Category: Twitch APGs

Pitcrew Postmortem


Download the build here. Link to audience interface:

Initial Goals

Pitcrew was intended to be an experimental exploration of the design space of Audience Participation Games (APGs) made in tandem with research into a framework for understanding the mechanics of existing, intentionally designed APGs. (This excludes games that are APG-ified through external tools or streamer intervention, such as naming your XCOM soldiers after viewers, though we did take note of those occurrences during research.) We wanted to learn the sequence of design choices that lead to the audience experiences, streamer experiences, timing mechanisms, and audience aggregation we see in the space. We wanted to make an APG that pushed the boundaries of how audiences interacted with the game through an interface; we wanted to expand upon the common voting mechanic as audience aggregation. We wanted to gain insight into the motivations of each level of the APG hierarchy, from viewers to participants to subscribers to streamers.


Pitcrew was designed and developed by Rachel Moeller (me) and Nathan McKenzie, lead designer on the APG project. Nathan created the Ludolab APG networking tools, developed the HTML interface for the audience, and advised the project, while I developed, created the assets, and designed it. I created five APG prototypes in the Fall of 2016, and the full Ludolab team chose the Pitcrew prototype for development. From January to May, Nathan and I met weekly for coworking (Nathan was also designing and developing his own APG, Gods of Spoons and Socks, as well as iterating the networking tools) and design discussion. The Ludolab designers and stakeholders met biweekly for extended design discussions. The research team (Jessica Hammer, Safinah Ali, Judeth Oden Choi, and I) met biweekly to discuss progress and emerging trends. The research team produced Analytic Frameworks for Audience Participation Games and Tools which went to the Spectating Play conference in Tampere, Finland on April 24-25, 2017. I wrote the ludographic portion of the paper. While not in coworking, I worked on Pitcrew alone or at CMU Playtest Night for about 7 hours each week.

Expanding on the first Pitcrew prototype, I first created the main racing experience. I playtested this basic racing functionality with racing game players and non-genre players at CMU Playtest Night. Nathan and I then implemented the networking, adjusting the design around technical hurdles. As the networking tools became more robust, we used Pitcrew to iterate them.

Pitcrew Prototype’s audience interface mockup
Pitcrew’s current audience interface

Acquired Knowledge and Skills

Pitcrew was my first racing game and my first networked game. Pitcrew’s intended tone was absurd and while that is mirrored in how the car handles, I learned that racing games are a cavalcade of fine tuning. I learned the basics of networking and IRC events, how to package and receive messages, and the effects of information travel time on player/streamer experience.

Quickly we learned that streamer and audience relationship is affected by APG mechanics and streamers, as the financial gatekeepers of APGs, will only repeatedly play what they perceive as fun (regardless of the audience.) This led to safeguards for the streamer so that the audience could not remove their agency through their choice of car build. This included revealing the car builds of the pitstops to the streamer so they can choose which build to receive (though obfuscating the streamer’s information about the current state of their car by not giving them concrete numbers but a dial on a spectrum from low to high), allowing the streamer enough gas/time to choose between pitstops (though this is limited to force the streamer to interact with the pitstops on some level), and extra handling controls like the emergency brake. I was also careful to make all the choices the audience can make situational, meaning that there are never choices that affect the streamer in purely negative (ex: automatically killing off the streamer) or positive ways (ex: restoring the streamer’s health.) All of the audience effects on the streamer depend on context. In looking at other APGs we found that streamers quickly get tired of consistent negative behaviors against them, and audiences generally feel very strongly about their streamers (though how this adoration is expressed varies wildly depending on the stream’s culture.) Context driven effects lead to more complex opportunities for stream generated meta narrative and I felt it provided more options for forming and expressing stream culture.

Audiences are often aggregated into a mass of input in APGs. For example, voting turns an audience of 15 into a god force or a player 2 force. We learned that expressability and individuality diversified the experiences audience members had when participating, even if those mechanics did not directly affect gameplay. This led to the visual representation of chatting viewers and their individual usernames as fans in the stands of the in-game track. This was also an attempt to include viewers who have no desire to participate in the game, as pure spectatorship is a strong motivation for users of livestreaming platforms.


Setbacks and Solutions

Scaling became a big question for the game, including the question of “is this game playable without an audience? /is this game fun without an audience?” I opted to answer yes by including some basic AI that changes the pitstop builds every 30 seconds if there are no audience members playing. We also needed to address how audience members were allocated across the pitstops and capping the number of total audience participants. The cap was necessary to appropriately route and make agent the choices of each participant. Few existing APGs do this, an aggregation technique we called “Dedicated Participation.” Originally, I designed each pitstop to scale from 5 to 15 players. Car builds would have five areas to change: interior, paneling, peripherals, engine, and tires and up to 3 audience participants could change car parts in those areas. For example, the engine block contained pistons, sparkplugs, and a casing and an audience participant would be assigned to each part, with three choices per part. Each choice affected two areas of the car’s score attributes: Weight (rotation angle), Speed (acceleration rate), or Suave (how far the camera was zoomed out over the car). The combination of all of these part values would result in the total car build. If there were less than 15 players, only one audience participant would get assigned to each block and would only change one part (one audience participant affecting the engine’s pistons). If there were a number between 5 and 15, audience participants would get distributed evenly over the blocks. If there were less than 5, players would get spread over different pitstops with their single choice affecting the build at that location. This choice ended up being much more compelling, especially since most streamers have very small audiences. In the current iteration of the build, the number of audience participants is capped at 3, with each audience participant getting control over a set of parts in the engine. There are three pitstops in the game, and all carry the same build constructed from the audience input. If there are 1 or 2 audience members, their choices are the only changes to the build (we do not add a dummy player.) We made this decision due to scope and hope that future iterations of the game can better address the issue of scaling.

