Prey Tell — Team Squirrel: Daniel Contreras-Esquivel, Michelle Li, Xi Liu, Sizhe (Alex) Wu

 

Table of Contents

— — — 

1. Artist’s Statement

2. Playtesting and Iterations

3. Concept Mapping and Ideation Exploration

4. Formal Elements

5. Print and Play

6. Full Representative Playtest

7. AI Disclosure

— — — 

Artist’s Statement

— — — 

Prey Tell is a story where everyone is prey, and everyone is a hunter. For friends old and new who enjoy social deduction games, we invite you to explore our  social game full of  trust, betrayal, and group decision-making. Society is not a simple, black-and-white binary opposition. Instead, it is made of more diverse perspectives, and people’s positions often change based on their current survival environment—they are not set in stone. We take this at its extreme, this game simulates the complex social traits of humans (vegans in this society) living as a group within a “real” society. Enjoy the paranoia of asymmetrical information while battling it out for survival. The three factions that balance each other, Vegans, Cannibals, and Zombies,  will completely break you out of  the old “good guys vs. bad guys” mindset.

In our design, survival is no longer a single-dimension game. The hidden infection of the Zombies, the predatory survival of the Cannibals, and the defensive survival of the Vegans build a social ecosystem full of conflicts. The game is driven by a structured day-and-night cycle: during the day, discussions and arguments form a rational social buildup, while at night, physical contact (delivered by the host) symbolizes hidden power interventions. This mechanic forces players into a fog of information. They can no longer simply look for “the bad guy” as their single goal. Instead, they must build short, fragile alliances through complex trade-offs based on the different survival logics of each faction. “Prey Tell” reveals how we define a “threat” when facing group pressure. This is a game of perception: by forcing different factions with different survival goals into the same space, we seek to examine how players try to define truth within an unstable social structure full of deception.

— — — 

Playtesting and Iterations:

***Legend: [🌟] what happened/changed; [❤️] what people enjoyed; [🤔] what people didn’t like/were confused about.***

— — — 

Playtest 1: CS 247G Tuesday 2A, 30-45 min (7 players):

[🌟] An initial spark from modding midnight werewolf turned our game from partially flushed out (“two rooms and a werewolf?”) to its first breath of life. Our mods went from corporate espionage to Sherlock investigators to food chain. And within this creation, somewhere along the way: the vegans < cannibals < zombies food chain was formed. We added investigator tokens to reflect our Sherlockian espionage origins. [❤️] Our first pitch was made in time to be tested during the back and forth of modding midnight werewolf still, so the players were all assigned by convenience and a second time by shuffling the teams. Right off the bat the concept garnered attention and laughter, people enjoyed the outlandishness of our theme. The three factions made it arguably different from midnight werewolf, and players enjoyed finding the balance in strategy themselves. [🤔] Upon playing, we immediately realized the role of the moderator is way too heavy. A moderator juggling explaining the rules to everyone, assigning roles, tapping the right people to wake up at night, all while trying to walk full circles around and around the table to ensure people don’t hear who got tagged as who. It was just too much to keep track of, and made playing that role too involved and not fun. Furthermore, even in the very first round we play-tested, the role of vegans were a bit tricky and felt overpowered. [🌟] This was our sign to develop game components that could guide the structure of the game. We made color-coded identity cards and changed the physical structure of the game so the moderator is in the center of a circle instead of the outer rim. We still try to enforce walking a full circle (so you can’t auditorial deduct identities) but the circle is hopefully smaller 🙂

— — — 

Playtest 2: CS 247G Thursday 2B, 30 min (11 players):

[🌟] Our very first full playtest with 11 randomly assigned class members (ages 16+). We incorporated our newly made game components, having zombie tokens (yellow cubes), investigator tokens (mushroom looking token), identity cards, and freshly made instructions. [❤️] Positives were still high energy and high excitement towards the game. [🤔] Downsides, the game instructions took forever to explain. There was just so much detail we were excited to share that we forgot people have short attention spans lol. Many people didn’t fully read the instructions before starting, resulting in people asking “what does yellow mean again” after viewing their identity card, the answer is zombie, which the rest of the group have also figured out as their identity 😭 Special roles were unclear, theoverall gameplay was stunted by the fact that our instructions were too information dense. The gameplay we wanted to elicit with our factions can’t be drawn out when people don’t understand the full rules. [🌟] Went back to the drawing board (literally) and drew up better instructions!!

