One of the announcements from the January Monthly Update from GMT Games was the news that there would be a new version of the solitaire bot for Fire in the Lake. The designer for the new bot system will be Bruce Mansfield who created the very cool and easy to use Arjuna bot used in his game Gandhi. We reached out to Bruce and he was very willing to give us the scoop on the bot design process.
Grant: Why do you think Fire in the Lake needs a new solitaire bot? Wasn’t the bot just updated?
Bruce: From the very start of my work on Gandhi, I knew I wanted to design a different bot, but we didn’t settle on the card-based system that became Arjuna until late in the project, near the end of 2018. By that point, the update kits for the existing COIN games were done and starting to ship; Arjuna was an untested design. It wasn’t until the release of Gandhi in the summer of 2019 that we got a chance to see how players would respond to the new solitaire system. From talking with players and reading online reviews, we were happy to see they liked the new system and they understood our design goal of building a bot that was quick and easy to run but felt like playing against a human opponent. We have since received several requests to backport the system to earlier COIN designs, but it wasn’t until Volko Ruhnke asked me to design an Arjuna-style bot for Fall of Saigon that I seriously considered retrofitting an existing design. I knew that players would want to play the full scope of the Vietnam War from 1964 to 1975 with both Fire in the Lake and Fall of Saigon, and I also knew that many of the design problems that I would have to solve to make the system work in Fall of Saigon would also work for Fire in the Lake, so this seemed like an obvious place to start.
Grant: How much did Arjuna influence Trưng? What things did you keep and what did you have to create from scratch?
Bruce: We wanted to keep the design for Trưng (“Troo-ng”) as close to Arjuna as possible, even copying the graphic design of the cards and the various tables. But Fire in the Lake is a different game than Gandhi. For one, as the designer of Gandhi, I not only knew that system inside-out, I was able to make changes to the multiplayer game to accommodate the solitaire game. For example, in order to eliminate the need for additional Event instructions, COIN Series Developer Jason Carr and I rewrote many Events to clarify how Arjuna would execute that Event, and we streamlined the Operations and Special Activity procedures to facilitate bot play. This isn’t something I can do with Fire in the Lake; nor would I want to mess around under the hood with Mark and Volko’s masterpiece! But I also found that the Arjuna system is very flexible, so we didn’t need any major new mechanics. The Trưng cards look very similar to the Arjuna cards and the Space Selection Priorities and Move Priorities tables look similar, tailored of course to the unique needs of Fire in the Lake. I was also able to streamline the system. For example, the Move Priorities table handles all the ways that pieces move in Fire in the Lake: Patrol, Sweep, Air Lift, Transport, and March.
Grant: Where did the name Trưng come from?
Bruce: Jason and I were inspired by Mark Herman (Erasmus, Brasidas, et al.) to name each system after an important cultural or historical figure; I chose Trưng because the story of the 1st century CE Trưng sisters is an important one in the development of Vietnamese nationalism and was a favorite of my students when I taught world history.
Grant: How much has Jason Carr been involved in this design process? What ideas has he offered for the bot?
Bruce: Jason has been an important part of this project from the beginning. We work very closely and discuss design ideas almost daily. He is particularly good at seeing the big picture and keeping me focused on the end goal of creating a functioning bot when I get lost in the weeds of details.
Grant: I understand that your lead playtester is Scott Mansfield. Who is this to you and how do the two of you work together?
Bruce: Scott is my brother and lifelong gaming partner. Along with Jason, Scott is the third member of my design team. He is very strong in both visual design as well as providing feedback on playability. He helped me develop the original design of the Arjuna cards, making sure we found a balance between information density and useability. Scott is focused on other projects now, so he’s taken a step back from being lead playtester for Trưng. He is also a professional photographer and does all the photography for GMT; you can see his work on their website and social media posts.
Grant: Does this bot cover the new Fire in the Lake expansion Fall of Saigon? Why not?
Bruce: This bot only covers Fire in the Lake, but we will also build a card-based bot for Fall of Saigon. Jason and I felt that the strategic needs of Fall of Saigon were sufficiently different compared with Fire in the Lake that it warranted new cards and tables, especially for the ARVN and NVA.
Grant: One observation from our uses of the bots in COIN Series games is that the bots play very skillfully. Why do they play so much better than us? What is this devilry?
Bruce: COIN is such a wonderful system because the play space is so open. There is enough complexity that each title is worthy of study, and each new gaming session offers insights into the design and topic’s history. But this also makes COIN games challenging to learn. I have heard from several players that they use the bots when first learning to play because they provide examples of play for each Faction and narrow the decision space to more manageable levels. They do this because the bots primarily serve to limit choices, ideally down to single move (e.g., “of all the things that the Raj can do right now, what is the one move that it should take?”). The ability of the bots to separate the signal from the noise of the game is, to my mind, also the skill it takes to play these games well as a player. There are moments when as a player you may be susceptible to analysis paralysis given a particularly complex board state, but the bot never suffers from this. So it sometimes makes plays that catch the player off guard.