Time and delay were also significant factors in the design of the game. Messages between client and server are sent within seconds, but the effects of those messages won’t show up in the stream for at least 15 seconds. To mitigate the feeling of waiting, I chose to display the full car build in the participant’s interface. They see the specific effects of their part choice, the grand car build, the names of their teammates working in the same block, and their teammate’s choices (in the form of changing sprites; each part has 3 options, each with a different corresponding image) changing in real time. Participant changes are collected and sent down at regular, but delayed intervals to prevent spamming. I chose not to display this collection interval to keep this visual activity going and to provide the illusion of immediate change/agency. The streamer sees this change in-game within a second, so their reactions can occur in somewhat real time as well. So even though the stream that audience participants watch will lag to show their changes, the participants are still getting immediate feedback from their choices.

Getting the networking working was a challenge and took the longest time to implement. Making sure to set aside lots of time for networking debugging is a must.



Pitcrew has been an exploration of how IRC, streaming, and networking tech constraints effect streamer/audience experiences; how APG mechanics affect stream cultures; the motivations of livestream platform users, the possibilities and current boundaries of the APG design space; and APG interfaces. Ludolab hopes to continue work on Pitcrew in the future, iterating based on user experience and new opportunities in the design space of APGs.

Check out all my posts on the APG project with CMU’s Ludolab (listed in chronological order):



Trends in Audience Participation Games

Of the 22 APGs we studied:

6 were arena matches (many with platforming elements):

  • Wastelanders
  • #IDARB
  • Move or Die
  • Streamline
  • Ultimate Chicken Horse

4 were rouge-likes:

  • Choice Chamber
  • Legends of Dungeon Masters
  • Party Hard
  • Upsilon Circuit

With the next widest common trait being zombie games (3):

  • Dead Nation
  • How to Survive 2

and Party games (2):

  • Quiplash
  • Superfight

The singleton outliers in genre were: Horror (Daylight), Fighters (Salty Bet), Platformers (Cluster Truck), Creative (Poly Bridge), Adventure (Rise of the Tomb Raider), Survival (Don’t Starve Together), Tower Defense (One Troll Army), JRPG (Akiba’s Trip), and Other (GeoGuessr).

Poly Bridge

Nearly all APGs studied allowed the audience members to influence gameplay to increase difficulty for the streamer in some way, an effect I referred to as “influencing obstacles.” The exceptions were: Salty Bet, which does not allow audience members to interfere with the game; Superfight, in which the audience only votes on options provided by the streamer; Quiplash, another party game in which the audience only votes on options provided by the streamer; and GeoGuessr, in which all audience and streamer interactions occur one-on-one through challenge setting. From this it seems like party games tend to use the audience as a determining force, rather than a modulating force like other genres do. Only the party games allowed the audience to determine the winner of the game, giving the audience a higher power position in the community dynamic of the stream.


Within arena match style games there were several mechanical trends such as interval voting, individual commands, obstacle influences, and resource influences (which were less common than obstacles influences.) These function as gameplay modulating tools for the audience, allowing them to control the difficulty and pace of the game play they watch. This is important to consider when thinking of livestreaming platforms as a view-first platform: the audience comes to Twitch to view, so tools that allow them to change the viewing experience (pace, the performative quality of the streamer, etc.) might be more widely accepted than tools that affect gameplay for the sake of talent, etc.


Of the Audience Aggregation methods, interval voting was the most common. This isn’t surprising, as it has been stated that streamer prefer games with a player-driven or predictable pace that gives them time to read the chat and interact with their audience. Interval voting also offers a clear solution to timing issues and networking traffic when designing the APG: if the game is naturally divided into turns or rounds, data gathering periods can be easily attached to the parts of the game dictating the existing intervals.

Betting was he least common Audience Aggregation method, with only Salty Bet (a social/betting overlay for existing fighter matches) and Streamline allowing players to bet on the outcome of matches. Interestingly, there were many APGs with persistent currency for other audience interactions: (Upsilon Circuit and Legends of Dungeon Masters let audience members spend items on streamers, Party Hard let audience members spend experience points on NPC actions and expressibility, Superfight had experience points that function as a status symbol.)

Chat participation was by far the way audiences input data int the APG. Only 5 games did not use chat: Salty Bet, Poly Bridge, GeoGuessr, Upsilon Circuit, and Legends of Dungeon Masters all had their own APG interfaces. Poly Bridge, GeoGuessr, and Quiplash (which used chat to gather votes) used their APG interface for content generation as well.