— — — 

Playtest 3: Post CS 247G Thursday 2B, 30 min (8 players):

[🌟] In an attempt to get one where we can receive some more concrete feedback, we ran another play test. Around half of the players self-identified as those who enjoy talking a lot in a game and social deduction-like play. [❤️] The circle formation worked! Moderator could easily get to people and had a smaller turnaround throughout the game. [🤔] Once again received request for images on instructions, but this time we were able to go further than that. We learned important info like people on two ends of the spectrum of volume (spoke too much/little) got voted out first, the first person to talk was also immediately voted out. Zombie tokens were nice to look at but physically impossible to hide in a palm without others teaming up and demanding them to open their palms. The ratio of good to evil was way too low (5:2:1), even with only 2 cannibals and 1 zombie, the vegans quickly got overpowered to the point of no return. Anyone feeling like there is no chance to win is not having fun. Another fascinating thing happened, the zombie, somehow someway, got voted off immediately, we had to temporarily and secretly change zombie to someone else to even test this in game play. [🌟] New changes included clockwise discussion enforcing (everyone gets a chance to speak, no suspicion based on how much of a yapper someone is), remove physical zombie tokens and use thumbs up/down (much more natural and easy to hide), decrease the number of cannibals and make special roles (doctors, psychic) wake up every night to give vegans more power, make zombie go first. All this to help drive more chances to win. Finally, a new and exciting game mechanic we added was paired conversations. The moderator gets to randomly pair people up for short discussion every 3 rounds. We were very excited to see how this would further contribute to the chaos and allow cross team mingling and bluffing on a close-up scale (easier to bluff when hiding amongst a crowd, what if everyone was forced to talk face to face with someone else?).

— — — 

Playtest 4: CS 247G Thursday 3B, 45 min (11 players):

[🌟] This was one of our official final game plays and was super fun and engaging. It’s worth noting that at this point our players were rather well aquainted and had known each other for a while now. It was truly magical watching someone else moderate the game well (we added a moderator script to make things easier) and players being able to explain the rules to each other (our creation came to life thanks to the players yay!!). The players looked attentive when reading the rules (1:12-6:19), and the diagrams helped a lot in offering quick explanations of the textual instructions. Many players being able to answer each others’ questions seems like a good success metric on the instructions iteration 😀 (6:30, 10:43, 10:57, 13:18, 15:18, etc) [❤️] Our paired off discussion mechanic paved way for other game mechanics that have previously not been activated! Immediately upon returning from a paired discussion, the group voted to use their one investigator token chance (29:45-30:20) and successfully gave it to a Vegan to use (who did not lie about the identity though they could’ve, in fact they accidentally revealed the actual card to the crowd 35:55 which drew quite a laugh from the crowd). This means that the paired discussions were able progress the narrative successfully and provided even more opportunities for asymmetric information! It was very enjoyable watching the pairings: e.g. a zombie got paired with a vegan and fully trusted each other, vs another group of two vegans where distrust formed over the course of the paired conversation. The clockwise discussion also really helped certain players not get wiped out immediately. Over the course of playing with the class, certain players have played our game and almost always get chosen out first because they spoke first or more than others (coming off as suspicious), thus this allowed everyone a chance to participate and offer information 13:31. [🤔] Slight problem, perhaps this was because the moderator script wasn’t emphasizing this well, but the moderator forgot to use the doctor every night until the last few nights (23:23), and then the doctor immediately got killed off 😭(26:37), making the vegans suddenly realize they were rather powerless at that point. To further exacerbate the problem, the cannibals and zombies both pretending to be vegans kept convincing the vegans not to vote anyone out in the beginning rounds until they were fully outnumbered. This resulted in vegans feeling like there was very little chance of them succeeding (22:52, 34:30 etc). It got to the point where they’d rather become zombies to avoid this issue. Thus something definitely needs to change for vegans to feel more empowered. [🌟] There were a couple of things we considered doing after this play test: one interesting idea would be no matter which team you switched to, your goal is to make your original team win (42:59 where cannibals are upset they didn’t win as cannibals but as zombies). This would require everyone to only open their eyes for their original team, which we thought would have been a bit sad for zombies who have to just carry out the whole mission alone. Some other things we considered were zombies getting healed, scaling up the number of players, increasing cannibals (to counteract zombies), etc. The final few we actually decided on are the following: [1] change the win incentive for cannibals, instead of directing all their hunger towards vegans, make the game an ALL vs ALL game. This takes away the unbalanced pressure the vegans are feeling. [2] let everyone hunt: alternate the nights from zombies to cannibals, to vegans too, where vegans can have a designated investigator role every 3 nights to help them “hunt” as well [3] let psychic vegan be able to receive a direct hint from a forfeited ghost player. The changed winning incentive and increased vegan powers (hunting, more special roles) should help vegans feel more empowered as a whole. Taken together, we had a LOT of fun play-testing this with such a lively crowd, and are very glad the players had an overall positive experience with it. To test our gameplay to make sure it’s now more balanced, we ran another (less recorded) game play with the same sized group. [video of the paired discussion, all timestamps in Playtest 4 Reflection refer to Full Representative Playtest].

