If you click on a link and make a purchase we may receive a small commission. Read our editorial policy.

Specially Level With Me, Adam Foster - Part Two

GLaDOS To Be Here

In this very special installment of "Level With Me", Robert Yang speaks with former celebrity modder / current Valve level designer Adam Foster about his work and process. In Part 2, we cover his work on Portal 2, game architecture, and how to design for players.

Robert Yang: Let's talk Portal 2. It didn't have any health-kits or power-ups, so how did you go about guiding players without those breadcrumbs?

Adam Foster: The game often worked best when it kept to the basic structure of the 'test-chamber' - walking through a door, GLaDOS says a funny line, you complete the test, GLaDOS says another funny line... It's that constant, near-ritualised loop which can be subverted now and then, but you have to show the normal routine before the abnormal becomes surprising.

We had to scatter a ridiculous number of clues through the game to quietly help players through the puzzles. Sometimes they would be pretty obvious, such as flickering lights, other times there'd be a gentle gradation of lighting colours - a cool, blue at the beginning through to a warm yellow at the end. But yes, there'd be no obvious supply caches to confirm to the player that yes, they were going the right way. Instead, the reward could be GLaDOS speaking, and getting to see the next test.

We did experiment with much less linear, more natural exploration in some of the underground areas, but everyone got instantly, irretrievably lost.

RY: Oh, I remember when I first got to the Portal 2 underground levels, it felt so Adam Foster-y to me. I don't know if it was the scale or the proportions or what.

AF: One thing I did was help with finding reference materials that eventually gave rise to many of the environments seen in Portal 2. Things like the destroyed modern Aperture (heavily inspired by some photos of a collapsed, abandoned Soviet space shuttle I found), the grid-like nature of the surrounding infrastructure (a combination of the innards of rocket test hardware, industrial architecture and deep inside CERN) and the deep underground sections (power station cooling towers, neutrino observatories, salt mines and so on).

(left: Portal 2 concept art -- right: the Large Hadron Collider)

Having a vague scientific background was ridiculously useful. I'd remember something interesting I'd seen or read about years previously, then go looking for imagery - like for the neutrino stuff. Occasionally I'd hit the jackpot, as with the CERN photo archives. They've got perhaps hundreds of thousands of pictures online, going back to the 1950s - a near-complete evolution of particle physics research, along with incidental details like office hardware, architectural styles and the fashion stylings of the scientists and personnel involved. So many wonderful things.

(left: an underground-themed Portal 2 level -- right: the remains of the "Buran" Soviet space shuttle program)

Then, we went exploring NASA's archives. The process continued...

Design-wise, I've got really used to sketching stuff out in Hammer to get an idea of shape and scale. Huge blocks of featureless geometry to figure out sight-lines, perspectives, angles, form, lighting and whatever. It was a little surreal seeing screenshots of a couple of these quick Hammer-sketches printed out and stuck on the wall of the Portal 2 art room, surrounded by artwork from some seriously talented actual-artists.

(left: a geodesic sphere-shaped test chamber in Portal 2 -- right: Buckminster Fuller with a model of a geodesic dome)

Once we'd nailed down a style, I then went on to bash holes out of formerly pristine test-chambers as part of the visual-pass stuff. Some of the initial training puzzles I worked on are closely based on those from the original Portal, so at certain points, I was taking this much-loved game out into a darkened alleyway - before kicking and tearing it apart into corroded heaps of twisted metal and asbestos panelling. I felt like a complete monster!

RY: ... But you loved being a monster, right? All level designers love two things: (a) aimlessly flying around in the level editor, and (b) destroying levels and dropping rubble everywhere.

AF: At one point we were figuring out the way destroyed test chambers would twist and distort with the aid of brown paper bags and sweet wrappers...

(left: a ruined test chamber -- right: a paper bag)

RY: Given your work on MINERVA, did you have any input into GlaDOS's character? What, do you think, is it about omnipresent female cyborg constructs that appeals to players?