Party Games preferred option voting to interval voting, as well as Rise of the Tomb Raider, which gave audience members card based options to vote for. Interestingly, this case, many of the options were not overtly negative or positive (though there were some), but cosmetic such as “big head mode.” These options, an effect I called “event summoning” were prevalent in many of the singleton genres. Horror game Daylight allowed audience members to cause sounds in the environment that would not affect gameplay, but would affect the streamer. These kinds of events are another example of view-first events that are triggered to change the viewing experience rather than the game (watching the streamer freak out, for example.)

Hidden Information Power Mediators in APGs

Some of the predecessors of  contemporary live-streamed APGs were games the research team calls “couch APGs.” These are multiplayer audience games played locally, many of which of relied on asymmetric or hidden information to construct dynamic relationships between players and set up mechanically relevant situations. Examples of this span social interactions (Legend of Zelda: Four Swords Adventure using anonymous voting through individual game boy controllers to vote on who the worst player of the round was, with that player being publicly shamed) or mechanical, rewarding actions (Pac Man Versus giving control of the ghosts to player 2 and above) or to ease the load of UX (Final Fantasy Crystal Chronicles using gameboys as an extra menu screen for character adjustments that don’t interrupt the main gameplay loop.)

Pac Man Versus

Due to their visual capacity and the sharing of the the game interface between audience and streamer, the number of contemporary APGs that accomplish this kind of information balancing are rare. In Choice Chamber, for example, the streamer knows what voting options the audience has during this interval. This creates an interesting, if repeated, tension of the streamer asking for an option and waiting to see how the audience responds.

The APGs that do pull off hidden info are usually games in which the audience input is content generation or command based, but in each of these cases the hidden information is instantaneous and controlled by an individual audience member. The streamer is always hidden from, and their information (the stream and what’s happening inside it) are always public. Poly Bridge allows audience members to propose changes to the track or a new track entirely, and that information (of success or failure should the streamer chose to play that track) is mostly hidden to all but the bridge creator, an audience member. There are cases where the audience is in control of the options to vote on by voting with commands that are always available (Ultimate Chicken Horse). To maintain the power of the streamer, mediation between the effects of the audience and the streamer are almost always present. In Ultimate Chicken Horse, the players decide which of the obstacles voted on will end up in the level. In Poly Bridge, the streamer decides when and if to play an audience bridge plan. In contrast, APGs with non hidden information use the presentation of information as this power mediator. The begging of the streamer for option 1 of 3 allows them time to prepare for an option that negatively effects their playstyle, gives them a social opportunity with the audience, and allows them to gauge the attitudes of the audience. This multi-purpose power mediator is important in keeping the streamer playing (the streamer is the audience’s gatekeeper) and maintaining enough audience agency to keep both sets playing.


Pitcrew Networking Demo

Pitcrew is now connected to Twitch! Audience members can join and make choices (though they don’t quite map to the architecture for mapping the choices to pitstop values yet.)

Things we (Nathan and I) learned from this:

  1. Tech: Using Nathan’s networking scheme in which the game uses a chat and logic channel to send client messages, the streamer cannot use the client while logged into their logic channel. The chat channel is fine (the app will load and choices will register.)
  2. Lobbies: Pitcrew has a lobby that will wait for the streamer to start the game when enough audience members have joined. This kind of structure needs a cap on how many can join. Feedback in the stream and the app are integral for communicating when choices can be made.
  3. Tech: Tie all of your messages to explicit events or put them on a timer. If you accidentally spam messages to Twitch (say you just set messages to launch in update), Twitch will ban you.
  4. Tech: You must register your html app with Twitch using a permanent(?) URL for multiple people to play. Right now Pitcrew does not have this, so only one person can play.
  5. Scalability: It made sense to space pitstops equally over the track, and that if one person was playing, to have their build occur at every pitstop. Having one pitstop would have to change the rate of re-fuel/power consumption, which seemed too large a constant to change up on the streamer. Instead I decided to always have three equally spaced pitstops with influence shared. In the edge case of an odd/edge case number of players, padding the other stops with average value AI car builds might also provide the streamer with an opportunity to regain better control of the car from troll pitstops.

Expression in Audience Participation Games

Opportunities for individual audience member expression are a less common feature found in APGs. Interestingly, there are many external tools made for this purpose, and the prevalence of these tools indicates a desire from both streamers and their audiences to hear individualized voices in the crowd. This design concern is challenging considering that all but one of the APGs researched at the time of this writing do not represent individual audience members in game. Chat is the main interface for individual communication on livestreaming services and is where (in addition to profiles) individuals are represented. The desire for expression may arise from the prevalence of aggregation mechanics used to represent audience input in many APGs. Voting consolidates audience member voices into an identity-less input. Even in APGs that use voting, expressive options that may or may not link to game mechanics can give the audience a sense of individuality and empowerment when making participation decisions.

