My unique design sensibilities determine which problems I choose to work on. The important point is that my problems are specific enough and unambiguous enough to suggest a direction.

Indeed, on the day I clearly articulated these problems to myself, I had the following game idea: assign a hidden character card to each player, and then try to discover the owner of a different card each round.

This hypothetical proto-game addresses all four of my problems, to varying degrees. It requires no moderator or eye-closing. It works with as few as three players. It doesn’t put an individual player on the chopping block for an entire game. And if players are ever eliminated, they’re allowed to stick around and keep talking, because they don’t have any game-ruining information.

This isn’t a complete design, but there’s enough there for an initial playtest. We tried it, and the results were promising. Of course, we then had a new list of problems to solve. Good! Problems are our signposts in the vast and trackless land of game design.

We iterated on this cycle of problem-solving and playtesting (with Dave Chalker doing much of the heavy lifting) until we didn’t have any more problems to solve—or at least, until we were able to live with the remaining unsolved problems. The result is a cute little card game called Criminals.

Money, Love, Originality, and Fun

With this concrete example in place, it’s easier to see why the desire to make money is unlikely, in isolation, to lead to good game design. That desire will never help me come up with a specific idea like the one that lies at the core of Criminals, because it doesn’t (and can’t) tell me where in design space to search. The desire may be an excellent motivator, but it’s a terrible navigator.

If I declare that I love Werewolf and would love to design my own psychology game, my desire is also unlikely, in isolation, to help me design Criminals or any other interesting game. There are millions of ways to modify Werewolf, and almost all of them are bad. I need some methodology that actually helps me figure out what to try, and why.

The honest desire to create something original is similarly useless, because it has nothing to say about how to find non-derivitave moves through design space. If the only clear problem I can articulate about my current rule-set is that it’s too much like some existing game, my solution is not to search for ways to differentiate it. My solution is to scrap the project. Otherwise, I’ll end up with a derivative result which differs from its inspiration in ways that exist only for the sake of differentiation, and likely make the game worse. I want an actual methodology that helps me find unexplored, high-quality regions of design space.

Finally, the honest desire simply to create a fun game fares no better than any of these other motivations. If the only clear problem I can articulate about my current rule-set is that it’s “not fun enough,” I’m strongly inclined to return to an earlier, better iteration and branch out from there, or scrap the project entirely. “Not fun enough” is navigationally useless, unless I can translate it into a specific, unambiguous problem-statement that doesn’t include the word “fun”. If I can do that, great! Now I have a problem to focus on which will likely suggest new ideas and directions. Otherwise, I’m just wasting my time.

Closing Thoughts

Why might I design a game?

  • To make money.
  • To pay homage to an existing beloved game.
  • To create something original.
  • To create something fun.

There’s nothing wrong with these motivations, but they’ve never helped me design good games. I’ve always obtained my best results when I’ve ignored these motivations and focused on solving interesting problems and answering interesting questions.

Katamari vs. Peggle

KatamariPeggleKatamari Damacy is a game built around a brilliant core concept, but its overarching goal structure nearly ruins the experience for me. Here’s the brilliant part: you push around a growing ball which can grab literally any object in the game world if it grows large enough. Here’s the annoying part: in each of the main “Make a Star” levels of the game, if you don’t grow your Katamari big enough within a given amount of time, you have to start the level over again from the beginning.

It’s easy to turn a fun, sandboxy pastime into a game by adding a time-pressure to it, but that’s almost always the least interesting way to do it, and it usually has negative side-effects. For instance, Katamari’s final level is 25 minutes long, and most players will need to play through it multiple times in order to win. At the end of my first 25-minute attempt, my Katamari wasn’t even close to the goal of 300 meters. That was okay—at least I’d gotten the lay of the land, and had a better idea of what I needed to do. When the timer ran out at the end of my second attempt, my Katamari was 298 meters large, and rolling at high speed toward a juicy island full of skyscrapers. Oof. Now all I have to do is spend another 25 minutes doing the exact same thing I just did for the last 25 minutes, but a tiny bit faster. This is not good game design.

There’s a more emergent negative consequence of this goal-structure: if you get off to a bad start in a level, you can know before you’re even halfway through that you can’t succeed. In such a case, playing out the rest of the level feels like a waste of time. This is analogous to playing through the last painful stretch of Monopoly when it’s obvious that you’re going to lose. Since Katamari Damacy is a single-player game, I can at least quit and restart without annoying a bunch of other players. But if I’m doing this on almost every level, there’s something wrong with the game design.

