00:00:00
If there’s one episode of this show that
I’ve wanted to make, but have never had
00:00:04
the guts to actually attempt, it’s this
one: puzzle design.
00:00:08
Because I love puzzle games - like the time-travelling
platformer Braid, the comedy sci-fi gem Portal,
00:00:14
and the cult hit Stephen’s Sausage Roll
- but I have no idea what actually goes in
00:00:18
to making one of the puzzles for these games.
00:00:21
How do you make something that leaves a player
stumped and scratching their head, and then
00:00:25
makes them feel very smart when they finally
figure out the answer? What makes a puzzle
00:00:29
too hard, or too easy?
00:00:31
And so that’s what I’ve been trying to
figure out these past few months. I’ve been
00:00:35
talking to the creators of great puzzle games,
tried making my own puzzles in different editors,
00:00:40
revisited my video on Jonathan Blow’s puzzle
design philosophy, and have analysed loads
00:00:45
and loads of different levels from different
games.
00:00:48
And in this episode, I’m going to share
what I’ve found out. So this is Game Maker’s
00:00:54
Toolkit, I’m Mark Brown, and here’s what
I think makes a good puzzle.
00:00:58
Part One: The Mechanics
00:01:01
Every puzzle game starts with its mechanics:
a set of ironclad rules that govern how the
00:01:06
game works.
00:01:07
So in a game like Cosmic Express, you can
draw train tracks on a grid - but you can’t
00:01:12
cross over the tracks. One alien can jump
into each train car, and then hop out into
00:01:17
the first box of the same colour they pass
by.
00:01:21
These rules, and perhaps more importantly
these limitations, are used to create puzzles.
00:01:26
The fact that you can’t cross over tracks,
for example, might stop you getting back through
00:01:30
a tight gap - forcing you to find a different
approach.
00:01:33
The overall… let’s say, cleverness, of
the main mechanic will ultimately decide the
00:01:38
number and difficulty of the puzzles in the
game. And so this favours outlandish concepts
00:01:43
like time travel and wormholes - as well as
funky movement controls like those found in
00:01:48
Snakebird.
00:01:49
The way these creatures move - how their body
follows their head, how they’re affected
00:01:53
by gravity, the way eating fruit makes you
longer, which is both a blessing and a curse
00:01:58
- all leads to creative puzzles.
00:02:00
Of course, temporary tools can be used to
augment the main mechanic - like light bridges,
00:02:04
coloured paint, and turrets in the Portal
games. And they can even be used in place
00:02:09
of a main mechanic, as in a game like The
Talos Principle which generates all of its
00:02:12
puzzles from external sources like jammers,
cubes, and repeaters. Mechanics that can combine
00:02:18
together create even more possibilities.
00:02:20
A puzzle game also needs a goal. This is usually
just an exit door, or some kind of collectible.
00:02:26
The important thing is that it’s clear what
the player is trying to achieve. The player
00:02:29
shouldn’t be figuring out what to do - just
how to do it.
00:02:33
Okay. Now it’s time it’s time to actually
make a puzzle.
00:02:35
Part Two: The Catch
00:02:38
So I think a good puzzle is often built around
a catch. Which is a logical contradiction,
00:02:43
where two things are seemingly in direct conflict
with one another.
00:02:47
Here’s the absolutely most basic version
of that idea, just to help explain what I
00:02:51
mean. There’s a door and a button. Standing
on the button opens the door, but when you
00:02:56
walk to the door the button raises back up
and the door shuts.
00:03:00
You need to stand on the button. And you need
to walk to the door. But you can’t do both,
00:03:04
because doing one makes the other impossible.
00:03:06
The solution, of course, is to put a box on
the button. So that’s a really crappy example,
00:03:11
but I think you can find some version of this
conflict at the heart of every good puzzle.
00:03:16
Here’s an example from The Talos Principle,
where, at this point in the game, we’re
00:03:20
using these tripods to route coloured light
beams from these orbs, to these panels - which
00:03:24
makes nearby doors open.
00:03:26
So, after a bit of messing around we’ve
got the puzzle like this. Two tripods are
00:03:30
being used to get blue light to this panel,
and open a door. And one tripod is being used
00:03:35
to get red light to this panel, and open another
door.
00:03:39
But, unfortunately, we need to put one of
those tripods on this pressure sensitive switch.
00:03:44
Now the plan actually seems quite simple.
Send red light to panel C to open this door,
00:03:49
and then use the opening to send blue light
to A with just one tripod, instead of two.
00:03:55
Except... here’s that catch.
00:03:57
You can’t get red light to panel C, without
already having blue light in panel A. So if
00:04:03
you remove either of these tripods, this door
will shut and put a stop to your plan.