Expressive Features in Existing APGs

Party Hard’s Badges: Audience members can enter a chat command which, for the duration of the stream, will create one of 12 Party Hard specific badges next to that audience member’s username. The badge does not change anything about their participation in the game; it’s purely for expression. Badges like this can denote team or streamer allegiance (i.e. the Horde and Alliance badges that popped up on Twitch a while ago.) Since this badge’s only functions are social and do not have an effect on the game, this kind of participation raises an interesting question: If the audience member is using a badge specific to this game and the badge is not persistent over time or in other streams, is this player playing the APG if they’re only chatting and using the badge? If the answer is yes, then design decisions around expressiveness must be addressed. How can the APG create and manipulate the streaming experience for these low level participants? How can this feature push the social play of the APG? Perhaps a streamer can unlock more badges for audience members to use by finding hidden secrets in the APG. This sets up asymmetrical goals between the streamer and the audience: the audience may encourage the streamer to search for these audience benefits that accomplish the audience goal, but are only tangentially or not related to the streamer’s goal. This tension sets up opportunity for stream bonding, the formation of stream culture, and the progression of a stream community.

Expressive Features in APG-ified Games

Screenshot (325)
Naming an XCOM 2 soldier for a Twitch viewer

Individual Recognition: Streamers frequently name their pokemon or XCOM soldiers for audience members. Streamers pick the individual for any number of reasons: recent sub, recent donation, funny or supportive chat comment, mod, community engagement, etc. By using the name of the audience member in play, that audience member is made visibile to the others and performs alongside the streamer (even if this performance is constructed by audience members.) For example, it might be frustrating if a soldier in XCOM dies, but if xxxSickMemes2010xxx is about to die, the audience members may feel more attached to that representation than they normally would. While this event is not purely expressive-based, it is an example of how individuality changes the experience of the stream. Allowing this kind of opportunity for individualism in APG design is another avenue to

Expressive Features on Livestreaming Platforms

Badges: Persistent icons that appear next to a username in chat. Badges indicate status, loyalty, and authority within a stream. Many are earned, rather than chosen. This distinction is important, as expressive features must inherently have some option in order to show the holder’s choice.A feature that shows status is more sought after, but a feature that is chosen from a group may become a symbol or icon for a community within the stream or represent an event that occurred within the stream that the community wishes to remember. When designing expressive badges, their use as symbols should be understood.

Soundboard: Beam has a soundboard, a panel on every stream which audience members can spend currency at to play a sound of their choosing over the stream. This noise is attributed to them, but is momentary. Interestingly, this expression feature is a conjunction of status and expressiveness, as currency is earn by watch minutes. The accumulation of currency (combined with a cool down period for when sounds can occur in succession) limits the amount and ability for audience members to express themselves through this feature. This makes the soundboard a a brief status reminder to other audience members.

Screenshot (324)
Beam’s soundboard

Designing APGs with opportunities for individual expression are an avenue for providing audience members with power, either through status or freedom.

Audience Participation Game Longevity

Audience Participation Games (APGs) have become something of a novelty in livestreaming. The top games on Twitch are League of Legends and Counter Strike: Global Offensive, interestingly these two e-sports typically have remain in the top watched spots for long periods of time. Usually a newly released game is also in the top three, but these games usually leave popularity relatively quickly (few days to a week after/prior to launch depending on streamer advertising deals) for the next new release. For example, Conan Exiles, a relatively unheard of game, rocketed to the top spot on release for the fact that it was new and, I suspect, for its novelty nudity. As of writing this, it’s no longer in Twitch’s top 25.

Screenshot (320)

APGs tend to fall into this novelty category. Choice Chamber, arguably the most famous and widespread APG, is still played by streamers on Tuesdays at 11:30 AM (the time of this writing), but the game is not being watched by enough people to get it into Twitch’s top 300. Considering Choice Chamber’s audience facing mechanics, this seems strange. Incorporating e-sport elements and augmenting them with APG specific content could help APGs stay in the spotlight longer and increase market longevity as well as fill out the open critical space APGs have hinted at since their inception.

 Design Considerations

  1. Allegiance and Spectatorship: E-sports, like all sports, are sports because they are fun to watch. People flock to them, know their rules (usually in an inherent way that can be hard for newcomers to understand simply because they aren’t explained publicly any more), and viewers form attachments to individuals or teams. Individuals may be team members and/or stream entertainers who cultivate a stream community.
    Screenshot (321)
    League of Legends even has a massive following (for a streaming individual) at 4 in the afternoon on a Tuesday.

    The bonds audiences form with their teams and entertainers fuel the persistence of e-sports on livestreaming platforms. A viewer can always check in with their favorite team and chat with/be a part of that team’s community. APGs have deep intentional for community building due to the way their mechanics form or degrade social bonds. Teams of audience participants could become just as popular as teams of streamers.