AF: Not directly, but I think a lot of the environmental design stuff we were doing rubbed off on the writers. Storytelling at Valve is remarkably fluid - while the writers will request particular scenes and locations and make many, many suggestions, they'll also absorb all kinds of stuff from concepts, gameplay and other designs. It's very much a two-way process!

As for the AI GLaDOS and the non-AI Minerva, I suspect a fair number of players simply quite like being ordered around. I was raised by feminists, so it's been a pretty standard part of my existence.

RY: Yeah, I liked how Portal 2 is about two women and the men / system who conspire to exploit them. There's that part where GlaDOS mocks Chell for being fat, and it offended some players who read it as the game condoning bullying or fat-shaming... when I thought it was clear that the insults were petty and unfunny, that the game actually condemned that offensiveness in this complex ironic way.

AF: If Portal 2 condoned such petty bullying, does that mean Half-Life 2 spoke out in favour of brutal, oppressive dictatorships?

Right now, I'm imagining a game where some character keeps congratulating the player in a wonderful, wholesomely glowing manner - for doing stuff which, when you scratch past the surface, has some utterly disgusting undertones. Actually, that would be pretty nasty - even if the underlying subtext is still that it's very much a Bad Thing. For once, could insults take the edge off it?

RY: But what if the player is an impressionable earnest sweet innocent child? We can't choose what our players comprehend or think. (… Yet.)

AF: At one point we were experimenting with architectural glitches, with repeating sections of test chambers, misaligned sections and three-dimensional compression artefacts. Unfortunately, bugs in transcription by self-replicating robots looked just like bugs in the game.

RY: Ahh, that sounds great! I can see that as a bug and poor design, but part of me also thinks, “you're letting the least literate players decide what the most fluent get to see.”

AF: Before I was hired at Valve, I visited their offices and playtested a near-finished version of the first Portal. I got painfully stuck on one of the very first puzzles - it must have been agony for the Portal team watching. At that point I realised everyone is potentially an idiot when playing games - they can misunderstand or fail to spot some vital clue or control or gameplay hint, and everything becomes frustrating.

If something in a game is actively, unintentionally confusing people, then the idea's not being expressed properly. So while Portal 2 ended up with loads of stuff which only super-observant players would notice, we had to be super-careful that it didn't break puzzles for those who weren't looking quite so closely.

RY: That's what's great about being super-observant, you focus on some stuff and overlook a lot of other things. Like, this puzzlemaker level, "Test 07: Concentre d'espace" – yes, it's possible to get stuck, but I still love how elegant of an idea it is. It's a level that's more than the sum of its puzzles.

AF: In the screenshot, it was like seeing the insides of a tiny, compact piece of engineering. Every last voxel packed with hardware, this three-by-three cube is quite ingenious. A tiny little state-machine of puzzling.

Of course, I inadvertently trapped my fingers in the metaphorical machinery - lifting up the reflecto-cube through the grating, later using it for something I shouldn't have, getting myself completely wedged in the puzzle. I had to restart. Oops.

RY: Yeah, the same happened to me. Seems like a huge flaw, the more I think about it, but I don't really think about it. Maybe that means I love the idea of it more than the actual designed thing.

AF: Something I really appreciated on the first Portal 2 DLC in making a good puzzle was having that central, interesting idea to build everything else around. I figured out the basis for the 'triple-axis' map - requiring a bridge, tractor-beam, laser and reflecto-cube to be clamped in mid-air in order to power a lift to the exit. Every other gameplay element in the map is simply to support that moment.

Cover image for YouTube video

Portal 2 is such a pure 'game' in so many respects that the environmental design works best when it alludes to that fact. The whole fiction is constructed around super-streamlined testing of scientific participants - the sleek test-chambers attempt to minimise external distraction. Introducing external details can potentially derail everything, but done with care, subtle not-quite-right aspects can appear hugely important. See the rat-man rooms from the first game, for example.

RY: But these important subtle details are actually unimportant, in that they're not tied to progress through the game?

AF: For me personally, the non-game aspects of a game can be the most memorable - exploration, plot revelations, general atmosphere and so on - but again it all needs that central, interesting gameplay idea to rest on.