00:04:08
Now, resolving a conflict like this can come
in many forms. Sometimes it’s about changing
00:04:13
the sequence of events that led up to the
conflict. Other times it’s about rethinking
00:04:17
your spacial position, perhaps starting the
puzzle from a different location. But there’s
00:04:21
another way, that I think is the gold standard
that every puzzle designer should be shooting for.
00:04:27
Part Three: The Revelation.
00:04:29
So, the solution to that puzzle in The Talos
Principle is to make this tripod connect to
00:04:33
the other tripod, and panel A - even though
the door is in the way. Because when you then
00:04:39
open the door with the red beam, the connection
is made and you can remove the second tripod
00:04:44
without breaking the link.
00:04:46
This puzzle is incredibly simple once you
know the answer - and it’s effortless to
00:04:50
actually execute the solution, which is always
a plus in my book. But it’s still really
00:04:55
challenging. And that’s because it asks
you to think outside the box, reconsider how
00:04:59
the game works, and approach the concept
in a lateral manner.
00:05:03
And, beyond that, it also reveals a non-obvious
- but also totally logical consequence of
00:05:10
the game’s rules that now becomes a part
of your toolbox going forward. And, in fact,
00:05:14
this Talos Principle solution does crop up
in future puzzles as just one part of a larger
00:05:19
conundrum.
00:05:20
So solving the puzzle is like a revelation.
A discovery. An epiphany of some deeper understanding.
00:05:26
And I think that’s, often, where those “eureka!”
moments come from.
00:05:30
Now they can be quite significant revelations.
So in the time travelling puzzler P B Winterbottom,
00:05:35
you’ve got this conundrum where you need
to record a clone to help you pick up pies
00:05:38
in numerical order - but picking up pie three
cuts off access to pie four.
00:05:44
After a lot of messing about, you’ll eventually
realise that clones loop when they reach the
00:05:48
end of the recording. So if you have the clone
start at pie number four, it will appear there
00:05:54
when it finishes its recording and loops back
around. Boom. Revelation.
00:05:59
But often they’re just tiny, subtle things
that you might not even think of as being
00:06:03
important lessons. Like, in Snakebird, where
you need to understand that the bird can change
00:06:08
shape and fall in the same turn, to create
shapes that protect you from spikes.
00:06:13
Now, this is actually a very delicate balance
to hit. Because when you’re asking the player
00:06:17
to think outside the box and do things that
are perhaps not obvious, or not entirely intuitive,
00:06:23
you could leave the player thinking “oh,
I literally didn’t even know I could do that”.
00:06:28
Often after looking up the answer in a walkthrough.
00:06:30
Here’s an example of that from Braid, which
largely has excellent puzzles but there’s
00:06:35
one that stumps a lot of people. So in the
puzzle you essentially need to have an enemy
00:06:39
bounce off your clone’s head, and then you
can bounce off the enemy to jump up very high.
00:06:45
Ultimately, yes, it makes sense. It is a natural
consequence of a game where characters bounce
00:06:50
up when they kill other characters. But for
many, it felt more like a trick than a revelation.
00:06:56
And it really didn’t help that there’s
only one specific moment when it can happen,
00:06:59
meaning players couldn’t easily experiment.
00:07:01
So. Anyway. Lemme give one more example of
a puzzle with a catch, and a revelation.
00:07:08
In Lara Croft GO, there are tiles that crumble
when you first stand on them - and then break
00:07:13
if you stand on them again. And you can use
that to deal with lizards that chase after
00:07:18
you - just lead one over a crumbling wall
tile, and it will fall to the floor below.
00:07:23
That happens in this puzzle too, but if you
go to break the tile, the lizard will kill
00:07:28
you before you can get back. That’s the
catch. The solution is to pre-break the tile
00:07:34
once, then go taunt the lizard, and actually
use the tile’s falling effect to make Lara
00:07:39
fall down, not the lizard. That’s the revelation.
00:07:43
But here’s something else interesting about
that puzzle. This other lizard. It’s not
00:07:48
really part of the solution. You could actually
remove all of these elements and the puzzle
00:07:53
would still make sense. So what’s the point?
Is it just something to waste your time? No,
00:07:58
I don’t think so.
00:08:00
Part Four: The Assumption
00:08:02
The first lizard is actually there, I think,
to trick you into making the wrong assumption
00:08:07
about how the puzzle works.
00:08:09
Because you will use the ol’ walk over a
tile trick to defeat the first lizard, and
00:08:14
most players will assume that they need to
do the same on the second - which leads to
00:08:18
failure. It’s only when they break that
assumption and start thinking about other
00:08:23
avenues, that the solution can be found.
00:08:25
And you can find this sort of cheeky misdirection
all over the place.
00:08:29
Take this puzzle from Stephen’s Sausage
Roll. The goal of this game is to roll sausages
00:08:34
over grills to cook them on both sides, and
like Snakebird, the weirdo movement controls
00:08:39
leads to many tricky levels.