    Screenshot (322)
    CS:GO rankings
  2. Persistent Identities: This can only occur because audience members are repeatedly exposed to playing entities (teams/individuals) in a positive way. The channels for exposure may vary. For example, televised Starcraft tournaments get team names and faces out to fans on TV, while a streamed tournament relies on Twitch’s front page status or pre-existing bonds with a viewerbase. These entities must exist in multiple channels over time to become persistent, or to become “findable” and “checkable” to a viewer. This allows a viewer to associate themselves with that playing entity over time, creating two persistent identities (the entity’s and the viewer’s.) Profile mechanisms and Twitch badges, as well as platform currency like bits, provide an opportunity for audience members to express their allegiances and become stakeholders in the games they watch. This keeps them watching/participating longer.Screenshot (323)
  3. Combinatorial Events: Surprises in sports arise from performance. They are not built into the game’s system. This is also a factor of replayability. Party games like Superfight and Cards Against Humanity (again, popular streaming titles, but not in the top 25 at the time of this writing) in which novelty is built in, contain their surprises in their content, which gets stale after repeat plays with the same players. Sports gave clear goals with an uncountable number of events that can occur between start state and solution. APGs have an advantage here: the audience represent as many events/opportunities for surprise as there are viewers. This is embodied in chat. Designing ways for audience input to stack or build off each other can create this uncountable number quickly.
  4. Metagame: Stream communities are a part of metagaming. They create their own fanfiction, art, question boards, and spin off games to celebrate the community and the game it’s centered on. Sports communities do this too: they sell t-shirts, autographs, pre-game passes, etc. to feed their communities and reward them for being involved in the community. APGs have a clear advantage in this arena: nominated and acknowledgement. Nominated participation is appointing an audience member to be a participant. A form of this occurs with sub gizmos in Choice Chamber, in which subscribers get extra control of a gizmo if they subscribe during a stream. Acknowledgement is publicly identifying an individual(s) for their contribution, such as the way Ultimate Chicken Horse reveals the usernames of the audience participants who voted for a chosen obstacle. These extra rewards encourage the formation of a metagame community which will keep an interest in the game for the sake of the community (but this still keeps the game alive longer!)


APG Participation Hierarchy and Role Acknowledgement

In examining existing APGs and commonly streamed non-APGs on Twitch, I’ve found that there are levels of participation. Whether or not these levels should be described as a hierarchy is debatable and I do not attempt to ascribe worth to the levels, just to note that the motivations of stakeholders in the stream are different at each level. In designing Pitcrew, I want to reward participants at as many distinction levels as I can, as this provides layers of secondary feedback (see previous post), increases audience engagement with the stream, and can be used to strengthen community bonds/culture.

Audience Hierarchy Roles (in order of authority level)

Streamer: Here I define the streamer/players/host/broadcaster as the individual(s) playing the game and streaming content. The streamer is actively playing the game. In Pitcrew, the streamer drives the car around the track with the goal of beating their best time.

The streamer controls the purple car. *Most of the following figures are in white to allow for computational color grading.

Moderator: A moderator is an appointed (by the streamer) community manager whose task is to maintain the culture of the stream community by enforcing rules, gathering viewer data, participating/organizing community events, etc. Moderator status is often considered an honor, and moderators stand out in chat with special badges and bold text. This status is indicative of some relationship/trust with the streamer. In Pitcrew, I added a commentator box that appears next to the track when a mod participates in the chat.*(Chat participation is a social contract indicating that the chatter is ok with having their identity in the stream. See previous post for details on this social contract.). Since moderators function as a voice between the streamer and the community, it made sense to me to theme their representations as a race commentator.

Screenshot (268)
This box would normally have a figure inside.

Subscriber: A subscriber is an audience member who pays a fee to access exclusive content in association with the stream (ex: private chat rooms, etc.) Paid subscriptions constitute a large portion of streamer revenue and are an indicator to the streamer of their entertainer status. Overlay tools exist for acknowledging a subscriber when they subscriber during a stream. Some games, including Choice Chamber, also acknowledge subs with special in-game events.

Screenshot (265)
Note the cat overlay thanking the subscriber. This shows off their name in front of the whole stream for a few seconds, making it a status boost for the subscriber. Image from archived stream of Chris from Phonecatss.
Screenshot (266)a
The subscriber has control over a gizmo in-game. The gizmo remains on screen until the subscriber cashes it out for resources given to the streamer. The gizmo can also aid the streamer in navigation puzzles. Image from archived stream of  Chris from Phonecatss.

In Pitcrew, I’m mapping during-stream subscribers to cheer leaders that appear around the inside rim of the track, very close. This proximity to the track makes sure they are always seen, regardless of the streamer’s suave level (which changes the camera height, allowing the streamer to see more of the stands). These cheerleaders are in stand-viewer figures that are randomly given a hat/costumes to wear.


Donor: A donor is an audience member who donates money to the streamer to support them. They may already be a subscriber or not, but a donation can be any sum of money and does not give the donor and subscriber benefits. Donations can get quite large and seeing a streamer’s reaction to a large donation is often half the fun. I want to add more feedback for donors, so I’ve given them advertising space, literally. Donor usernames appear on the side of the track as an ad sign. They say things like “BUY usernameX”, “Elect usernameX”, and “marry me, usernameX?”

Screenshot (267).png
Donor sign feedback on the left, general chat participant on the right.

Chat Participant: Chat participants will populate the stands with figures with and usernames. These basic role representations will populate from the outer edge of the stands inward, requiring a higher suave level to see them. I’m hoping this will create tension between the differing goals of the audience (get the streamer to be more suave) and the streamer (finish the track faster.)