— — — 

Playtest 5: Friday July 10th, 2 hr+ (13 players):

[🌟] After making all of our final changes, we wrapped up the game design and fixed a couple of final small details before testing it again on a large crowd — this time with players who were friends and knew each other more. [❤️] The results were great! The final round ended with one Vegan, two Cannibals, and one Zombie left — we’d finally balanced out the power dynamics. The game received high praise from the group 😀 [🤔] One piece of feedback we got: flip the hunting order from Zombie → Cannibal → Vegan to Vegan → Zombie → Cannibal. The reasoning holds up — if Zombies and Cannibals act before the Vegan phase, there’s a real chance the Investigator gets eliminated before ever using their ability, cutting off the only lead the Vegan faction has. Going first instead guarantees Vegans get at least one clue, giving the group grounds for an informed vote in the early rounds instead of a blind one. It also fixes the Doctor problem from the previous playtest: the Doctor’s whole role is to protect whoever the Investigator flags, but with no clues yet to act on, they had nothing to go on and defaulted to just guarding themselves every night. [🌟] We would adopt the reordered hunting in future rounds of testing. A second suggestion — that a Vegan who gets infected and wins as a Zombie should score “less” than an original Zombie, to discourage Vegans from wanting to defect — is still an open question for us rather than a settled change; we don’t have evidence yet that this is an actual problem in play, and it would require adding a graded-win concept the game doesn’t currently have. We’re holding off on it until we see it show up as a real issue. 

***Player testimonies after the game ended***

1. 1. From Eugenio (The doctor vegan): https://drive.google.com/file/d/1KQ6mCn8NqKUa7t4laBOG6fKI92T_RPNr/view?usp=sharing
   2. From Hurst (The moderator): https://drive.google.com/file/d/1weF3Ym-2yJXrSXBRz1o6yaOcq5b_Aodf/view?usp=sharing
   3. From Jason (The zombie): https://drive.google.com/file/d/1ypeZsnp4E28Zv6PWbX1Q1LCbQ387rN2m/view?usp=sharing 
   4. From Aileen (The vegan): https://drive.google.com/file/d/1AwSqqTpd2I4ptHvkWXleQhSN74XXOkUX/view?usp=sharing

— — — 

Playtest 6: Saturday July 11th, 1 hr+ (7 players):

[🌟] We’d wanted to hit 10+ players again for this round but couldn’t get enough people together, so we ran it small instead — moderator plus 4 Vegans, 1 Cannibal, and 1 Zombie, the floor of our stated player count. It turned out to be useful in its own right: testing below our target size is exactly how we start to understand where the game’s balance actually breaks as player count shrinks, giving us our lower bound. [🤔] Playtest 6 ran with two powered Vegans active at once (Doctor and Investigator), and that combination made Vegans noticeably too strong to beat at this smaller scale — with fewer Cannibals and Zombies to begin with. [🌟] Our takeaway: at 7–9 players, only one powered Vegan role should be active — the Investigator specifically, since it gives Vegans a way to act instead of just react, without stacking a second power on top of it. The 7:2:1 ratio and full role set (Doctor, Psychic, Investigator) should hold as-is for any game of 10+ players plus moderator. We haven’t playtested anything beyond ~13 players yet, but based on how the ratio scales, we’d guess the format realistically caps out somewhere around 22–31 players before the night phase and moderator overhead become unwieldy — that’s still an open question for future playtesting, not a conclusion. [video  of playtest 6]

— — — 

🤖 Attempted Agentic AI Playtesting 🤖:

