Epistemic Conflict Engine v3

Philosophical, dialectical reasoning with LLaMA 3.1 — built for researchers, not chatbots.

ECE v3 retrieves contradictory sources (thesis vs. antithesis), measures their ideological distance, and forces LLaMA 3.1 to synthesize a grounded answer instead of averaging opinions. It works locally with Ollama, stores knowledge in Neo4j, and connects directly to your own Zotero library.

One-click Windows launcher included. No API keys are stored in the repo.

---

Why this is different

• Dialectical Attention: scores pairs by (relevance + contradiction + balance) × temporal decay, not just cosine similarity
• Human-in-the-loop: pauses for structured material grounding before synthesis
• Private by design: your Zotero API key lives only in your local .env
• Reproducible: LangGraph checkpointing (SQLite/Postgres), full audit trail

---

Quick Start (Windows, 3 minutes)

1. Download the latest release and unzip
2. Copy .env.example → .env
3. Right-click start-local-ece.ps1 → "Run with PowerShell"

The script will:

• create a Python venv
• install dependencies
• pull llama3.1 and nomic-embed-text via Ollama
• start Neo4j locally
• load demo corpus
• open interactive menu

Then choose:

1 → start → topic: "Prison reform" → year: 1975 → thread-id: demo1

---

First-Time Setup

Requirements

• Windows 10/11, PowerShell 5.1+
• Python 3.11+ (from python.org, check "Add to PATH")
• 8GB RAM minimum

Connect your Zotero (optional but recommended)

Each user connects their own library. Nothing is uploaded.

python setup_zotero.py

You’ll be asked for:

• ZOTERO_LIBRARY_ID – find at https://www.zotero.org/settings/keys (numeric User ID)
• ZOTERO_API_KEY – create a private key on the same page
• Or set ZOTERO_LOCAL=true to use Zotero 7 desktop (no key needed)

Test:

python zotero_connector.py list-collections
python zotero_connector.py preview --tag philosophy --limit 5

Import to Neo4j:

python zotero_connector.py import --topic "Heidegger" --tag philosophy --default-class critical --epoch-start 1920 --epoch-end 1976

---

Usage

Single-file mode (recommended)

python ECE_v3_ALL_IN_ONE.py start --topic "Welfare administration" --year 1975 --thread-id welfare-1975
# ... system pauses for grounding ...
python ECE_v3_ALL_IN_ONE.py resume --thread-id welfare-1975 --grounding-file examples/material_grounding.example.json

With your Zotero data

1. Import: python zotero_connector.py import --topic "YourTopic" --collection ABC123
2. Run ECE as above – it will retrieve from your imported nodes

---

Privacy & Security

• No credentials in code. zotero_connector.py reads only from environment variables
• .env is in .gitignore – never commit it
• ZOTERO_LOCAL mode uses local Zotero database, zero cloud traffic
• All inference runs locally via Ollama

---

Project Structure

ECE_v3_ALL_IN_ONE.py      # full engine (retrieval + workflow)
zotero_connector.py       # Zotero → Neo4j importer
setup_zotero.py           # interactive credential setup
start-local-ece.ps1       # one-click launcher
requirements.txt
.env.example
examples/material_grounding.example.json

---

License

MIT – use freely, cite if you publish research.

Built for critical social science, philosophy, and anyone tired of consensus-biased LLMs.