Nominated Individuals: Something that I have not yet seen in APGs is the nominated audience member. This is an individual singled out by the streamer for whatever reason they chose. I think this presents an opportunity for the Streamer to decide who is acknowledged for reasons game designers cannot explicitly design or plan for (considering the longevity of the APG.) In Pitcrew, I have given this role to the flag waiver, a figure at the lap mark who waives the starting flag.


Other Roles

Dedicated Audience Players: A dedicated audience player joins a play state that is limited in scale or otherwise delineates chat and viewership from a more involved play interaction. In Pitcrew, dedicated audience members click a link in chat to join a pit stop. The link sends them to an alternative interface from which they can modify their car plans. This alternative interface signals to the audience player that the game is registering their input. For this reason I have chosen not to include another representation of them in game.

Viewer: A viewer who does not participate in chat may do so for numerous reasons. However, if the viewer does not participate in chat, they do not want their presence public known and identified at that point. I have talked about this in a previous post, but I believe it’s a violation of social norms on Twitch (and downright creepy), to acknowledge the non-participating viewer in game.

Roles Not Applicable to Pitcrew

Teams and self differentiated audience participants (Party Hard allows audience members to assign themselves one of twelve badges in chat) are not designed into pitcrew in representable ways. (In the case of Pitcrew, I lump teams with dedicated audience players) However, these roles present another space for audience acknowledgement in the stream community.

Considering the complexities of stream cultures and status within those communities, I think it is important to differentiate feedback for each kind of stream participant. This allows each action to be discovered by the community (a bonding exercise) and allows status and desire for status in relation to the streamer to grow, aiding community management and revenue opportunities for the streamer.

Audience Choices and Streamer-Audience Relationships in APGs

When designing livestreamed APGs for community centered platforms like Twitch, audience choices are commonly distilled into a few options to vote on. For example, in Choice Chamber, audience members are given a list of options to determine weapons, enemy difficulty, resource drops, etc. A trend I have noticed in popular APGs is that audience options tend not be a mix of objectively negative towards the streamer, objectively positive towards the streamer, or entertaining. There are a few instances of subjective options to affect the game too, which manipulate streamer-audience relationships in more subtle ways. Even when that is the case, streamers tend to percieve the audience’s choices as binary good or bad. See PewDiePie’s introduction to Choice Chamber:

Objective Negative Choices

These are actions that are taken against the streamer with the intent of making the game more difficult or unplayable for them. #KILLALLZOMBIES’ deplete streamer health command, Choice Chamber’s husky enemies, etc. In voting based games that do not provide on screen input, these negative effects lack accountability to the streamer. The streamer doesn’t know exactly who voted for this option, but they do feel a consensus that the group is trolling them or at least being openly antagonistic. In some cases this can be a streamer asking for higher difficulty, but in both cases this changes the streamer audience relationship. In games that provide only binary good or evil choices for audience members, this can be interpreted as a deliberate trolling message to the streamer. In designing APGs that consider the streamer-audience relationship, it’s important to note that repeated antagonistic behavior may degrade the relationship or change the culture of that streamer’s community. Streamers with a family friendly, moderator heavy channel with a long blacklist of chat terms may find that negative mechanics encourage antagonistic mirror behavior in chat. Streamer popularity is also important. These games may be fun for an established streamer with a dedicated viewer base (such as Move or Die, which has exclusively negative effects), but an emerging streamer trying to create and retain a community may find the experience hinders them. For audience members, seeing the reaction to their negative effect can have the same emotional payoff as a successful troll or watching a master overcome their obstacle may provide recognition and awe at the talent of the streamer.

Objective Positive Choices

Screenshot (246).png
An audience member heals the player in #KILLALLZOMBIES. Image from KDStudios.

In our recent paper, the CMU APG team found that audience members in general prefer helping the streamer to hindering them. They also prefer acknowledgment of the streamer rather the acknowledgment by other audience members. An objectively positive choice here is an audience choice that helps the streamer in some quantifiable way. This may be choices like the unlimited ammo command in #KILLALLZOMBIES, or double jump in Choice Chamber. These choices are a uniting force for the streamer and their audience. The audience feels supportive and part of the community including the streamer. This is in contrast to a negative choice, which divides the community into streamer and audience.

Entertaining Choices

Lara Croft with a big head, as voted on by Twitch viewers. Image from Dan Allen.

Entertaining choices no not have explicit gameplay effects, but are important in providing user feedback and generating community bonds. I think of them as inside jokes (which can get old.) These are commands like Rise of the Tomb Raider’s big head mode and rainbow trail cards. While it’s easy to dismiss these interactions as cheap or a novelty, the evolution of the entertainment role is important to consider. APGs with entertainment roles designed into them are a form of engagement driven by the audience. It makes them feel included as culture and content creators within the community. Entertainment choices create metagame moments that persist between streaming sessions and bond the community.

Subjective Choices

Compound and contextual choice in Choice Chamber. Image from BrownMan.