RY: I'm similar. Don't you think that's weird, or even a little bit crazy? That we make and play games just so the non-game aspects can live in our minds?

AF: For me, 'winning' is pretty uninteresting. In a big, open-world game, I'll often be deliberately sabotaging my character's actions and choices in believable ways - to create interesting new events and possibilities, subverting the game for new storytelling opportunities.

RY: That's what I like about your levels, you apply that sensibility to your layout. You use a lot of symmetry, like a spine runs down the body of the level. Few games use symmetry outside of monolithic architecture, I think, because symmetry seems so “designed” and thus "unreal," so we always have jagged concrete ruins with 1 of 10 lights being flickery.

AF: Something I'd get a lot when editing geometry, especially terrain, would be Rorschach-style ink-blot mirroring of random noise. Vague blotches in the textures would bring up butterflies, grinning devil-faces, all kinds of features.

I suppose adding symmetry to a pretty arbitrary lump of gameplay-oriented geometry can give the resulting structure much more of a sense of being 'designed', with otherwise arbitrary shapes and details taking on a new importance. Then, because of the Combine aesthetic of distinct asymmetry, you have to go back and mangle things up again, with the underlying symmetry still suggesting some deeper meaning...

RY: What makes a level seem "real" or plausible, or is that even important? Like, I'd argue a Doom 1 or Marathon level is more coherent / plausible than a Modern Warfare level -- it's not about perceived plausibility of details, it's about the thoughtfulness of the layout.

AF: I imagine some people mistake my weird approach to design as building entire levels before shoe-horning gameplay into the architecture, whereas to me, it's more both the plausible design and gameplay-oriented features going together simultaneously. Gameplay might suggest a lookout point at a certain position, then the world design will give stern recommendations as to what it should be. Or the world design might suggest a certain feature, then gameplay leaps on board and says 'yes please'. But that believable world design is very, very important to me.

("E2M2: Containment Area", Doom.)

As for simpler layouts in modern games, it might partly be due to design programs changing from 2D overviews to more fully-fledged 3D systems. It would seem logical to make an attractive architectural plan when all you have to look at is a 2D overhead view, especially when the player is going to bring that same 2D overview up using the in-game map feature.

RY: If mini-maps affect the complexity of a level, then I think the side-step is to make the mini-map more vague. Thief's maps were sometimes these symbolic scribbles, which is convenient given how big and complicated those levels were.

AF: When there's no in-game 2D map, then perhaps it's how it looks in 3D that's what matters most? But I'll still be building things with vast, overall structures only truly visible from the outside...

RY: I imagine you really like the Portal 2 puzzle maker and looking at all the levels from the outside, then.

AF: I've barely had a chance to look at any custom maps for Portal 2, sadly. I need to - I played about with the simple editor just before release, and really, really wish we'd had it for prototyping gameplay ideas in the original game. I love Hammer [(the complex level editor)] and all it stands for, but the simple editor is incredibly important - hopefully it introduces many more people to this fascinating world of level design.

RY: Last time I tried collaborating on a Portal 2 level during an interview, I wasn't interviewing someone who worked on Portal 2... So, um, here's an awful level that I made. It starts with the player walking in and seeing a turret staring right at them. (Let's break all the rules, whoa!)

AF: First time I played it, I died horribly. Second time, almost as horribly. After loading it up in Hammer to mix things up again, first compile had the turrets kill me again. I like the idea, but the door locking behind the player as soon as the turret starts firing is just a little too cruel?

RY: Yeah, your placement works better. The main idea of the level was that the solution should involve undoing progress toward the solution. So many Portal puzzles with multiple cubes typically let you leave them somewhere forever.

AF: About the puzzle - if anything, I felt the cube powering the tractor-beam was too strongly locked in position. It took me far too long to realise I had to carry it past the piston. Perhaps just two cubes total would work better?

