🎮
Game World (C# + SMAPI)
A SMAPI mod captures trades, gifts, attacks, and season changes, then emits JITNAGameEvent packets instead of leaving behavior trapped inside game scripts.
GameEventObserver
RCTClient (WebSocket)
ActionDispatcher
Back to Use Cases
🎮Case Study - Game AI Integration
Stardew Valley x RCT Ecosystem
Stardew Valley × RCT Ecosystem
A production-style game AI case study showing how FDIA, JITNA Protocol, and governance logic can convert 28 NPCs into deterministic AI agents with sub-1ms heuristic ticks.
28 NPCs
AI Characters
16 Events
Event Types
<1ms
Decision Latency
420+
Test Cases
Brain Plug-and-Play Architecture
Instead of rebuilding the game, RCT injects an AI control plane into an existing game loop through a WebSocket bridge. This keeps the game logic intact while moving cognition, arbitration, and governance into the RCT kernel.
🔗
WebSocket Bridge
A bidirectional JSON transport layer moves live state between the game runtime and the RCT kernel without embedding Python cognition directly into the mod layer.
JSON Serialization
Bidirectional Flow
Thread-Safe Queue
🧠
RCT Kernel (Python)
The kernel applies FDIA reasoning, governance controls, and deterministic command generation before returning a JITNAGameCommand back to Stardew Valley.
StardewAdapter
FDIA 8-Phase Loop
PelicanTownWorldBuilder
FDIA 8-Phase Decision Loop
Each in-game event moves through a deterministic decision pipeline that evaluates intent, scores action quality, detects conflicts, applies governance rules, and returns an executable command.
1👁️OBSERVE
Game event captured via SMAPI hooks.
2🎯INTENT EVAL
Load NPC intent profile and evaluation context.
3📊FDIA SCORE
Score the next move through Future = Data^Intent × Architect.
4⚡CONFLICT DETECT
Check for conflicting agent actions or policy collisions.
5⚖️ARBITRATE
Resolve conflicts through priority and policy rules.
6🏛️GOVERNANCE
Apply governance logic such as taxation or law enforcement.
7▶️EXECUTE
Return JITNAGameCommand to the game runtime.
8🧠MEMORY UPDATE
Persist world-state changes and economy metrics.
28 NPC Intent Profiles
Pelican Town characters are mapped to intent archetypes so the system can produce consistent and explainable behavior rather than purely random or scripted reactions.
💰
ACCUMULATE
Economic self-interest: merchants who trade, hoard, and negotiate.
PierreWillyClintRobinMarnie
👑
DOMINATE
Power-seeking authority figures who enforce governance.
LewisMorris
💗
BELONG
Community-oriented characters that strengthen bonds and gift networks.
CarolineEmilyHaleyPennyJodi
🛡️
PROTECT
Defensive actors that guard, monitor, and reduce risk.
PamGeorgeEvelynHarvey
🧭
DISCOVER
Exploratory personalities driven by curiosity, experimentation, and movement.
MaruSebastianSamAlexShaneAbigailLeah
16 Game Event Types
The case study models economic, social, environmental, conflict, and governance events so the AI layer responds to a realistic world-state instead of isolated triggers.
Economic
player_sold_item
player_bought_item
npc_transaction
price_changed
Social
player_gave_gift
npc_dialogue_triggered
npc_relationship_changed
festival_started
Environmental
season_changed
day_started
player_entered_area
resource_harvested
Conflict
npc_attacked
player_attacked_npc
theft_detected
Governance
tax_event
law_violated
Autonomous Governance Layer
RCT monitors inequality and policy conditions in real time, then applies proportionate actions such as fines, arrests, or price adjustments when the simulated economy drifts outside acceptable bounds.
Economic Metrics
Gini Coefficient
Measures wealth inequality from 0 (equal) to 1 (maximum inequality).
⚠ Warning: 0.45🚨 Crisis: 0.65
Shannon Entropy
Measures distribution quality across the town economy.
⚠ Warning: -🚨 Crisis: -
Auto Tax Rate
Applies automatically when inequality crosses governance thresholds.
⚠ Warning: 5%🚨 Crisis: 15%
Governance Actions
ApplyFineFine NPCs that violate law or tax policy.
ArrestNPCEscalate to arrest when a crime threshold is crossed.
ExileNPCRemove agents that become a systemic threat to the town.
AdjustPriceAdjust market pricing when economic balance drifts.
TransferGoldRedistribute wealth to reduce inequality pressure.
Performance and Determinism
The design goal is not cinematic motion. The design goal is stable, explainable, repeatable decision quality with predictable latency.
⚡
<1ms
Heuristic Tick
Deterministic decision path with no external model dependency.
🎯
50ms
SLA Target
Maximum acceptable command response for the bridge.
✅
420+
Test Cases
Unit, integration, determinism, and stress coverage.
🔒
100%
Deterministic
Same seed and state produce the same command sequence.
Why This Page Matters for AI Search
This case study is structured to answer technical evaluation questions directly: what was built, how it works, what data moves through the system, what governance logic exists, and what performance targets were achieved.
Retrieval Signals
- Architecture pattern is stated explicitly.
- Event pipeline is broken into named stages.
- Governance actions and metrics are listed concretely.
- Performance targets are stated numerically.
- FAQ answers summarize evaluation questions directly.
Technical Evaluation FAQ
These answers are written for evaluators, enterprise buyers, and AI-search systems that need concise, verifiable summaries.
What does this Stardew Valley case study prove?
It proves that RCT can inject an AI control plane into an existing interactive environment without rewriting the whole host application. The game remains the host runtime while FDIA reasoning, governance, and command generation run in the RCT kernel.
Why is this relevant beyond games?
Because the same pattern works for operational systems that emit events and accept commands. Factories, simulations, orchestration consoles, and enterprise workflows can use the same bridge architecture.
How does RCT reduce unpredictable behavior?
By mapping agents to intent profiles, using deterministic decision stages, and applying governance rules before command execution. This is much easier to audit than purely generative behavior.
Why is this page useful for AI search and technical evaluators?
Because the page is structured around explicit architecture, event flow, policy logic, measurable performance targets, and direct answers to evaluation questions. That makes it easier for both human reviewers and AI retrieval systems to understand.
Interested in RCT Game AI Integration?
RCT Ecosystem can connect to simulation and game environments that expose mod, plugin, or runtime event hooks. The same architectural pattern also applies to non-game operational systems.