These choices are the most complex in the current space. These are choices which are often perceived as good or bad but in reality their effects are dependent on other factors. One example is Choice Chamber’s pogo jump, which is reviled by many streamers. When paired with high flying enemies, this ability can help the player better take out enemies (as there is no angular aiming from the ground). These choices are replayable, in the sense that their outcomes are expected to change give the situation of game events. This makes them more compelling than entertainment choices, which have one expected outcome for the most part (if you vote !bighead, then you already know what you’re getting into with the outcome of your vote). The effects on streamer-audience relationships from subjective choices are time driven. This means they are more sensitive to issues of delay, input intervals, and audience member communication with each other. For example, an audience member who wants to help the streamer would need to know about the upcoming room in Choice Chamber before they picked an option for jump height. I’ve tried to play around with this information dispersal with the obstacle minimap in Pitcrew, allowing audience members to plan their car builds with time to talk to team mates and the viewing group. In general, designing for chat communication, pushing audience members to communicate, is a deep social structure which bonds communities on large and small scales. Friends and alliances are made when the chat is given a clear set of goals (see Twitch Plays Pokemon and democracy vs anarchy). This manipulation of the audience is complex and builds stream communities differently than explicit anti-streamer and pro-streamer behaviors.

Implications of APG Interfaces

Most of the APGs currently on the market have chat-play (typing and sending messages through Twitch’s chat) or external interfaces which allow the audience members to influence the game. These input methods have different player responses and encourage different kinds of relationships and behaviors in play.

How do viewers who desire to influence an APG become players? In chat-play, the viewer types a command or vote and sends it to the game (and the public).

Note the chat messages with associated user identity in the right column. (Image Credit: Austin Chronicle)

This action comes with a social contract. By participating in chat, the viewer knows that their name and message are now publicly viewable and knowable. This can be interpreted as opting-in to play because the player is aware that their contribution and identity are both linked and public. In Pitcrew, the racing game I’m making for the project, we tossed around the idea of having viewers’ usernames scraped and assigned to figures in the stands. Over time I cam to realize that without this opting-in phase, the feedback of your viewer being a viewer in the game might feel creepy (as a breach of privacy.) To solve this, we decided to use chat participants instead of all viewers, because chat participants already feel comfortable in the community (enough to have a voice, at least.) This delineates two other roles/places in the audience hierarchy: lurker viewers and chatting viewers. Designing for these motivations is a new opportunity and could potentially raise questions of privacy in Twitch games.

Salty Bet’s interface is external to Twitch, but uses Twitch’s chat. (Image Credit: CD Gamer)

External interfaces do include explicit permission to play. The viewer must click a link or go to a unique webpage to become a playing audience member. This action is agreeing to an exchange of information, making feedback mechanisms engaging rather that creepily unexpected. External interfaces provide another challenge, however: they add barriers to entry. Learning an interface takes time, which is an investment for players. Part of Twitch’s appeal is that it’s easy to access games that normally take long downloads and expensive purchase amounts to experience. There’s no installation or complicated settings to tick off a list. If a viewer wants to be a part of a community or experience a game, they need only click a few buttons. This ease of entry begins to get muddy with more and more layers. In the APG project’s current html external interface communication tool (created by our amazing lead designer Nathan Mckenzie), we do this by having a link to the external interface appear in chat every few seconds. This link takes the player to a different webpage owned by the game. This also requires another channel for logic processing, requiring the streamer to have to pass the game through a second channel. This process, while not hard, is asking more than a traditional installation. We don’t know what effects this will have on streamers playing the game and sticking with it, but we are aware that it is a hassle. The link that appears in the chat is also several lines long, which clutters up the chat and takes up about half the size of the reduced sized chat in the external interface. Reducing or removing this link (where appropriate) is important in keeping the chat a legible hub of communication, both for the streamer and audience members who need to communicate about their team’s involvement in Pitcrew. Having access to the link in the external interface also encourages one player to quickly open a bunch of new tabs and fill the game with just one human audience player.

There are very few existing input methods for audience-players on Twitch. Technical constraints should be kept in mind when making new input methods, as should the way input manipulates a player. Does their input method reinforce their identity, such as in chatplay? Does their input method reinforce/challenge their status in the community, such as highlighting subscriber chat messages in play? What relationship and stakeholder does the input method effect, such as lurkers vs chat paritipcants vs opted-in players?

Audience Participation Games: Delay, Audience Feedback, Necessity, and Relationships

Audience Participation Games (APGs) are games played through livestreaming services like and Beam during which the viewing audience can directly influence the events of the game. An example is Twitch Plays Pokemon, an automated experiment in which Twitch users typed commands in chat to control the player avatar on screen. Another example involving a streamer, a content facilitator/host/entertainment personality streaming gameplay video on the service, is Choice Chamber. Audience members use Twitch’s chat interface to vote on obstacles, resources, and difficulty of levels for the streamer playing the base 2D platforming game. Design concerns games for APG play include delay, feedback, audience presence, and community relationships.

Audience Feedback

The minimum delay time for an audience member to see their effect in the stream on Twitch is 15 seconds (and this is assuming the streamer and this audience member have a high quality, high speed internet connection.) This time amount varies across internet speeds too. However, the delay between the audience member input and the streamer seeing that effect (on their side, not in the stream) is relatively quick, maybe a second or two. Without any feedback from the game or the interface, players question whether their input even registered and/or have no idea if their input effected the streamer’s actions. Their identity within the game is buried, and their role becomes unclear. Fixing this requires giving individualized audience member feedback at every possible opportunity. Ultimate Chicken Horse does this by having the names of audience members who voted on chosen obstacles explode out of the object when it is used. However, there is still a 15 second delay (at minimum) before the voter’s input is even recognized, and this input may never be recognized if the streaming players don’t pick that obstacle.