The designers of Peggle were faced with a similar design challenge. In Peggle, you’ve got a pachinko-machine screen full of pegs, and you’ve got ten shots to try to hit all of the orange ones. On the surface, this feels nothing like Katamari Damacy, but in fact, the two games bear a deep structural similarity. In Katamari Damacy, you’re trying to collect a given amount of stuff in a given amount of time, and in Peggle, you’re trying to hit a given number of pegs with a given number of shots. The games share a pair of design problems: how do we make it fun to replay levels after failing, and how do we keep players from sometimes feeling hopeless halfway through a level? Peggle’s designers actually managed to solve these problems.

They solved the first problem by randomly distributing the orange, blue, and green pegs at the beginning of each level. The physical layout of the pegs is the same each time—that’s what defines the level—but the distribution of peg-colors within that layout is different each time you play it. You never really play the same level twice, because each replay presents you with a unique situation that you’ve never seen before. Studying the initial layout of a Peggle level is fun in the same way that studying the initial layout of a Settlers of Catan island is fun.

And then there’s the Free Ball Bucket—a steely well that sweeps back and forth across the bottom of every level. The first time I played Peggle, I wondered what it was. After a ball fell into it, I understood, and thought “I wonder why the designers did that?” A few turns later, I realized that it was a brilliant solution to a serious design problem, and is in fact the key to making the whole game work. No matter how badly you’re doing in a level—no matter how many shots you’ve wasted on poorly-aimed or ill-conceived plans—it’s always possible, in principle, that the Free Ball Bucket will save your ass and keep you going long enough to mop up the remaining stragglers. In fact, the more dire your situation, the more glorious your comeback will be if you manage to pull it off. That’s top-notch game design, and like all non-band-aidy design solutions, it doesn’t just solve the immediate problem at hand, but ends up adding strategic depth to the entire game.

There’s no easy way to apply Peggle’s design solutions to Katamari Damacy. I don’t think that randomizing the item layout at the beginning of each Katamari level would improve the game, and while it might be nice to have some way to increase your timer in the middle of a level, it’s still hard to think of a way to do this that’s always capable of providing hope and that improves the fundamental gameplay. I don’t have a good solution to offer for Katamari’s design problem, but my design experience tells me that there there must be a solution—in fact, multiple solutions—out there. The key is to insist on looking for them.

I don’t mean to argue that Peggle is an objectively better game than Katamari Damacy. Katamari Damacy is brilliant. I love the music, I love the art, I love the writing, and I love the gameplay. I just wish the designers had found some way of avoiding the “do it within X minutes” paradigm.

Blockhouse is Now Available

BlockhouseLevel059Blockhouse is now available on the iTunes App Store!

Give it a download and let me know what you think. I’m working on putting up some solution pages here on the website, but in the meantime, if you get stuck on a puzzle, just post a comment here and I (or someone else) will lend a hand.

Also, I’d like to hear which puzzles were your favorites, and why, since I’m already starting to think about what mix of puzzle-types to put into Blockhouse 2. You can let me know which individual puzzles you liked the best, and you can also tell me which categories of puzzles you preferred. I would divide the puzzles into four main categories:

  • Single-block mazes (like puzzles 26-28).
  • Single-polyomino puzzles (like 29-32).
  • Segmented puzzles (like 33-35).
  • Double-polyomino puzzles (like 36-45).

Of course, there are also puzzles that are harder to categorize, like 14-15, 46, 49, etc.

All thoughts are welcome!

Building Blockhouse

BlockhouseLevelsBy the time I decided that I was going to implement Blockhouse as a commercial iPhone app, the game design was already done. Tilt or swipe in one of the four cardinal directions to cause all blocks to slide as far as they can, and try to put them on their respective targets. What could be easier? Many aspects of the aesthetic design and the user-interface were still up in the air, but I had an inkling of where I was headed: chicklet-style blocks and walls, bright candy colors, minimalist interface. Implementing this stuff wasn’t going to be particularly difficult.

