[In this complete and comprehensive article, Rob Jagnow of Lazy 8 Studios (developers of Cogs) take you behind the scenes of the Portal 2 ARG which lead up to the early release of Valve's highly-anticipated sequel.]
Way back in March of 2010, Valve added a mysterious new achievement to Portal. In no time at all, players figured out that if they carried the in-game radios to specific locations within each level, they could change the radio transmission. In the parlance of Alternate Reality Games (ARGs), this is referred to as a "rabbit hole" -- a point of entry into a much deeper universe.
From the crackling radio transmissions, players managed to decode embedded images. One clue led to the next to the next, sending players on a massive scavenger hunt that included dialing into an old-school BBS and retrieving cryptic ASCII images, culminating in the announcement that a sequel to the popular game was in the works.
The puzzles were challenging, but the whole sequence, from beginning to end, only took the community of players a few hours to solve.
Fast-forward to December, 2010. Following the success of the prior ARG, members of the Portal 2 team suggest that they'd like to do something similar in the lead-up to launch; only this time, they'd like to wrangle the creative talents of some of the indie developers with popular Steam titles.
The next day, an invitation goes out; a week later, on December 16, 2010, 20 indie developers fly in to Seattle, all somewhat confused as to why they are there in the first place.
With everyone gathered in Valve's sixth-floor fishbowl conference room, Gabe Himself lays out the core idea: Somehow use our Steam games to usher in the arrival of Portal 2. That's it. There's no agenda and no grand plan. Just pack a room full of fun, creative people and give them unrestricted access to Valve's resources, including Portal 2 assets. No constraints. No NDAs. This was a project built on trust and mutual respect.
During our two days at Valve, we formulated the minimalist story that we would use to drive the event: GLaDOS has infiltrated our games and she slowly makes herself visible as she seeks to make her comeback. New content will be added to the games in three phases, starting with only vague mysteries and working up to GLaDOS making an explicit appearance. To make it easier for hard-core fans to participate in every aspect of the ARG, we plan to release all of the participating games for 75 percent off in a bundle called The Potato Sack.
We schedule enough time between the content releases to allow players to solve the puzzles and to give the media time to pick up the story. The event is designed to culminate with the crescendo, in which players will spend time in each of the participating games to roll back the Portal 2 launch clock.
The Portal 2 ARG is a great example of Valve's philosophy of digital distribution: let's give new value to our old games in a way that engages our existing fans and draws in new ones. This philosophy sets Steam apart from other portals, where content updates are difficult, if not impossible.
I suspect that the ARG converted quite a few pirates to legitimate customers. If they want to have access to the new content and participate in the metagame, they'll need to own the game, and have an authenticated connection with the Steam servers.
Everyone wins. Valve draws attention to the Portal 2 launch. Indie developers, who don't have the benefit of a multi-million-dollar marketing budget, draw attention to their titles and benefit from the Potato Sack sales. Old fans get hundreds of hours of new content in their favorite indie games, plus a metagame built on top of it all. New players get all of this, wrapped up in a package discounted by 75 percent.
To help the teams coordinate from our offices spread across multiple continents and time zones, Valve set up a secure wiki where we could share our ideas and critical information that was needed by all of the teams. This is where many of the game and metastructure puzzles were formed. In addition, we had a high-traffic internal mailing list that hit around 150 messages a day during the height of the ARG.
There were no conference calls and no IRC chats. Other than the meeting in December 2010, the only time we all met in person was during a March 10 meeting in Seattle where we finalized our plans and made sure everyone was on the same page for the April 1st #PotatoFoolsDay launch.
Jeep Barnett at Valve spearheaded the ARG metastructure by combining ideas from the wiki into a mega puzzle and breaking it into components that could be spread across all the games. The first revision was completed on February 3.
Phase 1. April 1, 2011, every developer will release new content tagged in the update notes with #PotatoFoolsDay. Hidden somewhere in each game will be a glyph with a consonant equivalent and a nonsense sentence with exactly 16 consonants. Taken together, this data can be used to create 13 images with the letters "COLLABORATION" -- a password that will be needed in Phase 2.
Phase 2. April 7, games are patched with subtle clues that may be construed as being related to Portal 2. Each game also has a hidden password and an action that players can perform to open an Aperture Science login page. If the player enters the correct password from another game, a .zip file is transferred to their computer.
Inside the .zip files are Portal 2 concept art. Hidden in the alpha channel of the images are black-and-white photographs of real-world Seattle locations that will need to be connected together in Phase 3. Also in the .zip files are data chunks that can be sewn together into a larger .zip file, password protected with the previously collected password, "COLLABORATION". Inside is even more great concept art and other Seattle locations.