I also think that the experience of seeing the bot play well is a feature of the kinds of narratives these games create. After all, we have brains evolved to think in terms of stories, and will happily invent causality and intentionality even when absent because stories about invested agents are more interesting than stories about algorithms. In terms of the design of Arjuna and Trưng, we exploited this feature (it’s not a bug!) by working to remove the feeling that you’re playing against an algorithm, knowing that your brain will fill in the details it needs to maintain the narrative. So I agree with you, the bot sometimes acts like it’s playing you as much as playing the game—especially when it makes just the right move needed to frustrate your planning. Of course, the bot also does a lot of dumb stuff too, but so do humans!
Grant: When selecting Operations and Special Activities how does Trưng choose?
Bruce: Trưng uses the same table that guides Arjuna: it prefers to take Operations plus a Special Activity in lieu of an Event, but will select a critically important Event when available (typically, ones that allow it to immediately gain victory points). Trưng will also select Operations only to block another Faction selecting its own critical Event, like Arjuna.
Grant: Can you show us a few examples of the Trưng cards and interpret their chosen action path?
Bruce: Here’s an example of a VC card. Each card attempts to meet or more towards one of that Faction’s strategic goals. In this case, the bot is looking to see if it can send US Troops to the Casualties box. The VC do not have troops like the NVA, so the best way for them to remove US forces without exposing themselves to much risk is via Ambush. So first this card checks to see if any spaces meet the basic conditions of Ambush—spaces with US Troops and Underground Guerrillas. If so, the card next checks to see if there are more pressing strategic needs. First, it checks to see if even has the forces it needs on the map by rolling three dice against the number of Available Guerrillas; if successful, VC will Rally to get forces onto the map. If there are plenty of Guerrillas on the map, however, the card is flipped over. Then, the bot looks to see if there are any spaces where it could execute Terror—spaces with two or more Underground Guerrillas not at Active Opposition. Although it’s looking to hurt the US, if there’s a chance it could gain victory points, it will do that first. If not, it selects Attack and will first interrupt the Operation to execute Ambush, targeting US Troops per their Space Selection Priorities table. For other Operations, the VC will select a Special Activity either to add to its Agitate Total (which allows it to Agitate during the Coup Round) or Subvert to remove ARVN and control Patronage.
Grant: In my experience the Factions in Fire in the Lake are fairly complex. How does Trưng deal with that complexity? Does it feel like you are playing a human opponent?
Bruce: This has been one of our biggest design challenges. Many Operations and Special Activities have multiple elements and several consist of different types of actions. US Advise, for example, involves elements of Assault, Sweep, and a version of ARVN Raid. We solved this in several ways. For one, we created components that can handle multiple actions. We do this by creating instructions for a type of action, rather than the Operation or Special Activity itself. This is most evident in the Space Selection Priorities Tables, which form the heart of the system. Need to remove an enemy piece? Consult the Remove or Replace column. Need to put cubes on the map? Use the Place Cubes or Special Forces column. It doesn’t matter if this happens during an Operation, a Special Activity, or an Event. This also speeds play because players do not need to learn and internalize separate systems for different actions; once you know how to use the selection tables, you know most of the system. Another way we handle complexity is by adding additional conditional statements and checks to the cards and tables. Gandhi needed a limited number of these, but Fire in the Lake requires more simply because there is more going on. So Trưng will look at several different measures of the board state before it acts.
Grant: What is the logic behind the random Space Selection Priorities tables? Can you show us a few of the Faction’s tables?
Bruce: The cards get a lot of attention, but the Priorities Tables are the heart of the system. These were originally inspired by Adam Zahm’s work for Labyrinth: The Awakening. Trưng uses three kinds of priorities tables: one for each Faction to select spaces, one for all Factions to facilitate moving pieces, and one for all Factions to select pieces. The power of these tables lies in their ability to handle several different types of actions with the same table. The Move Priorities table, for example, handles movement as a result of Patrol, Sweep, March, Air Lift, and Transport. The Pieces Priorities table handles selecting pieces, whether it is enemy pieces to target during an Assault, or friendly pieces to move during an Air Lift.
The logic of these tables is complex and layered because they are both nested and sequential. Take a look at the ARVN Space Selection Priorities Table. Say the ARVN bot selects the Transport Special Activity. We use the Sweep or Transport Destinations column to select each destination. First, we look for spaces on the map within South Vietnam that do not have COIN Control. If there are several, we look for such spaces that have the most Population. If there are several of those, we look for such spaces that have an enemy Base. Finally, if there are several spaces in South Vietnam with no COIN Control, the most Population, and an enemy Base, we select the one with the most COIN forces. If there is still more than option, we select the destination space at random.