RY: I like how you put it: "too strongly locked in position"... this came up in the other Level With Me installments, a discussion of the "economy" in a Portal puzzle. You always assume you have exactly what you need. Are extra pieces poor design, or not hewing to a "right kind of puzzle"? It depends on how you see your player. Are you their benevolent puppet-master, their pleading parent, or their tired spouse, or what?

AF: 'Annoying sibling', probably. Showing off some weird thing I find terribly interesting!

MINERVA had a section in which the player's weapons were taken away, and they had to run past various locked armouries full of tempting stuff. Some players seemed to think the game had broken, that they should be able to get into those armouries straight away - subverting the game's language is one thing, but you have to reassure the player somehow that no, the game is operating correctly. Unlike most other modes of fiction, games can and do break down at times.

Early in Portal 2, there's a section where you have to portal from one ruined chamber to another, in order to catch a dropping Wheatley. A fair few playtesters simply didn't get their heads around this, and spent ages searching for a puzzle to solve in that first chamber. So Wheatley's dialogue became remarkably insistent on the actual, seemingly trivial solution.

I've also made a couple of other tweaks to the map: the grating in the middle has been changed to glass with a visually-cube-pickup-allowing slit down the middle -

RY: Wait, I don't understand the purpose of the slit -- you don't really need it to solve the puzzle. Or is it a red herring to get the player to put the cube in the tunnel?

AF: I... think I must have found an alternate solution?

RY: What? Oh... oh... right.

AF: … And the distant button has gained some invisible trickery to encourage the box to land on it. Hoopy the Hoop is now there by default.

RY: (Hoopy the Hoop for president!) I'm a big fan of “invisible trickery.” My favorite Portal 2 level entity is probably info_placement_helper [a special invisible doodad which snaps portals to a specific point and angle, an aim assist].

AF: Many important portal-flings had invisible catapults on the exit surfaces, lasers would automatically create placement helpers when they hit portalable surfaces, cubes being flung would funnel themselves through portals. Kind of sneaky, but Portal is more about the puzzling than pixel-perfect accuracy. Solving a puzzle, then having the solution fail because a fling fell slightly too short, or a cube caught the edge of a portal, tends not to be much fun. Skill can be one thing, random failures are another.

RY: And now, this is weird. I want to work on this more with you, but I feel like just calling it done. Usually there's always so much more to add and fix. How do you (and/or Valve) decide when it's done, or do you just wait for a cathartic epiphany accompanied by angelic choir?

AF: Puzzle design could be considered 'done' when most playtesters would find the puzzles fun, and make it through without getting irretrievably stuck. That last one is important - it's absolutely fine to take a while, but if someone's spent an hour trying the same, unsuccessful thing over and over again without spotting the real solution, that's not good. Getting the puzzle immediately is bad, but it still has to be solveable.

The path-hinting thing can be terribly subtle. The triple-laser puzzle in the single-player game used to be one of the hardest, with people inevitably getting permanently stuck - but then suddenly, with some incredibly subtle changes in the cubes’ initial positions, it was fixed. We don't really know why.

Cover image for YouTube video

The usual route when dealing with a puzzle that seems too complicated is not to change it directly, but modify previous puzzles so that they vaguely hint at particular mechanics.

With Portal 2, we'd have these pure, bare puzzles, which got slotted into the plot at appropriate points - with artwork and audio and more playtesting and signs and lighting and performance fixing and so on... And eventually it's time to fix the bugs, tidy it all up and ship the game.

RY: Okay, one last question, totally unrelated to the subject of shipping games. When's Half-Life 2: Episode Three coming out?

AF: ^H*DSd]]/&s Wlb-%£--NO CARRIER

RY: Thanks for your time.

AF: No problem.

You can play the resulting Portal 2 level, "Hoopy the Hoop", if you really want to.

Rock Paper Shotgun is the home of PC gaming

Sign in and join us on our journey to discover strange and compelling PC games.

Related topics
About the Author
Robert Yang avatar

Robert Yang


Robert Yang is an artist, writer and game developer, and makes surprisingly popular games about gay culture, such as The Tearoom, Rinse And Repeat and Radiator 2. Previously, he taught at NYU Game Center.