Skip to content

HANDOVER.md

Latest commit

 

History

History
90 lines (64 loc) · 3.26 KB

HANDOVER.md

File metadata and controls

90 lines (64 loc) · 3.26 KB

Handover Notes

What We Built

A system to model a person's worldview as an adversarial equilibrium of competing internal tendencies.

Key insight: Ideas are atomic, position is relational. The same observation means different things under different value lenses.

Documentation

See /docs for comprehensive documentation:

Quick Start

from world_model import create_world_model, Arena

# Load observations
model = create_world_model("Andrei", "observations.json")

# Run adversarial debate
arena = Arena()
trees, result = arena.run_full_debate(
    observations=model.observations,
    agents=model.agents,
)

print(f"Winner: {result.winner}")  # Which tendency's claim won

Current State

Implemented:

  • 7 human tendencies as competing agents
  • Adversarial debate: propose -> stake -> resolve
  • Weight propagation: net_score = direct + sum(pro) - sum(con)
  • ML-style training with convergence detection
  • Validation with statistical significance testing
  • JSON and Firestore persistence
  • FastAPI service

Validation Results (first run):

  • Accuracy: 27.3% (baseline: 7.1%)
  • P-value: 0.001 (highly significant)
  • Model predicts tendency ownership 4x better than chance

The Spark (Why This Matters)

This isn't just a data structure. It's a hypothesis about how meaning works.

Binary distinction (pro/con) is the primitive. Everything is this/not-this. Compose recursively -> arbitrary complexity.

The motivating error: First extraction said Andrei "distrusts DAOs." But he built Jurisdiction with fractal DAO topology. Wrong because no mechanism to contest the claim. Pro/con structure surfaces tension, forces refinement.

Adversarial dynamics: Agents don't just categorize - they COMPETE. Propose claims, stake evidence, win or lose. The competition IS the life of the system.

Agents as internal plurality: A person isn't monolithic. They're a coalition of drives finding equilibrium. The structure of that equilibrium IS the personality.

Worth building because it might be true, not just useful.

Key Files

life/
├── world_model/          # Core package
│   ├── models/           # Observation, Agent, Tree, Node
│   ├── dynamics/         # Arena (debates), Trainer (ML loop)
│   └── storage/          # JSON, Firestore persistence
├── api/                  # FastAPI service
├── docs/                 # Documentation
├── observations.json     # 165 observations about Andrei
├── test_arena.py         # Single debate test
└── test_training.py      # Full training test

The Debate Model Foundation

Extends tokenized debates from c:/code/werule_new:

  • Pro/con binary tree structure
  • Weight propagation formula
  • Arguments stake voting power

We extend it: multiple simultaneous trees, same observations appearing across them, internal tendencies as agents instead of external voters.