[🌟] Human playtesting doesn’t scale — every session needs 10+ people in a room, which made it hard to run enough trials to catch balance issues like the ones above. So alongside in-person sessions, we built a parallel harness: a pure-Python rules engine with LLM agents plugged in as the players, each one scoped to only the information their role should legitimately know. [🤔] We tried three model backends before running into a wall with all of them. Gemini capped out at a free-tier limit of 20 requests/day — a single 10-player game alone burns pass that. DeepSeek’s API uses dynamic pricing so while at times the cost of class/tokens per day were very low but not guaranteed to be zero. We migrated to OpenAI (GPT-5, via Stanford’s ChatGPT Edu Open API) last, but hit API quota limits before we could get a full batch of simulated games through — so we don’t have red-flag stats (win rates, early-death rates, etc.) from this harness yet. [🌟] What we do have: the full pipeline debugged and working end-to-end, set up to run on Stanford’s Farmshare cluster. In fact, the rules engine is ready to plug a live client into if we had additional time to build and debug a functioning digital version of the game. You can find our repository here. [CS247G-P1-Playtesting.git] 

— — — 

Concept Mapping and Ideation Exploration

Our concept mapping wasn’t a single upfront exercise — it was iterative, running alongside actual playtests rather than ahead of them. Each session surfaced a structural tension (Vegans’ blindness, the Doctor’s lack of leads, hunting-order asymmetry) that we mapped out as competing ideas before deciding which to keep, reject, or leave open. We found concept mapping useful in developing direction and reflecting on iterations — e.g. origin theming, mechanic patches, then a larger structural pivot — with rejected ideas leading us closer to a balanced and engaging game. Many of our diagrams made it to the final Print and Play ruleset of the game (please reference the Print and Play section).

 

 

 

— — — 

Formal Elements and Values of Prey Tell：

Description and Format: A bluffing 3 faction (Vegans, Cannibals, Zombies) social deduction card game. 

— — — 

Objective and Win Conditions: Zombies win by silently converting players and gaining numbers, Cannibals win by loudly consuming from both factions, and Vegans win by out surviving and out informing both threats using their identity-based powers (Doctor, Psychic, and Investigator).

— — — 

Players: The game has two modes (1) for 7-10 players we recommend 1 moderator, 1 Cannibal, and 1 Zombies, and (2) for 10-22 players we recommend one moderator with an initializing 7:2:1 split of Vegans:Cannibals:Zombies. 

— — —

Resources:

• 🪪 Role cards 🪪 — Vegan (optionally Doctor/Psychic/Investigator printed on it), Cannibal, or Zombie identity
• 👍 Thumbs up/down 👎 as the silent infection signal
• 🗣️ Private pairing conversations 🗣️, held every 3rd round, 1 minute, in addition to group discussion

👻 Ghost forfeit 👻 — a dead player’s option to trade their one remaining vote for staying awake at night

— — —

📝 Procedures and Rules 📝: The moderator/narrator dispenses identity cards, ensuring players familiarize themselves with their respective role. Special-role cards (Doctor, Psychic, Investigator) are printed directly on the Vegan identity card.

😴 Night — a 3-night rotation (Zombie / Cannibal / Investigator), repeating 😴: Eyes closed, thumbs up. The Doctor Vegan wakes first, every night, and protects one player. If there are forfeited Ghosts, the Psychic Vegan wakes up next and observes who the Ghost(s) point at to target (Psychic doesn’t know the Ghost’s faction affiliations). Afterwards, either Zombies hunt, Cannibals hunt, or Investigator Vegan inspects (if alive) before ending the night. 

• Mechanic: this rotation means Vegans face exactly one hunting threat per night instead of two simultaneously, and the fixed Zombie→Cannibal→Investigator order is public knowledge, so the timing of danger is predictable even though the target isn’t.
• Dynamic: players can reason “this is a Cannibal night, so whoever’s acting strange today probably worried about being eaten, not infected.” Once the Investigator’s mechanic (moderator directly reveals ground truth) enters this rotation, the Investigator becomes the only player in the game with unmediated certainty about anyone — everyone else is working from inference, claims, or a Ghost’s motivated silent point.

☀️ Morning ☀️: The moderator announces who was eaten. Infections are never announced. The agenda for each day is (1) Pair Discussion (mod 3 rounds only), (2) Group Discussion, (3) Voting to eliminate. 

• Mechanic: asymmetric disclosure — public for eating, silent for infection.
• Dynamic: suspicion calibrates almost entirely off the public eaten-list, letting converted Cannibals continue as double agents since their conversion leaves no morning trace.