Phase 3. On April 12, games are updated yet again. This time, GLaDOS makes prominent appearances and the ARG community really starts to pick up. Each game has one or more actions that can be performed to open an Aperture Science page where GLaDOS speaks a peculiar sentence that references two of the aforementioned Seattle locations.
When the locations are connected together accordingly, they spell out the word "nelipot". This, in turn, is the name of a Steam group where users can find some Portal 2 screenshots and a QR code that points them to the [email protected] page and thus the start of the crescendo.
Crescendo. Originally scheduled for April 16 or 17, the crescendo was designed to give players an opportunity to play all of the participating games to chop time off of the Portal 2 launch clock. In a last-minute decision, we decided to move up the start of the crescendo to give more time for international fans to participate and also to give more time for the media to key in to the event. In retrospect, this was a bad decision. I'll discuss this later.
Launch. Based on the efforts of the players, Portal 2 launched at 9:29pm on Monday, April 18 -- nine and a half hours earlier than the originally scheduled Tuesday morning launch.
The ARG metastructure ended up working well for both designers and players. For the designers, we had a roadmap of where to go, but the constraints were loose, so we were free to pursue puzzle ideas that are suitable for our own games and consistent with the Portal 2 fiction as well as the fiction of our own game universe.
This resulted in an astonishing variety of challenges, including story that spanned far beyond the metastruture requirements. For instance, the Dejobaan team had an expansive collection of blogs, tweets and videos that revolved around a whole new set of fictional characters with clues that tied loosely back into the main ARG.
For the fans, being able to piece together the metastructure limited the space they needed to explore and helped save them from pursuing too many red herrings. For instance, once it became clear that in Phase 1, each game contained one glyph and one nonsense sentence (something that was further hinted at by a brief image in one the Cave Johnson promotional videos), the players were able to focus their efforts on games where they had not yet found the clues.
One potential drawback to the weak constraints of the metastructure is that it didn't encourage the developers to collectively advance a strong fiction alongside the ARG puzzles. In hindsight, the ARG players seemed happy to have such a huge variety of puzzles and were very forgiving of the weak fiction. Rather than filling in a back story of GLaDOS, we chose to make her a commentator within most of our games, mocking the player in her characteristic style.
The indie developers created far too much content for the ARG to cover it all here. Thankfully, many of us took the time to talk about our puzzles on our own blogs. If you're curious (and don't mind a few spoilers), follow the links here:
- Toki Tori and Rush: http://twotribes.com/message/what-we-did-for-the-portal-2-arg/
- BIT.TRIP BEAT: http://www.gaijingames.com/?p=2255
- The Ball: http://www.theballthegame.com/?page_id=1219
- Cogs: http://www.lazy8studios.com/2011/Portal2_ARG_puzzles
- Defense Grid: http://www.hiddenpath.com/blog/18/
- Killing Floor: http://www.killingfloorthegame.com/potato-arg-a-kf-postmortem/
- Amnesia: The Dark Descent: http://frictionalgames.blogspot.com/2011/04/portal-2-arg-postmortem.html
As is often the case with an ARG, the players set up their own wiki where they could track the community's progress and coordinate efforts on specific puzzles. A few individuals made an impressive feat of keeping the community organized. In fact, the developers used the fan wiki extensively to keep abreast of progress. I would guess that none of the individual developers knew about more than 20 percent of the total puzzles, so we were often pleasantly surprised to learn about fun content from the fan wiki.
Due to the high volume of traffic, the wiki had to twice be relocated to other servers. In retrospect, we think it would have been a good idea for Valve to offer to host the bandwidth-intensive servers. On one hand, this type of offer can compromise the fiction of the ARG. On the other hand, I think the player community would have really appreciated this assistance.
Fans also used IRC channels to coordinate their efforts. During the later stages of the ARG, it was common to find more than 500 individuals in the main IRC channel.
The fan-maintained resources were impressive, but despite that, we think that the ARG may have benefitted from an official Valve-hosted webpage where players could check in to stay abreast of any progress. If done right, a clearinghouse page could serve multiple purposes:
- Hint at when players are spending too much time chasing red herrings.
- Key players in to when they're making good progress.
- Provide a central location where new players can get quickly caught up on everything that's happened so far with screenshots, videos and other easily digestible summaries.
- A gallery of fan content.
- Allow press to easily stay abreast of what's going on.