00:08:41
So this puzzle, The Clover, looks really easy.
The player assumes that they can just roll
00:08:46
the three sausages onto their closest grills
and finish the stage. But actually, no, because
00:08:52
doing that means they cannot then manoeuvre
themselves onto the exit.
00:08:56
The developer, Stephen Lavelle, has used an
assumption to walk the player right into the
00:09:01
puzzle’s central catch. And it almost feels
like a joke at your expense, with this moment
00:09:06
being a cruel punchline.
00:09:08
But setting up the puzzle in such a way that
the player will make these wrong assumptions,
00:09:12
actually offers some key benefits.
00:09:15
One is that the player is not completely overwhelmed
when they start the puzzle. Luring the player
00:09:20
into thinking they know how to solve the puzzle
gives them a starting point.
00:09:24
And then, two, while they’re working on
this wrong assumption they’re actually seeing
00:09:29
how the puzzle works and they get to build
a mental model of how this conundrum is put
00:09:33
together.
00:09:34
Three is that it largely ensures that the
player will fail the puzzle their first time.
00:09:39
They’re not going to just waltz into the
solution, but will be carefully led astray
00:09:43
to create that feeling of being stumped.
00:09:46
And four is that it really focuses the player’s
attention on the catch at the heart of the
00:09:51
puzzle. That Talos Principle puzzle isn’t
really about “how do I get to the collectible”,
00:09:56
but it’s “how do I get these two doors
open simultaneously”.
00:10:00
You want the player to be thinking critically
and logically about the situation. And getting
00:10:04
them to walk themselves into the puzzle’s
catch is a good way to achieve this.
00:10:09
So here’s an example of the assumption,
the catch, and the revelation working wonderfully
00:10:14
together, in Snakebird level 10.
00:10:17
So, to finish the level, you need to eat these
two fruits. You’re too short to get the
00:10:22
bottom one, so the assumption is that you
should get the one on the left, go down, get
00:10:27
the bottom one, and then turn around and come
back... except...
00:10:32
You’re now too long to turn around.
00:10:35
So that’s the assumption. Which focuses
us on the catch: that you’re either too
00:10:39
short or too long to get the bottom fruit.
And this forces us to reassess what we know,
00:10:45
and come at the puzzle from a very different
angle - and do this.
00:10:52
Yeah. Not only is it a clever solution, but
it’s also subtly revelatory as it it teaches
00:10:57
you important stuff about how Snakebirds move,
which you can use in future puzzles.
00:11:02
Part Five: Presentation.
00:11:05
Now all of this stuff we’ve learnt so far
can fall apart if you don’t present the
00:11:09
puzzle properly. Check this out.
00:11:11
There’s this really cool puzzle in Portal
2 where a laser beam powers up an elevator,
00:11:16
and a button opens the exit door. It has a
small assumption, where you might think that
00:11:21
you can just release the laser beam, and then
place the cube on the button. But then you’ll
00:11:26
realise that the elevator has gone up without
you - revealing the catch.
00:11:30
You need to use this cube to weigh down the
button. But you also need to use it to temporarily
00:11:35
block the laser beam. Huh!
00:11:37
Now the solution is pretty clever. You need
to place the cube on a light bridge so that
00:11:41
it blocks the laser. Then stand on the elevator
and remove the bridge so the cube falls down,
00:11:48
releases the laser, and lands on the button
- simultaneously lifting the elevator and
00:11:52
opening the exit door.
00:11:54
I really liked this puzzle. It had that revelatory
moment of being like “yeah - I can use gravity
00:12:00
to move blocks from afar”. And while it’s
a very simple puzzle with very few moving
00:12:05
parts, the lateral thinking needed meant it
took me a good few minutes to figure out the
00:12:09
answer.
00:12:10
It definitely took me longer than when I encountered,
essentially, the exact same puzzle in another
00:12:16
game, called The Turing Test. Now it’s not
because I remembered the solution from Portal
00:12:20
2. I played the games like five years apart
and didn’t recognise the set-up at all when
00:12:25
I first played The Turing Test.
00:12:26
No, the reason it’s so much easier in the
second game is because of how the puzzles
00:12:31
are laid out.
00:12:32
So in The Turing Test, the light bridge is
already over the button. You just have to
00:12:37
remove it. Whereas in Portal, you have to
both make and remove the light bridge yourself.
00:12:42
Also in The Turing Test, the button serves
two purposes: it opens one door and shuts
00:12:47
the other. So it’s a lot more obvious that
you need to press it when you’re in between
00:12:51
the two doors. In Portal, you’ve got to
juggle both a laser and the button.
00:12:55
And finally, Portal requires a bit of manoeuvring
to get the cube up on the light bridge, whereas
00:13:00
The Turing Test makes it obvious and effortless.