What is the ARVN bot doing here? It has two primary goals: gain victory points by adding COIN Control and gaining victory points by increasing Patronage. Since Transports doesn’t affect Patronage, it instead tries to add as much COIN Control as possible. Obviously only spaces without COIN Control already qualify, and among those the spaces with the most Population will earn it the most victory points, typically one of the many 2-Pop spaces (as Saigon usually has COIN Control barring some disaster like a massive NVA March). If ARVN were only looking to add COIN Control, we should have it move into spaces with the fewest enemy pieces. But during testing we found that with this instruction, the ARVN bot was too wary of entering enemy strongholds, and thus moved within an ever-shrinking territory as it lost spaces to the VC and NVA. To counter this, we have ARVN push into spaces with an enemy Base, likely places where enemy forces will build up. We let the US worry about those enemy pieces. Likewise, if there are several enemy strongholds, we move into the one with the most COIN forces, knowing that the US could then use these ARVN Troops to Assault during its turn.
Grant: What has been the play experience of you and your test team? What does Trưng do extremely well?
Bruce: We have a small but dedicated playtest team who has been putting the system through the wringer for the past two months. The playtesters have been terrific: they play a lot of test games and their feedback is prompt and useful. We couldn’t have done this without them. They report that Trưng runs smoothly and offers a compelling game narrative that lets the player focus on the game, not on running the bots.
Grant: What are problems you had to overcome and how did you accomplish your aim? Please give a few examples.
Bruce: One of the ways that Trưng is different than Arjuna is that it needs to work with a much more complex game. There are simply more moving pieces in Fire in the Lake compared with Gandhi. One area that required extra efforts was the way we handle Resources. Like Arjuna, Factions run by Trưng do not use Resources. We found that having the bots track Resources only gives the player a way to game the system. Instead, we use a die roll to control total spaces selected during Operations: roll above a certain number and the bot selects another space. For the NVA and VC, the system we used in Gandhi works well.
But for the COIN Factions, who share Resources, this wasn’t sufficient. Although the US doesn’t spend Resources for its Operations, it does spend ARVN Resources during Pacification and Train Operations that place ARVN cubes. When both COIN Factions are bots, we use die rolls to limit their actions. But when only one COIN Faction is a bot, this simple system caused all sorts of troubles. Say the US is a player Faction while ARVN is run by the bot. How many Resources should the US player get? How should we handle Aid and Econ? When ARVN gains Resources, how many of those should be available to the US player? And so on. Initially we had an overly complex set of rules to handle these situations. But it never worked correctly. One of our core design philosophies is to prefer simplicity over cleverness. So we just kept the original, four-player system: whenever one of the COIN Factions is a player Faction, we track ARVN Resources normally. So if the US were a player Faction and ARVN were run by the bot, ARVN Operations are limited by a die roll but they spend Resources that could be used by the player US Faction. It’s a simple solution.
Grant: When do you anticipate the design wrap up to be?
Bruce: We plan to have primary design done by the middle of March, then we send the cards to GMT for final art. After that, we work on getting the rulebook and playbook done (which we’ll package in one booklet). We would like to be completely finished in the next two months so we can shift our attention to adapting Trưng to Fall of Saigon.
Grant: Have you applied your bot skills to any other existing games? What might we expect to come out in the future?
Bruce: Not yet! Currently, the system is tailored to fit COIN games. We are working on retrofitting several existing COIN games at the moment, and are working on developing card-based bots for several upcoming COIN designs.
Grant: What other designs are you working on?
Bruce: Currently, Fire in the Lake takes up all my design time. After this project wraps up in a few weeks, I will start working on the bots for Fall of Saigon. After that…well, you’ll have to wait and see!
It is amazing that solo gamers have so much to look forward to with these new and improved bot systems for the COIN Series as well as other games. Having an easy to follow and use system makes playing solo games so much more satisfying and feasible, especially for someone like me who struggles sometimes with rules interpretation.
If you are interested in the Fire in the Lake: Tru’ng Bot Update Pack you can pre-order a copy for $17.00 from the GMT Games website at the following link: https://www.gmtgames.com/p-856-fire-in-the-lake-trung-bot-update-pack.aspx
These “interviews” certainly give a great deal of detail about design considerations and I’m grateful for them. I would suggest they might be even more engaging if they didn’t stem from what are obviously written questions … in other words, if they read more like actual conversations, not interrogations. You could always follow up with written questions to help fill in any gaps you’d like explored further. Perhaps that’s all a bit hard to schedule and facilitate.
At any rate, thank you for the valuable service you provide.
LikeLiked by 1 person
Thanks for the comment. As it is a written blog it is what we can do. We do video interviews for the YouTubeChannel as well so maybe check those out for more conversational style.
LikeLiked by 1 person
Fingers crossed they adapt Arjuna to every other COIN game. But A Distant Plain next. 🙂
LikeLiked by 1 person