We owe the ARG community a lot of credit for enforcing limits on their own behavior. At one point, the actions of a few hackers threatened to bypass a big section of ARG content. A few individuals managed to crack the system that Steam uses to distribute beta versions to developers for testing, giving them early access to the content. Thankfully, Valve quickly caught on and pulled the content down. At the same time, some of the indie developers hastily beefed up their own in-game security measures. This quick response sent a clear message that this type of hacking was unexpected and unacceptable.
When leaders in the ARG community figured out what was going on, they sent an official apology to Valve and issued a statement on the wiki that this type of behavior was unacceptable. While some amount of hacking continued, this scolding helped.
Over the course of the ARG, we observed several types of player behaviors. When planning future ARGs, I think it's worthwhile to take these player types into account to try to design puzzles that target each audience and also to try to predict player behavior.
Librarians. These players had little interest in actually playing the games. However, they were intensely focused on the metagame. This is the type of player who organizes the wiki, directs conversation in the IRC channel, and collects screenshots from other players.
As soon as it became clear that #PotatoFoolsDay was an ARG, these are the types of players who quickly came on board. It took longer for each of the indie teams to motivate our core fans. Some of the librarians were instrumental in advancing the ARG, yet they didn't own any of the games or earn a single potato.
Gamers. These are the core fans of an indie game who were thrilled to offer their player expertise to help advance the ARG. It took a while for us to draw their attention to what was going on, but once they came on board, their familiarity with the games was essential for spotting suspicious changes that would otherwise escape the attention of a new player.
Hackers. For a lot of ARG players, their first instinct is to hack. Hackers enjoy decompiling the code and searching the executable binary for plain-text strings that may provide hints. In some ARGs, light hacking has often been a legitimate and expected way to move the game forward. It's not unusual, for instance, for an ARG to hide an important clue in an HTML comment so that the player needs to view the page source to advance. This precedent sometimes makes it difficult to draw a clear line to indicate to players what is and isn't acceptable behavior.
Collectors. In the second and third phases of the ARG, players were awarded a small image of a potato in their profile for every puzzle they successfully solved on their own. This incentive drives a significant number of players to seek out walkthroughs that show them exactly how to earn every potato. As with rare hats in Team Fortress 2, limited-time collectibles provide a surprisingly strong incentive for this group.
Take-a-Hikers. In the first wave of ARG content, one of the games written by Two Tribes revealed the GPS coordinates of their Dutch office in a clue. The glyph and nonsense phrase were printed on a piece of paper and taped to a light pole about 3m off the ground. Once this established that some clues might involve physical locations, some players eagerly sought out any opportunity to explore the real world, for better or worse. This sometimes mean that they spent hours driving to a place, only to conclude that they were chasing a red herring.
The second and third rounds of content pointed to a slew of physical locations in Seattle, but in retrospect it may have been a good idea to target more content at the Take-a-Hikers. Clues hidden in the real world often make for great stories and powerful impact. For instance, the guy who climbed the pole at the Two Tribes office instantly became an ARG celebrity.
Even a player who chased after a red herring had a good story to tell. One guy got the idea in his head that he was supposed to find a clue at The Couch Potato, a furniture warehouse in Santa Cruz, California. Team Meat learned about this by monitoring the IRC channel, so Edmund McMillan of Super Meat Boy fame intercepted him at the warehouse and surprised him with a signed copy of Super Meat Boy.
Thanks to the decentralized nature of the puzzle development, the Portal 2 ARG had a truly astonishing variety of puzzles. For the sake of brevity, I'll attempt to categorize the puzzles to talk about what worked well. For specific descriptions of the puzzles, I suggest further reading at the fan wiki and the previously listed postmortems written by the individual developers.
Cryptography and Steganography. Cryptographic puzzles are a mainstay of ARGs. They can have a wide range of difficulties and may require players to apply class cryptology, data compression and analysis techniques to translate data into something useful. Encodings may use Braille, Morse code, ASCII, zip, bit manipulation, image manipulation, frequency analysis, etc.
Cryptographic messages may be used in tandem with steganography, which literally translates as "concealed writing." Codes may be hidden in text, images or audio files. For instance, the Seattle location images revealed in Phase 2 were hidden in the usually-invisible alpha channel of other images.
Expert Puzzles. These are puzzles that require a lot of domain-specific knowledge to solve. They are designed to be solved by one individual who is an expert in a particular game. Expert puzzles may require a player to observe a subtle change in the game universe or to solve a puzzle that would be difficult or impossible for a novice player.
These puzzles work well for leveraging a game's core fan base, which is uniquely qualified for the challenge. These types of puzzles tend to get completed fairly quickly with little fanfare.