00:13:02
So you’ve got two puzzles with almost the
exact same concept, but Portal’s presentation
00:13:08
is just so much more effective than The Turing
Test.
00:13:11
I mean, you could make Portal 2 even harder
if you wanted. The puzzle is actually full
00:13:16
of pretty obvious hints like how the cube
starts off being in front of the laser, showing
00:13:21
that you can use it to block the beam. The
only wall you can place a portal on will make
00:13:25
a bridge right over the button. And when you
stand on the semi-transparent bridge, you’ll
00:13:30
immediately see the button right below you.
But, hey, not every game needs to be as hard
00:13:34
as Stephen’s Sausage Roll.
00:13:36
So. Some other presentation tips. I think
a good puzzle is pretty minimalist, with almost
00:13:41
no extraneous elements. If you ask me, the
best puzzles are those that are so small,
00:13:47
with so few moving parts, that you can’t
believe that it’s not more simple to figure out.
00:13:51
A puzzle with too many elements is either
too complicated, or - more likely - most of
00:13:56
those elements aren’t actually part of the
core puzzle and are just busy work that will
00:14:00
frustrate you when you need to reset the level.
00:14:03
A puzzle’s presentation should also provide
clear feedback. Portal has lines running from
00:14:08
buttons to doors, which change colour when
powered up, to clearly explain how the room
00:14:13
is put together. The puzzle is not, after
all, just figuring out how the level is rigged up.
00:14:19
But feedback is also really important when working
with assumptions. There’s a puzzle in Rise
00:14:25
of the Tomb Raider where you make a platform
rise up and then run to the exit - but the
00:14:29
platform drops before you get there.
00:14:32
You definitely don’t want to make it look
like Lara could make it in time if she was
00:14:35
just a bit quicker. Instead, the platform
is positioned significantly far away so it’s
00:14:41
clearly impossible to get there in time - and
the player immediately knows to break this
00:14:45
assumption and try a different approach.
00:14:47
Part Six: The Curve
00:14:50
No puzzle is given to the player in isolation.
Every conundrum is designed to build on top
00:14:55
of the puzzles that came before.
00:14:57
Because if you randomly jumbled up all the
levels in, say, Portal, the game would be
00:15:02
practically impossible for a new player to
get into it.
00:15:05
For one, puzzles use all of the stuff you’ve
learnt so far. From stuff that’s explained
00:15:10
in clear tutorials, to the subtle revelatory
moments I discussed earlier. And secondly,
00:15:16
puzzles should generally ramp up in difficulty
from one to the other.
00:15:20
There are lots of ways to establish a puzzle’s
difficulty, but at Square Enix Montreal, where
00:15:24
they make the GO games, they use four criteria.
The number of possible solutions - the more
00:15:30
there are, the easier the puzzle is. The number
of steps required - more is more difficult,
00:15:35
but too many is tedious. The number of options
the player can choose from at each moment.
00:15:40
And which mechanics the player needs to be
familiar with beforehand.
00:15:44
Those criteria help put the puzzles in a sensible
order - but that’s not to mention some heavy
00:15:48
play testing - puzzle games perhaps need more
playtesting than most other genres, according
00:15:54
to the devs I talked to
00:15:56
So that’s what I learned.
00:15:58
I think a good puzzle is derived from the
game’s rules, and has a catch that makes
00:16:02
the puzzle seem impossible to finish at first
glance. The player can be made to stumble
00:16:07
upon that catch, if the developer exploits
an assumption that the player will make. To
00:16:12
overcome the catch, and resolve the conflict,
the best puzzles ask the player to think laterally,
00:16:18
and uncover a hidden nugget of knowledge about
the game’s rules.
00:16:22
Does every puzzle need to be exactly like
this? No, probably not. But I think you’ll
00:16:27
find that any puzzle worth its salt will have
some version of this stuff.
00:16:32
And puzzles that feel lacking are probably
missing a key aspect. Maybe they have a conflict
00:16:37
that’s too easy to resolve. Maybe it’s
missing the assumption, so many players just
00:16:41
stumble into the right answer. Maybe the puzzle
doesn’t offer enough of a revelation, and
00:16:46
just feels like busywork.
00:16:48
The main thing I’ve learned is that puzzle
design is a very difficult craft, and the
00:16:52
very best examples of the genre require years
of design, iteration, playtesting, and ruthless
00:16:59
cutting. If you’re a developer watching
and you want to make a puzzle game, be prepared
00:17:04
to put in some hard work.
00:17:09
Hey thanks for watching. And a huge thank
you to indie puzzle maker Alan Hazelden, Pierre
00:17:14
Mongrain and Etienne Jauvin from Square Enix
Montreal, and some puzzle making Patrons of
00:17:22
mine, who all took time to answer my questions
about making puzzles.
