An investigation shaped by your questions
Independent design and development · Prototype in active iteration
Blizzard Villa is a text-based mystery prototype set in an isolated manor. Players investigate rooms, question a cast of characters, revisit evidence, and build an explanation of what happened. The broader experience I am working toward also asks players to manage how others judge them: uncovering the truth and convincing people to accept an account are different problems.
I am independently developing the project across game rules, scenario and evidence design, character knowledge, dialogue behavior, and the playable web prototype. The current implementation uses Python, FastAPI, and a JavaScript interface.
The design question
How can players investigate in their own words while still trusting the mystery’s evidence?
An open conversation offers more room for curiosity, persuasion, and unexpected questions. But a mystery becomes unfair if its facts change to fit a player’s guess, if a character knows something they could not have learned, or if essential information depends on a lucky generated reply.
My guiding principle is to keep case facts stable while allowing flexibility in how players investigate and how characters respond. The model interprets language and proposes responses or actions; the game’s rules validate what can be revealed or changed.
Characters with different knowledge
I separate what each character knows, believes, conceals, and has personally witnessed. Discovering evidence as the player does not automatically inform every NPC. Showing it in a conversation is a separate act.
This distinction supports a central part of the intended experience: people can disagree because they have different information or interests, rather than because the world’s facts have changed. The prototype applies these knowledge boundaries to individual conversations and multi-character scenes, with checks on revealing protected facts.
A focused prototype: searching in your own words
The current natural-language investigation experiment is deliberately limited to one room. Players can ask for an overview, browse records, or investigate a specific question. The model selects from authored materials; the engine checks access and returns the original evidence text. It cannot invent evidence to confirm a suspicion.
Iteration exposed a more specific interaction problem: asking broadly what records exist is different from asking to read them, and one sentence can contain several requests. I introduced different reading depths and separate handling of each request. Players can still ask a direct question without being forced through a fixed sequence of menus.
An unsupported action is also different from a completed search that finds nothing. The prototype reports that boundary explicitly so the player can reason from the response.
Testing the experience before expanding the world
The project explored NPC conversations, gossip, and proactive behavior early on. I have since narrowed the active scenario to a more controlled investigation so I can isolate problems in evidence, dialogue, and player feedback before adding more autonomous behavior.
I use fixed-response checks for rules and information paths, then real-model tests for interpretation and dialogue. Passing those checks is useful evidence about the implementation, but it does not by itself prove that the mystery is satisfying to play.
Current focus
This is an ongoing prototype, not a released game. Current work focuses on investigation clarity, character information boundaries, and a reliable playable case. Broader social consequences, competing public accounts of the case, and autonomous NPC systems remain areas for further implementation and playtesting.
The question guiding the next iterations is whether open-ended language gives players meaningful investigative choices while preserving clear, dependable feedback.