Community Puzzles. Some puzzles require a large community working together in unison. For instance, the physical locations that were revealed in Phase 2 needed to be accurately plotted on a map of Seattle. This involved quite a bit of walking around Seattle or perusing Google StreetView to find the exact locations where the photographs were taken.
Social Puzzles. Defense Grid had some particularly interesting puzzles that were easy to solve, but only a select subset of players were given the tools needed to solve them (based on a hash of their Steam ID). As a result, the players rallied together to try to find the individuals they needed, resulting in a tighter social community. In general, these types of puzzles were relatively easy to create, took some time to solve, and were popular with the players.
Elusive Puzzles. With an ARG, it can be tempting to have puzzles with vague objectives that are simple to solve in the hope that, with the sheer volume of players, someone will stumble on the solution. This type of puzzle was the least popular type. If a puzzle doesn't have a clear objective, players tend to find it too frustrating. One simple fix for this problem is to have a way to indicate partial completion so that even if a puzzle requires a lot of experimentation, players know when they're headed in the right direction.
Hackers Gonna Hack
We knew hacking was inevitable, but we didn't realize just how much of an issue it would be. Michael Austin at Hidden Path has a passion for software security, so he played the role of the security expert and suggested precautions that we should each take with our own code.
As for his own code, Michael explained, "I went crazy with security." Not only did he put ample security in place, but he also wrote code that would notify the server whenever it appeared that a player was attempting to circumvent the security measures. "Everything got triggered... every single security measure I put in got hit."
Anticipate. Every time a new content revision is released, we came to expect that at least one hacker would run a full binary diff off the new version to see what's changed. This can make it easy for them to track down new textures our audio files that may contain clues that are supposed to be revealed in the game. A binary diff will also quickly reveal unencrypted strings that are hard-coded into the game's binary.
For many of the puzzles, players needed to a URL to a website hosted on Valve's servers. To make sure that these URLs couldn't be trivially shared between players, Valve set up a system that allowed us to hash the player's Steam ID on the client and verify the hash on the server. Even with this in place, some of these algorithms were hacked by the end of the ARG.
Defensive Strategies. At the very least, programmers should obfuscate strings that could be useful to a hacker. We shared some code between development teams, but for the most part, we encouraged each team to use their own encryption methods to make it harder for hackers to apply their findings in one game to another game.
The Hidden Path team came up with some particularly creative anti-hacking measures. For instance, one strategy is to look for unusually long frame times -- something that would usually only be triggered by a hacker stepping through the code. If this was detected, they had a subroutine that would corrupt all of the important keys so that they were useless to the hacker.
One important aspect of all cheat-detection strategies is to stay silent if cheats are detected. Alerting a hacker may suggest to her that she's playing a legitimate metagame and may encourage her to try a different approach.
Misdirection. For the most part, we tried hard to never lead players on a wild goose chase. That being said, if someone is hacking, they deserve a red herring (or in one case, a Rickroll). Encrypt the important strings in the binary, but throw in a few extras that point to other, unrelated ARGs.
Vocal Minority. The good news is that cheaters, though problematic, are a small minority. One hacker published a way to earn potatoes just by clicking a few URLs that bypass the usual security. Valve tracked this on their end and found that only 1.3 percent of the total potatoes were earned this way. Valve later quietly revoked all of those rewards.
For some people, the journey is its own reward. But let's be honest, most of us at least want a souvenir.
The 36 potatoes that players could earn served as a popular status symbol for tracking player progress in the ARG. They were difficult to obtain through hacking, which kept their value high. To those who earned all 36 potatoes, Valve awarded their Complete Pack with every game that Valve has ever made.
In addition, the developers who were carefully monitoring the fan wiki and IRC channel hand-selected specific individuals who we thought had made a big contribution to the ARG. For those who are under 21, Valve sent a hefty pile of schwag. The others were flown to Valve for the Portal 2 launch party on April 18.
April Fools. The decision to launch the ARG on April Fools' Day was deliberate. We wanted to blend in with the typical April 1 chaos and let it sink in over time that #PotatoFoolsDay is something much larger than April Fools' Day.
Unfortunately, this tactic worked better than we anticipated. We had a lot of trouble getting early media coverage, which is unsurprising since media outlets are wisely wary of April Fools jokes. Nobody wants to release a big story only to retract it later. For days, skeptical players were insisting that the ARG was just an extended prank put together by a few indie studios.
Kidnapped Champions. The second big mistake we made was pulling the ARG champions away from their communities too early. Based on contributions in various games, on the fan wiki and in the IRC channels, we hand-selected a few individuals who had contributed disproportionately to the ARG and offered to fly them to Valve.