☀️ Every 3 Rounds — Private Pairing ☀️: Random pairs, 1-minute private, discreet exchange, in addition to group discussion.

• Mechanic: a communication channel invisible to the rest of the table.
• Dynamic: enables targeted, unverifiable claims between any two players, living or dead.

☀️ Discussion ☀️: Two minutes, clockwise order, random first speaker (youngest, oldest, or guessing metrics can be used to set the rotation order the first morning).

• Mechanic: fixed turn order, rotating start.
• Dynamic: spreads social pressure evenly instead of concentrating suspicion on whoever happens to speak first — a direct fix from your Checkpoint 1 playtesting notes.

☀️ Voting ☀️: At most one player voted out per round, they become a Ghost. Ghosts may speak in discussion going forward but must keep eyes closed at night, unless they’ve forfeited.

• Mechanic: plurality vote, immediate elimination, but death no longer removes a player from the conversation.
• Dynamic: this changes the stakes of being voted out — a Ghost isn’t silenced, so the vote is a demotion (loses the ability to act at night/normal-vote) rather than a full removal from the social game. It also means limited information/opinions a Ghost has can still reach the table through ordinary discussion, not just through the Psychic’s private channel (post-forfeiting).

***Night and Day repeat until a faction’s win condition is met, then the moderator reveals all roles.***

— — — 

Conflict: Each faction weighs threats asymmetrically instead of everyone piling onto the most obvious enemy. Vegans and Cannibals both fear infection from Zombies; Cannibals and Zombies both fear expulsion by Vegan votes; Vegans and Zombies both fear being hunted by Cannibals. This triangle means no two factions are ever fully aligned against a third — each adversary is simultaneously a predator to one group and prey to another. This shared-threat structure produces two different modes of deception. Cannibals and Zombies each know their own teammates, so they bluff defensively — lying to protect their own identity from a table that can vote them out, with the safety of never risking a misfire against an ally. Vegans have no such certainty: they don’t know who the other true Vegans are, so their manipulation is blind, and a misdirected accusation risks landing on a hidden ally as easily as a real threat. Instead, Vegans rely on their special roles — Doctor, Psychic, Investigator — to convert that blindness into information. But those roles share a single point of failure: the Doctor can only protect one player per night. If both the Psychic and the Investigator are threatened on the same night, the Doctor can only save one.

This tension is the game’s aesthetic core, and it plays out differently for each faction. For Vegans, Discovery drives the whole arc — every claim, silence, and vote is a fragment of evidence in an identity puzzle solved with incomplete information — and Fellowship emerges from needing to trust strangers who might be lying, since there’s no shared team-knowledge to fall back on. For Cannibals and Zombies, the same triangle produces two different aesthetics rooted in their opposite relationships to teammate knowledge. Zombies’ aesthetic is rooted in Expression because infection is never announced and a converted Cannibal keeps waking up with their old team, the pleasure of the role is performative — maintaining a normal face while secretly building toward a silent majority — and their Fellowship is delayed and asymmetric, growing only as more players are quietly converted, rather than starting from a fixed, known team. Cannibals, by contrast, know their teammates and targets from night one, so their aesthetic is closer to Challenge/Competition/Tension — each night is an active resource decision since Cannibals’ numbers are fixed and countable from the start (2 of 10, never growing), unlike Zombies, whose threat never becomes visible as they silently expand.

— — — 

Want to Play?

You can download a print-and-play copy of “Prey Tell” by this link.

https://drive.google.com/file/d/1HOCRTB-zW9BhFQ_bRy5aWR5kaEEaxoxK/view?usp=sharing

— — — 

Full Representative Playtest:

https://youtu.be/ELbAdj_ivFY?si=eb8RXQCVxoWZh3Mv

5 timestamps (reference Playtest 4 for additional context and additional timestamps): 

15:00 — 27:45 — 36:00 — 43:10 — 44:50

First Ghost—Pair Discussion—Deprecated Inspector Token—All is Revealed—Feedback

— — — 

AI Disclosure:
Cover image and role images were generated by Gemini. Claude Edu was used to debug and clean up the rules engine and LLM playtesting harness. Gemini, DeepSeek, and OpenAI models were each tried as the LLM backend powering simulated players in the agentic playtesting harness. OpenAI Edu and Claude were used for spelling, grammar, and clarity edits of different blog sections.

— — — 

Thank you for taking the time to read 🙂