I knew that the hard part was going to be actually designing 100 unique puzzles. The only way I could think of to do this was to create some kind of puzzle-building tool that would allow me evolve, tinker, and tweak these puzzles into shape. As it turns out, I spent about six weeks on this task before I even wrote a line of iPhone code. The result was a rather clunky but serviceable OS X app called BlockhouseBuilder.

This was the first time I’d ever implemented a Mac app. Indeed, I’d only been using a Mac since about May of 2008, which is when I’d jumped into iPhone development with both feet. I knew there was going to be a bit of a learning curve. My goal was to make a fully-functioning document-based app, which would allow me to edit puzzles in groups of 100, save them, cut and paste them, undo and redo all editing actions, etc.

None of that was easy, but nevertheless I managed to cobble together an editor that allowed me to create and edit the layout of walls and blocks within a puzzle. It also allowed me to choose the grid-size of each puzzle, and I wrangled for a while about what range of sizes I should allow. Eventually I decided on an aspect ratio of 3:4 for all puzzles. This fit the screen of the iPhone well, leaving some room at the top for UI buttons. It also allowed for a nice range of puzzle sizes: 3×4, 6×8, 9×12, 12×16, and so on.

The really hard part was coming up with some way of displaying the abstract structure of the puzzles I was creating. I wanted to actually see the “puzzle-graph”—a visual representation of every possible position that the blocks could get into, along with the connections between them.

This turned out to be a tough nut for me to crack. There are algorithms for displaying arbitrary graphs, but they aren’t simple, and they often result in winding connections and crossed lines. I wanted all of the orthogonal relationships between states to be retained—in other words, if you can get to one puzzle state by swiping to the right, I wanted that state to appear directly to the right in the puzzle graph. I suspected that someone had already solved this graphing problem—if not for tilt-mazes, then for something isomorphic to them—but I didn’t know enough about graph theory to know where to look, or even to know for sure that what I wanted to do was possible in all cases.

I tried a number of different methods—even some evolutionary ones—but none of them quite did the trick. After working on the problem full-time for almost a week, I started to panic. What if the solution was simply beyond me? I couldn’t afford to keep pouring 8-hour days into it, and yet I couldn’t see how to do what I needed to do without it.

Fortunately, an new idea occurred to me involving a mix of two different techniques I’d tried earlier. I tried arranging all the puzzle positions along horizontal and vertical “struts” or “dowels”, like the beads of an abacus. Each strut contained all the puzzle positions that were connected to each other horizontally or vertically. Each puzzle position was connected to exactly one horizontal strut and one vertical strut. This created a large connected mesh of all the puzzle positions. This mesh was often too compact, overlapping itself at many points, but I could fix that by iteratively pushing the struts away from each other until no two horizontal or vertical struts lay on the same line. The result was a fast algorithm that displayed a flattened, 2D representation of the entire layout of a puzzle, with all orthogonal connections intact. Success!

Here’s a screenshot of BlockhouseBuilder in action, displaying puzzle #16 from the release version of Blockhouse.


The lower left portion of the window displays tiny images of all 100 puzzles and allows me to move between them, change their order, or cut-and-paste entire puzzles from one slot to another. The upper left portion of the window displays a close up of the selected puzzle, which allows me to add and remove blocks and walls, and change their colors. The main window displays the graph of all possible positions that this puzzle can get into, including the starting position (outlined in blue), the ending position (outlined in yellow), and the shortest path between them (represented by the weird black arrows). I can select any starting and ending positions in the graph, and it will instantly show me the shortest path between those two positions. I also created some buttons that automatically find the ending position that’s furthest away from the current starting position (in number of moves), the starting position that’s furthest away from the current ending position, or the longest path between any two positions in the entire puzzle. This was helpful, because in general I wanted the starting and ending positions of a puzzle to be as many moves apart as possible. Finally, I created a button that allowed me to export the collection of 100 puzzles to a property-list file, which I could then include in the iPhone application bundle.

Once I solved the graphing problem and implemented the above functionality, I started adding other features that I knew I was going to need in order to design 100 interesting puzzles. First, I added a bunch of criteria that all puzzles had to meet in order to be exported. Most importantly, all puzzles had to be “connected”. In other words, for any given puzzle, it had to be possible to reach every puzzle state from every other puzzle state. I wanted every puzzle to remain solvable, no matter how you moved the blocks. I did include a reset button in the final game, but it’s always optional. Mostly, it’s there so that people can easily input the online solutions when they give up on a puzzle, but sometimes it’s just nice to get a fresh start on a puzzle when you’re stuck in a rut.

