Week 6 – Technical Achievements

What We Did This Week

This week was primarily prep for our first playtest where we had other people relying on us.  We polished off our gold spikes and managed to hire actors to realize the three games we intended to run through on Sunday.

In terms of tech, we fully connected a chat scraper with Twitch extension visualization.  Generally, this means we can read any chat message and get it to whoever or whatever screen element will need it to have it displayed over the video feed.  A variation of this will be used for our most basic playtest of Freeze Tag. Randomly selected audience suggestions will be voted on throughout a performance, dynamically changing what the actors will be doing.  For example, the audience will be able to click to vote on four options during a scene. The winning vote will determine the next scene.

The second game we plan on testing is a TSA Game.  Audience members submit drawings of the contents of a traveller’s bag.  One performer as a TSA Agent will interrogate the traveller as they try to justify increasingly random and out-of-context security X-Rays.  Our barebones model will have audience members drawing on their own computers and submitting image files to a Google Drive folder for us to manually pull into the scene, but if the playtest goes well we hope to explore better image submission methods with less friction.

We experimented with a green screen for the TSA game, as it would be an easy way to get everyone on the same page about what image was being talked about, instead of a client side image updating at different times than the actors’.  For our Sunday playtest, we’re settling on a second monitor over a green screen until we get our game flow down, but still this is an interesting lead.

Something we didn’t consider with the green screen is the potential for it to provide immediate context.  Improv games generally build their scenes and characters up from nothing over time, and it can be confusing to jump in the middle.  While we are trying to minimize our focus on literal exposition, games that feature immediate, visual context narrowers like scene backdrops or specific video overlays seem like a good way to handle Twitch audience drop-ins and drop-outs.  This is still an idea in progress, but a method we’d like to pursue.

Our third game we intended to explore was Parasocial Activity, explained in a previous blog.  We’re uncertain that we’d be able to test it properly this Sunday, and will push it back until it’s in a playable state.

With that, we’ll report back how all of this goes after our playtest.  It’ll be exciting to work with real actors finally, especially after all the hard work getting interfaces and usability up to par for them.  Their insight will most likely greatly influence our project going forward. For next week, we have halves coming up fast, but we hope to use it as a good time to set our thoughts out more clearly. We’ve brainstormed future games with more nuanced questions behind them for next week as well, so we’ll see how this all goes.

Week 5 – What We Talk About When We Talk About Playtesting

What We Did This Week

Employing all of the questions and research areas raised over the last couple of weeks, we had fruitful near-daily brainstorming sessions this week. We turned last week’s foundational iterations into more interesting and complex games, aiming to answer some of our questions while coming up with even more.

We wanted to pin down ideas from the improv side of things first. We all are accustomed to games and the usual Twitch culture, so this approach let us see things from a different angle. Our qualitative method was slightly hard to make good games from, but breaking ideas down like this was important to our need to come up with design frameworks by the end of our project. We took these ideas and later mixed and matched them to come up with a smaller list of thought-out games, and then later pared that down to a list of three to try by the end of the week.

With these improv qualities in mind, we also wanted to stray away from voting as a primary game mechanic. This was what was used in nearly all audience participation games before, from Choice Chamber to Twitch Plays Pokemon and so on, but we feel Twitch can go much farther. We quickly realized the difficulty of this; many, many things can be reduced down to just a form of voting. With some finagling, some of our games to playtest are too.

The first was Freeze Tag, the roughly playtested game from last week. We developed this idea into more or less being a straight translation to Twitch. Important things we took from it are the need for attention management in the audience and how to best train the audience. Improv actors with consistent audiences tend to train their audiences on what to suggest and what not to through creating a positive culture. Spiffy retorts to off-color suggestions or making the crowd side with the performer are important tools, and things we didn’t consider before playtesting this straight translation. This will be a challenge if we don’t get a handle on having stream regulars and moderation, but these two things can be interesting tools added to an improvisers arsenal.

Basic visual aids were important to us to understand game flow now, and even more so for later transformation into interaction diagrams. Above is an example from Parasocial Activity

Another was Parasocial Activity, a game for two players. When a new customer arrives at a medium’s seance parlor, they don’t know what to expect. The medium must tell them something, and has to channel the spirits of Twitch chat to do so. Unfortunately, while channeling spirits, the medium can’t communicate that information directly. It’s up to the ghosts of chat to provide phrases for the medium to communicate with their customer. While this game is technically a single-item, single-vote, the necessity for conversation and selective answering lets it stand out. We’re excited to see this one through.

A Weakness

We are not expert improvisers, and this week showed that to us fairly well. While our playtests are decent at seeing a crowd react and their feedback, our own performances are lacking. We knew this going into the project, and are at the point where we need to start making this whole process worthwhile for those who love improv. We aim to fully roll out proper, ethically studied playtests with real actors next week, upgrading to those with proper training and talent. This will be a good step forward for our own theories and thoughts about improv on Twitch, to see if anything we’ve talked about can be used outside our own heads and friends.

Week 4 – Quarters and Servers

What We Did This Week

After last week’s consolidation of all our research, we were ready to present it to faculty on Monday for quarters.  This test of our knowledge went well, especially in how it solidified our own understanding of what we’ve been reading and absorbing the past couple of weeks.  We feel much more confident in our knowledge base after repeating and refining it in all these mini-presentations.

