IVThe Engine

Cardinal Engine

Infrastructure for persistent, deterministic simulations. Not a game engine in the graphical sense.

Version

0.1

State

Active

Mode

Deterministic

Engine Properties

Deterministic

Same input, same output. Always. Variance comes from behavioral state accumulation, not randomness at resolution.

Persistent

All state in SQLite. Sessions end and resume. Long-running simulations page to disk rather than bloating memory.

Server-Controlled

Every game-critical calculation runs on Node.js backend. Client receives results only. Nothing consequential in browser.

Heuristic-Driven

Combat AI uses scored heuristic functions, not learned models. Predictable, debuggable, tunable without retraining.

Narrative-Separated

LLM module isolated. It receives sanitized event payloads and returns text. No simulation state writable from it.

Modular

Combat, economy, progression, and narrative are discrete modules. Cardinal is not Crownless-specific infrastructure.

Architecture

Four layers. Defined interfaces between them. Separation enforced at the module boundary.

API Layer

Next.js API Routes

Transport

Receives Arbiter actions, validates input, routes to game logic. Returns display state to client. Stateless.

Game Logic

Node.js modules

Core

Simulation loop, combat resolution, behavioral state management, progression. No I/O.

AI Services

Groq SDK (isolated)

Narrative

Called post-simulation only. Input: sanitized event data. Output: text string. One-way pipe.

Database

SQLite / better-sqlite3

Persistence

All game state. Synchronous reads. Writes batched at simulation boundaries, not per-step.

The AI Boundary

The Boundary Rule

The AI services module receives a sanitized event payload after simulation completes. It returns a string. That string becomes a narrative record. The LLM has no access to live simulation state or Stray behavioral values.

Enforcement is architectural. The AI module does not import from game logic. The module boundary forbids it. Events in, text out, nothing writes back.

Data Flow

Engine→Event payload

Groq LLM→Text string

Database→Stored narrative

Game Logic×AI module (blocked)

Data Model

Core SQLite Tables

strays         (id, arbiter_id, name, role, stats_json, behavioral_state_json, status)
bastions       (id, arbiter_id, facilities_json, resource_reserves, capacity)
fractures      (id, type, depth, threat_composition_json, modifiers_json, status)
missions       (id, fracture_id, stray_ids_json, outcome_json, log_json, completed_at)
narrative_log  (id, mission_id, stray_id, event_type, content, generated_at)
arbiters       (id, name, authority_rank, progression_json, created_at)