I also implemented some other criteria. For instance, I didn’t want the starting blocks to obscure any portion of their destination footprints. I planned on representing the block destinations using painted marks on the floor of each puzzle, and I wanted these marks to be visible at the beginning of each puzzle. There are plenty of perfectly good puzzles in which two blocks (for instance) swap positions, but I decided to disallow these. Beyond this, there are other criteria that are too obscure to even bother describing.

I didn’t know what percentage of puzzles would still be valid after pruning out all of the ones that didn’t meet my criteria. There was some risk that it would be very low, which would make it a lot harder to come up with interesting puzzles. Fortunately for me, this turned out not to be the case.

After about six weeks of working on BlockhouseBuilder, I was finally ready to actually start working on the iPhone app that would read the property-list file and let me play the puzzles I was creating. The bulk of that work took about a month, and then I spent another couple of weeks integrating the two, adding more features to BlockhouseBuilder as the iPhone app required them.

When it finally came time to start designing all 100 puzzles in earnest, I still felt pretty overwhelmed by the task, even with my fancy new tool at hand. Therefore, I added an “Evolve” button to BlockhouseBuilder. This button caused the program  to create thousands of variations of the layout of the current puzzle, looking for various criteria, like a particular ratio between the longest path and the total number of positions. With the click of a button I could have the program spit out a random variation of the current puzzle that was “better” in some way. Determining what counted as “better” was a black art, and I didn’t bother building a UI to tweak these values, because I had no idea what variables I should even be tweaking. Instead, I just kept changing the actual code for the “Evolve” button, recompiling BlockhouseBuilder, and playing around with the results. So I’d set up the basic situation—the size of the grid, the shapes of the blocks—and then I’d start hitting this button over and over again, cranking out a ton of evolved variations. Of course, on top of this I did a lot “artificial selection” (i.e. tossing out results that were unacceptable for various reasons), and manual high-level tweaking.

It was kludgey, but it worked. Exploring the puzzle-space in this way was fascinating, and because of the evolutionary nature of the process, I didn’t automatically know the solution to each new puzzle that I came up with. Not only did this allow me to play (and enjoy) hundreds of these puzzles myself, but it allowed me to directly test how difficult various puzzles were. It also allowed me to answer a few questions I’d been curious about for years. For instance, what’s a good ratio of path length to total number of puzzle positions? I found that it was somewhere between 1:2 and 1:3. In other words, if it takes 15 moves to solve a puzzle, I wanted there to be about 30 or 40 total positions in the puzzle. How hard do puzzles get as you increase the total number of positions? I found that 100 total positions is usually too hard. Most of the puzzles that I included in the final product have between 30 and 80 total positions. On the other hand, there isn’t a one-to-one correspondence between total number of positions and difficulty. I found some 40-position double-block puzzles that I was incapable of solving after multiple hours of trying.

As I explored the space, the puzzles I came up with fell into four distinct categories, each of which had its own flavor. The most basic type of puzzle contained only a single square block. These puzzles were only interesting on grid-sizes of 9×12 or larger, and I decided that they should all be explicitly maze-like. As it turns out, my evolutionary process wasn’t very good at coming up with maze-like puzzles, so I had to do a lot of manual editing on these. Another type of puzzle contained only a single polyomino. These puzzles were also only interesting on grid-sizes of 9×12 or larger, but unlike the puzzles in the previous category, these puzzles didn’t work well as mazes. Instead, they worked well in more open spaces peppered with single-square walls. Another type of puzzle consisted of two or more single-square blocks, each in its own walled-off compartment. These had a unique flavor, and were too difficult if larger than 6×8. The fourth type of puzzle consisted of two polyominoes together in the same compartment. Most of these puzzles were way too difficult if larger than 6×8, but they became easier if the blocks were large, because large blocks fill the space and cut down on the total number of positions possible. This was my favorite type of puzzle, and they make up about half of all the puzzles in the final product.

All in all, it was a fascinating project, and I hope some people get a kick out of the result. You can count on seeing Blockhouse 2 at some point. The puzzle-space is huge, and I’ve got BlockhouseBuilder burning a hole in my pocket…