After making small refinements and focalization changes on our Role and Skill Level matrices, which you may remember from last week, we focused our presentation on them.  Explaining how we plan on exploring different role intersections seemed to be the best way to get across what exactly we planned on doing, along with clearing up that we will not be making one experience but multiple prototypes in these different areas.  Nailing down what audience we were going for was a more difficult area. Our views and scope on it were challenged, and led to us changing our plan.

Slightly refined with a more selected audience

We were construing audience research validity as a result of market appeal a bit too much.  As a discovery and research project first and foremost, we don’t really have to worry about making our experiences commercially viable, but we still had to keep true to both Twitch and improv.  We’re still working out this tricky line, but our impression is that it is safe to rely on expert/experienced improv performers as streamers for the purposes of our research rather than try and create for the existing sphere of normal streamers.  While a core part of improv and game design is implicitly teaching their processes, we’re more focused on finding a base level of fun at all in this intersection. We tightened our scope and our audience to be only experienced improvisers and average Twitch audiences.  This isn’t feasible in every playtest, but we do plan on working with professional improv troupes to make it happen.

Our list of design conventions was also heavily updated for the presentation, but ultimately didn’t get much use.  We think it is still integral to our understanding of the extratextual elements of each’s form, important to keep in mind when trying to translate between the two.

More or less the same as last week

Briefly before quarters we also had a phone interview with Seth Glickman, author of some of the research papers we’d been processing.  This was mostly a technical discussion on how to bypass latency on Twitch, but it also brought up the weird atemporality of interest curves with livestreamed content.  If someone can drop in to your experience, your stream, at any time on the interest curve, how could they feel the full experience? We considered this for a bit, but received the most helpful response to it during quarters.  If our content is good, then of course viewers would be encouraged to find out what was all going on. All we should do is keep our experiences episodic and manageable. New viewers should always feel that their potential to understand is within reach, and further be pushed toward finding it by the experience’s own engagingness.

Brainstorming, Ideation, and Playtesting

The other big event this week was kicking off our formal brainstorming for all the experiences we’ve been talking about.

Our goal starting out was still to translate the most basic improv experience to Twitch and stream it.  We’d have to first understand what works on that level before elevating or transforming it. We decided to implement a sort of freeze tag cycle focused on quick changes in scene, character, and whatnot.  So, we went with a rolling suggestion box for the audience and the ability to vote on what the improviser streamers would incorporate to their next brief performance. The main ideas at test here were multitasking and audience agency.

We ran a brief paper playtest of this process, and it went decently well.  It generally confirmed that this process is possible, and the faults in it, such as attention management and content quality, made us more excited to see how they’d change, if at all, on Twitch itself.  More on this after more formal playtesting next week.

Technical Gold Spikes

We had a lot of time to think about how to go about our first gold spikes for an audience input implementation.  We also had a great starting point with Joseph Seering’s bot codebase and various helpful parts from the HCI up on main campus.  Our first technical accomplishments had chatbots up and aggregating submissions quickly, followed by presenting this data for the audience to vote on, and then tallying them up.  We’re still deciding how best to spread the architecture between bots and Twitch extensions as it grows.

We have higher UI aspirations, such as letting audience members click on the screen and other visual input methods, but for now we are satisfied with strictly text based interactions to get playtesting off the ground.  After our paper playtest, we are ready to ease into the weird space of playtesting with people strewn around the world by getting a feel of Twitch next week.

Week 3 – Making Matrices

What We Did This Week

We focused on distilling the research we did over the past two weeks into more digestible formats, such as diagrams and matrices. What we found especially helpful was making a role matrix for improv and Twitch, which made it clear where the available roles in both areas intersected, and where there was the most room for exploration.

Another important diagram we made was an improv skill level matrix between performers (streamers) and audiences (viewers). Again, this made it pretty clear where there was the most potential for us to make something.

We concluded that areas with the highest amount of potential are where streamers take on the role of performers, viewers take on the role of audience members or performers, and where both streamers and audience members have an amateur amount of experience with improv. However, as a discovery project, we’re planning on designing vastly different prototypes that will place the streamer and audience in a variety of roles, and see how playtesting goes from there.

To help with that, we also came up with a list of design conventions for both improv and Twitch. We’re planning on referencing these heavily during our game design brainstorming sessions.

While some of our meetings this week were disrupted by the university snow days, we managed to meet with Joseph Seering, an author of some of the papers we read for research. We discussed the practical challenges of playtesting experiences developed for Twitch, as well as some other interesting developments in his research, such as a VR storytelling improv game that involved zombies. Cool!


We’ve decided that we definitely need to team up with a local improv group for design and playtesting purposes–some of our prototypes will require that the performers have a professional level of experience with improv, and having more professionals to bounce ideas off of will be helpful when iterating on our designs.

We’ve also decided that the audience for our designs will be the average Twitch viewer, who is someone who knows how to use Twitch, but may not necessarily have any experience with improv.

Research Paper

We’ve also decided as a team that we’re working towards writing a paper on our findings through playtesting of our prototypes. While the final paper won’t be considered a deliverable for our project because of the short time frame we have, we’ll be conducting playtesting in a more rigorous manner so we can preserve the data gathered for research.


With 1/4 walkarounds coming up next Monday, we’re looking to get valuable feedback from the faculty about our process so far, and what problems they foresee us running into.