In order to give sufficient time to notify them and make travel arrangements, Valve contacted them a few days before the start of the crescendo. We didn't require them to sign any NDAs, but to maintain the fiction, we encouraged them to take a step back from the ARG and claim that they had been kidnapped by minions of GLaDOS.
We failed to realize just how influential these few individuals had been in keeping the ARG running like clockwork. The wiki stopped getting updated, leaving other players confused as to what was going on.
In retrospect, a few minor changes may have prevented this. For instance, we could have altered the fiction of the story in a way that allowed the champions to keep rallying the ARG community in the anticipation that the kidnapping would come later.
The Crescendo. By far the biggest mistake of the ARG was the decision to move up the timeline for the crescendo. We originally scheduled this to start 24 to 36 hours before the imminent 7 am Tuesday launch, but we worried that this would give ARG players in other time zones very little time to participate. So we decided to set the completion threshold much higher so that the crescendo could stretch over multiple days. We moved the start to Friday morning to make it easier for international media to pick up the story before the weekend.
So what really happened? When the final ARG puzzle was solved, it pointed to a webpage with a count-down timer. Players immediately assumed that this was the count-down for the early release of Portal 2 -- a suspicion that they believe was validated by a clue that was sent to several media outlets that included the hint "4/19/2011_7AM=4/15/2011_9AM," where the first time was the published Tuesday release date. In retrospect, it wasn't very hard to see why the the community was surprised and disappointed when the timer reached zero only to be replaced by a page with another timer.
If our goal was to get media attention, then we got it. The extended crescendo gave plenty of time for the news to cover the event, but we sacrificed a lot of goodwill with the players. Those who had been participating in the ARG since the beginning felt let down because there were no puzzles left to solve. They participated in the crescendo by playing the games but found it to be anticlimactic.
New arrivals often weren't aware that the ARG had already been running for more than two weeks. Rather than seeing the many hours of new content that had been hidden in our games, all they saw was a cheap media ploy to get players to buy the Potato Sack in the hope of an early Portal 2 release. Even worse, players were let down each night when the crescendo failed to complete, and felt cheated when their Potato Sack purchase didn't result in being able to play Portal 2 over the weekend.
If we had been able to do this all over again, there would have been several changes. First, we would have stuck to the original schedule of having the crescendo be no longer than 36 hours. Given the toss-up between media attention and player goodwill, we'll choose the player every time.
Next, we would have continued to add new in-game puzzle content during the crescendo so that the core ARG players could maintain their excitement and have a way to contribute to the launch in the way that they knew best. One other option that we could consider for future projects is to move a crescendo-style event to the start of the ARG to alert the media and engage our core fans from the very beginning.
Aside from the mistakes listed above, we did a lot of things right. From the very beginning, Valve gave the indie developers incredible flexibility to use the Portal 2 intellectual property without constraints. This gave us the freedom to focus on great content rather than having to worry about legal concerns or a lengthy approval process.
Everyone involved worked hard to stay responsive. We monitored the IRC channels so we knew, for instance, when players were heading to physical locations. In a couple cases, we spied on those players and then wove their usernames or stories back into the content.
This was a huge hit and made players feel like they were part of the story. In response to the media's initial hesitation to report on what appeared to be an elaborate April Fools prank, we encouraged blogs and news outlets to cover the story by sending them mysterious emails, official summaries, screenshots of new content, and trailers depicting Portal-themed content updates.
This responsiveness was also important for dealing with hackers. Chris Douglass and Al Farnsworth at Valve did an incredible job of detecting suspicious behavior and shutting down hacked content, usually within a matter of minutes.
Finally, one of the big successes was simply weaving explicit Portal 2 content into our indie games. When Phase 3 launched and screenshots and videos of GLaDOS started showing up in all of our games, fans and media went crazy. The ARG gave all of the participating developers an amazing opportunity to create new game mechanics and new content, most of which will stay around indefinitely.
With less than four months from inception to launch, the development schedule for the Portal 2 ARG was tight. But I think I speak for all of the indie teams when I say that it was one of the most enjoyable projects of my life. As Michael Austin at Hidden Path put it, "It was the most fun crunch ever."
Some of the success of the ARG may be due to the nature of the Portal IP. The core story is both dramatic and comedic, making it easy to work with. The gaps in the history gave us a lot of room to play around. In Portal, GLaDOS often plays the role of a disembodied commentator, making it easy to weave her character into another world's fiction.
Despite a few mistakes, everyone involved in creating the ARG content considered the event to be a huge success. Hopefully the lessons discussed here will help us do even better next time -- and we do hope there will be a next time.
There's been a lot of specu