A toolkit built for the wrong reason — and a better one
It started as a personal reference library. The idea was simple: take complex frameworks from behavioral science, emotion theory, personality psychology, and strategic philosophy — and build them into interactive, cross-linked card explorers that made the material actually usable. Each system would be searchable, filterable, connectable. You could flip a card on Confirmation Bias and click through to Anchoring Effect, to Availability Heuristic, to the cognitive science underpinning all three.
What emerged was something more interesting than a reference library. Taken together, the six systems we built between March and July 2026 constitute a complete behavioral model of a human being: how they feel, how they think, who they are, how they navigate power and social dynamics, what ethical principles they operate under, and how they manage fear and uncertainty. That's not just useful for learning. It's the exact data architecture needed to make a fictional character behave like a real person.
The application to game NPC design wasn't the original brief. But once we looked at what we'd built — 298 emotions with intensity spectrums and combination states, 304 behavioral bias and heuristic concepts, 16 personality types with cognitive function stacks, 48 laws of power with social dynamic categories, 22 ethics principles with cross-linked contrasts, and 71 worry and resilience frameworks — the picture was undeniable. We had built a brain.
Six systems, one stack
Each explorer was built as a standalone interactive tool — but designed from the beginning with cross-linking in mind. Terms in one system reference terms in adjacent systems. The emotion "contempt" in the Emotion Explorer connects to "Dehumanization" in the Behavioral Science Explorer connects to Power Law XIII in the 48 Laws. The architecture is relational, not flat.
What happens when you give a fictional person a real psychology
The original question was straightforward: could we build a village RPG — a Skyrim-style top-down game with 30 NPCs — where each character had a complete behavioral architecture drawn from our knowledge systems? Not scripted dialogue trees. Not branching if-then logic. A genuine personality that shapes every response to every question a player might ask.
The answer required mapping each knowledge layer to a specific behavioral dimension of the NPC:
The result: when a player walks up to the village innkeeper and asks an unusual question, the response is shaped by six simultaneous filters — her ESFJ warmth that wants to help, her current anxious emotional state about the missing merchant, her Availability Bias (she only knows what's happened near the inn), her Carnegie people-skills that make her diplomatically deflect uncomfortable topics, her consequentialist ethics that let her shade the truth if it protects her livelihood, and her instinct from the Laws of Power to make the player feel important and heard while revealing nothing of substance.
That's not scripted behavior. That's a personality.
The relationship as a living variable
In the Thornhaven prototype, every NPC maintains a live affinity score with the player — displayed as a visible rating above their head. But affinity isn't a simple counter that increments on good answers and decrements on bad ones. It's shaped by the NPC's personality type, their current emotional state, and which of their cognitive biases the player has triggered.
An INFJ elder might start cold and suspicious but become deeply loyal once the player demonstrates genuine curiosity rather than transactional interest. An ESTP merchant will warm immediately to confident, direct players and grow contemptuous of hesitant ones. An INTJ wizard ignores social pleasantries entirely — their affinity tracks only the intellectual quality of the player's questions.
The affinity score also feeds back into what information NPCs will share. A guard with low affinity will deflect questions about the missing merchant. The same guard at high affinity will volunteer that they saw a cloaked figure near the north gate three weeks ago and didn't report it — because they felt guilty about it. Information gating through relationship rather than key-item triggers. Motivation rather than mechanics.
The rumor system: information as ecology
The most compelling emergent behavior enabled by this architecture isn't what individual NPCs say — it's how information moves between them. In traditional RPGs, world state is managed by flags: the player tells NPC A something, a flag is set, NPC B checks the flag. It's invisible plumbing pretending to be a world.
With a full behavioral architecture, you can model something much closer to how information actually spreads in a real community. Call it an ecological rumor system:
"A village that knows things the player doesn't — where NPCs have private information, private fears, and private relationships that shape their behavior independently — is a village that feels inhabited rather than decorated."
This is the emergent behavior that a flat dialogue system cannot produce. It requires the behavioral stack: emotion creates the distortion; cognitive bias determines which distortions each NPC is prone to; personality determines how they share it; Carnegie relationship principles determine who they share it with; power dynamics determine whether they share it honestly or strategically; ethics determine whether they feel any guilt about distorting it.
Remove any one layer and the ecology collapses into scripted behavior.
What comes next — and why it matters
The Thornhaven prototype demonstrates something that the games industry has been working toward for decades and hasn't been able to achieve through procedural generation alone: non-player characters who behave like people because they are modeled on the same frameworks used to understand real people.
The path forward splits in several directions simultaneously:
The broader shift in game design
What we have built is not a game feature. It is an epistemology — a structured way of knowing what a person is and how they behave. For the first time, game developers have access to AI systems sophisticated enough to use that knowledge in real time, and knowledge systems comprehensive enough to give that AI something real to work from.
The games being built with this architecture — including early prototypes informed by what we developed at Cat Game Research — represent a fundamental discontinuity in what players will expect from NPC interaction. Not better dialogue. Not more branching. A qualitatively different experience: the experience of talking to someone who has a real inner life and whose reactions are genuinely unpredictable because they emerge from character rather than from code.
The industry term for what traditionally animated characters do is "acting." What behavioral-stack NPCs do is closer to "being." The distinction matters because players feel it immediately — not as a technical achievement, but as the uncanny recognition that the person they just spoke to might actually disagree with them tomorrow.
"The goal has never been more realistic graphics. It has always been more real people. We finally have the tools to build them."
The six knowledge systems built at Cat Game Research between March and July 2026 are the earliest complete public implementation of this architecture. They are also, it turns out, the research library we needed all along — not for learning, but for building something that learns.