Skip to main content

NPC contradiction detector and belief repair for game worlds

Project description

worldoracle

worldoracle

CI PyPI Python Versions Downloads License: MIT codecov Typed

NPC contradiction detector and belief repair for game worlds.

Quick Start · How It Works · Features · CLI Reference · MCP · OpenAI Tools · Alternatives


Why

Game NPCs frequently end up with contradictory world models: the blacksmith "knows" the king is both alive and dead, the guard believes the bridge is passable while the quest log says it collapsed. These inconsistencies break immersion and cause dialogue bugs.

worldoracle gives every NPC a typed belief store with automatic contradiction detection and principled repair strategies — so your world stays consistent even when multiple event sources update the same facts.


How It Works

flowchart LR
    A[Game event / quest trigger] --> B[WorldPredicate]
    B --> C[WorldOracleStore]
    C --> D{ContradictionDetector}
    D -- contradictions found --> E[BeliefRepairer]
    E --> F[RepairFrame]
    F --> G[Updated BeliefState]
    D -- no contradictions --> G
  1. WorldPredicate — a typed belief: subject, attribute, value, source, confidence, timestamp. Content-addressed by SHA-256 of subject|attribute|str(value).
  2. BeliefState — an NPC's full belief set; also content-addressed.
  3. ContradictionDetector — scans for predicates with the same (subject, attribute) but different values.
  4. BeliefRepairer — resolves each contradiction using strategies: prefer_newer, prefer_higher_confidence, prefer_observation.

Features

Feature Status
Content-addressed predicates (SHA-256)
SQLite persistence (WorldOracleStore)
Contradiction detection
Belief repair (3 strategies)
Rich CLI (7 subcommands)
FastAPI REST server
MCP server for Claude Desktop
OpenAI function-calling tools
93 tests, >98% coverage
Fully typed (py.typed)

Quick Start

pip install worldoracle
from worldoracle import WorldPredicate, BeliefState, ContradictionDetector, BeliefRepairer

# Build a belief state
state = BeliefState(npc_id="guard-1")
state.add(WorldPredicate(subject="king", attribute="alive", value=True, source="quest-giver", confidence=0.8, timestamp=1.0))
state.add(WorldPredicate(subject="king", attribute="alive", value=False, source="observation", confidence=1.0, timestamp=2.0))

# Detect contradictions
detector = ContradictionDetector()
pairs = detector.detect(state)
print(f"Found {len(pairs)} contradiction(s)")

# Repair — strategies are applied automatically in priority order:
# prefer_newer → prefer_higher_confidence → prefer_observation
repairer = BeliefRepairer()
for a, b in pairs:
    frame = repairer.repair(a, b)  # repair(pred_a, pred_b) → RepairFrame
    print(f"Resolved: {frame.resolved_value!r} ({frame.strategy})")

CLI Reference

worldoracle [--db PATH] COMMAND [ARGS]
Command Description
add NPC_ID SUBJECT ATTRIBUTE VALUE Add a predicate to an NPC's belief state
check NPC_ID Detect contradictions
repair NPC_ID Generate repair frames for all contradictions
beliefs NPC_ID List all beliefs for an NPC
consistency Run full consistency check across all NPCs
diff NPC_ID TIMESTAMP_A TIMESTAMP_B Diff belief state at two points in time
status Show database stats
# Add beliefs
worldoracle add guard-1 king alive True --source observation --confidence 0.9 --timestamp 100
worldoracle add guard-1 king alive False --source rumor --confidence 0.5 --timestamp 50

# Check for contradictions
worldoracle check guard-1
# Found 1 contradiction(s) for guard-1:
#   CONFLICT: king.alive: 'True' vs 'False'

# Repair
worldoracle repair guard-1

REST Server

Install the API extra and start the server:

pip install 'worldoracle[api]'
uvicorn worldoracle.api:app --reload

The OpenAPI docs are available at http://localhost:8000/docs. See openapi.yaml for the full schema.


Repo Tree

worldoracle/
├── src/worldoracle/     ← Python package
│   ├── predicate.py     ← WorldPredicate, BeliefState, Detector, Repairer
│   ├── store.py         ← SQLite persistence
│   ├── cli.py           ← Click CLI
│   ├── api.py           ← FastAPI server
│   ├── mcp_server.py    ← MCP server
│   ├── report.py        ← Rich + JSON + Markdown formatters
│   └── py.typed
├── tests/               ← 45+ tests
├── docs/
├── tools/openai-tools.json
└── openapi.yaml

MCP / Claude Desktop

Install the MCP server:

pip install "worldoracle[mcp]"

Add to ~/.config/claude/claude_desktop_config.json:

{
  "mcpServers": {
    "worldoracle": {
      "command": "worldoracle-mcp"
    }
  }
}

Tools exposed: add_predicate, check_beliefs, repair_contradictions.

See docs/mcp.md and Smithery for hosted registry.


OpenAI Tools

The tools/openai-tools.json file defines function-calling schemas for GPT-4o and Codex CLI:

cat tools/openai-tools.json

See docs/openai.md for integration examples.


Alternatives

Tool Approach worldoracle advantage
Manual quest flags Unstructured booleans Typed, content-addressed, auditable
Event sourcing logs Append-only, no repair Built-in contradiction detection + repair
Prolog / logic engines Heavyweight runtime Zero-dep Python, SQLite storage
LLM world models Probabilistic, opaque Deterministic, inspectable, fast

Topics

This project is tagged: #npc #game-ai #belief-revision #llm #agents #mcp #fastapi

GitHub Topics: npc, belief-revision, game-ai, contradiction, agents, mcp, llmops


Case Studies

See how teams are using worldoracle in production:


Stay Updated

Subscribe to The Silence Layer — weekly dispatches on production AI infrastructure, new releases, and the failure modes that production AI systems don't surface until it's too late.

Star History

Star History Chart

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

worldoracle-0.1.0.tar.gz (1.5 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

worldoracle-0.1.0-py3-none-any.whl (21.2 kB view details)

Uploaded Python 3

File details

Details for the file worldoracle-0.1.0.tar.gz.

File metadata

  • Download URL: worldoracle-0.1.0.tar.gz
  • Upload date:
  • Size: 1.5 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for worldoracle-0.1.0.tar.gz
Algorithm Hash digest
SHA256 c3e2f2c88a4cd068a51af3e30b06f419fd761c0f5c852f99f700a498a1838d45
MD5 c5e667df5e5caf32aa739825c2b68b19
BLAKE2b-256 8403d5a3a80592bff9441df91c0ce890b638ef7e006b70bf4891782c7f8f7c16

See more details on using hashes here.

File details

Details for the file worldoracle-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: worldoracle-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 21.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for worldoracle-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 0d6c3bdc6cd9114ede75039d89e169ecb3cb60f8b5b8b623b601e56b57e7fcfd
MD5 94efb2b9e25023955ccdd8b71ec1ec16
BLAKE2b-256 a51e5459438e066dae014023f90101b7fd28780abd4892c1a4237fba724be90e

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page