Screenshot (228).png
Image from Clever Endeavor

For games with external Twitch interfaces (like our HTML5 client games), feedback should be given within a personalized environment that at least confirms audience input and at most is smart enough to acknowledge the kind of input in some way. Feedback should also occur in the stream to indicate participation to the streamer so that they can recognize active followers and purely spectating viewers can also see audience player contributions. This double feedback ensures the player is never left hanging and doesn’t attempt to spam input (which can get you banned from Twitch) simply because they don’t know if their input registered.

For games that use chat play, this registration is already handled by the chat message appearing. In my research on streamer desires, I’ve found that streamers prefer interactions that do not clutter their chats, as they use chat to communicate with their audience.

This real-time (relatively) feedback for audiences using the external interface presents a question for games that have input intervals, like Ultimate Chicken Horse or Choice Chamber. Choice Chamber does not show the explicit time interval for votes, which may be their way of solving this issue of accurate timing. Timers showing the time to acceptance of audience input cannot be trusted. The streamer will see an accurate countdown (if based on time or framerate), but the countdown is obviously uneven when viewing through the 15 second delay window. One solution may be to have false timers that tick solely on the client side. The game begins and the audience members see a “30 second” countdown in their interfaces. This countdown is running on their connection, so they perceive it as even (evenness is more important than accurate seconds to the perceiving player.) This countdown is counting until input will be collected from the client. Meanwhile the server is counting down accurate (to the streamer/game) 30 second intervals and pings the client at the end of this interval. Input is gathered regardless of the client side timer. This fuzziness is acceptable because the audience receives feedback at the end of their timer stating their choice was heard (double feedback) and eventually their name will appear in the stream, but the time between those events is not scrutinized because the audience member knows it will happen eventually. Meanwhile the client timer provides the illusion of urgency to the player. It achieves its emotional purpose.

The Essential Audience


Screenshot (230).png
One Troll Army’s Steam page. Notice the ability to opt in to having Twitch viewer play.

Another major design choice I’ve come across is whether this game is playable without an audience or fun without an audience. Both options are valid: a game that can only be played with an audience or a game that is fun with and without one. This question is important to answer quickly in development because it allocates resources. Does the team spend time making an AI to handle what the audience input would normally do, or can we always assume there will be an audience playing? Designing levels that are fun for both a single player and a mass audience is hard, and choosing to assume the audience’s presence cuts out balancing for a single player. This also changes the culture around online streaming services as a separate medium for play, and has the potential to democratize APGs. This question is also important for dealing with tech. Is this an installed executable or a fully browser based experience? Part of the draw for current APGs on Twitch is that the games are incredibly lightweight. Entering play is simple and fast, with no major barriers like time to install or a fee to play.

Screenshot (231).png
Streamline’s Steam reviews showing the APG is arguably less fun with no audience.

Designing for Relationships

Streamed games on Twitch that stay popular are designed with elements, intentional or not, that cater to at least 2 relationships on Twitch (and possibly 3 or 4.) Each relationship has differing player motivations. These relationships are:

  • Streamer-Game: The streamer (player) has exceptional talent or lack of talent at this game and their performance is a thing to behold. Games like League of Legend and DOTA 2 (esports) cater this kind of relationship both in their core mechanics and communities. This also encompasses the viewer motivation of being a fan of a certain game, but maybe not the streamer/player themself.
  • Streamer-Audience: The streamer (player) is an entertainer, host, celebrity, authority, or otherwise community facing entity that interacts with individual viewers in someway. This relationship is about community building and management. Games like XCOM 2 and Pokemon are frequently played by these streamers because they provide opportunities for the streamer to give back to the audience with non-integrated feedback like the streamer naming their soldiers for favored audience members, etc. This kind of relationship also generates false Twitch plays, or play throughs of games using third party bots in chat to gather votes that the streamer uses to play. This is also the home of commentators and hosts who comment on automated experiences like Salty Bets.
  • Audience-Audience: This is the relationship of of audience members with other audience members. Merit badges, persistent items/currency, subscriber status, and mod status are all facets of these relationships. Games with persistent interfaces independent of individual streams (like Salty Bets) cater to this.
Screenshot (229).png
Twitch streamer Clarelu‘s persistent viewer of the month tally
  • Audience-Game: This encompasses viewer motivations such as novelty: “Should I buy this game?”, which explains why recently shipped games trend on Twitch for a few days after their launch, like Conan Exiles. This also encompasses viewer motivations like: “I never had this console as a kid so I never got to play this classic game. I wonder what it was really like,” or “I feel nostalgia for this game.” This motivation can provide insight into why classic games usually remain within Twitch’s top 100.

There are other motivations for watching game play on twitch, but these relationships are a high level look at the stakeholders APGs can cater to. These relationships along with audience presence, feedback, and delay are a few of the major design concerns when making APGs.