quantum-context 0.2.0

Observer-relative knowledge graph using wave function compression

Quantum Context

Observer-relative knowledge graph using wave function compression for AI instance continuity.

What This Solves

Problem: Claude instances don't share context across sessions. Each new conversation starts from scratch.

Solution: Store measurements (subject, predicate, object, confidence) from different observers in a shared graph. Wave function compression preserves causal relationships while allowing multiple perspectives.

Install

Easy way:

git clone https://github.com/Unity-Environmental-University/quantum-context.git
cd quantum-context
./install.sh

Manual way:

git clone https://github.com/Unity-Environmental-University/quantum-context.git
cd quantum-context
pip install -e ".[mcp]"  # Includes MCP server support

For Claude Desktop: After install, add MCP server (see MCP_SETUP.md)

Quick Start

from quantum_context import observe_context, analyze_dependencies, act_record

# Record what you built this session (requires confirmation)
act_record(
    "skill-starter-template",
    "status",
    "complete",
    confidence=0.7,
    observer="claude-session-2026-02-05",
    confirm=True
)

# Next session: observe what was built
amplitude = observe_context("skill-starter-template")
# → magnitude: 0.7 (reasonably confident it's done)

deps = analyze_dependencies("skill-starter-template")
# → depends_on: ["complete", "friction-gradient-philosophy"]

Key Features

Epistemic Humility

• Confidence ceiling: 0.7 without evidence (can't claim >70% certainty from single observation)
• Evidence required to exceed 0.7 (provide URLs/citations)
• Default: 0.5 (moderate uncertainty)

# Capped at 0.7
act_record("concept", "is", "true", confidence=0.95, confirm=True)
# → Actually stored as 0.7

# Evidence allows exceeding ceiling
act_record("concept", "is", "proven", confidence=0.95,
           evidence=["https://paper.pdf"], confirm=True)
# → Stored as 0.95

Friction Gradient

• observe: No friction (read-only, always safe)
• analyze: Medium friction (computation cost)
• act: High friction (requires confirm=True)

Observer Attribution

Every measurement has an observer. Different observers can see the same concept differently (like your dialectical_mcp bias detection, but with wave functions).

Philosophy

The universe is a holographic projection from ℤ where any integer can be the origin.

Practical implications:

• Measurements = observations with confidence from an observer frame
• Wave functions = compressed representation (Shor-equivalent)
• Interference = relationships via quantum amplitude |ψ(A)·ψ(B)|²
• Divisibility = causality (if A divides B, A "causes" B)

This connects to:

• Shor's algorithm - prime factorization via wave interference
• Wolfram's Ruliad - space of all possible computations
• Causal set theory - time's arrow from observer entanglement

Storage

~/.quantum-context/graph.ndjson

Each line is a JSON measurement:

{"subject": "auth", "predicate": "requires", "object": "identity",
 "confidence": 0.7, "observer": "claude", "timestamp": "2026-02-05T12:00:00",
 "evidence": []}

Portable:

• Plain text (NDJSON)
• Git-friendly (line-based diffs)
• No database required
• Works anywhere Python 3.9+ runs

Use Cases

✅ AI continuity - Claude instances share context across sessions ✅ Multi-agent coordination - Different agents record different observations ✅ Learning analytics - Track how understanding evolves with confidence ✅ Bias detection - Compare observer frames (like dialectical_mcp) ✅ Research - Test if wave interference preserves causal structure

❌ NOT for legal evidence - No chain of custody ❌ NOT for medical/financial - Confidence levels too low ❌ NOT real-time - Wave fitting takes time

What's Implemented vs Stubbed

Working now:

• ✅ NDJSON storage
• ✅ Record measurements with confirmation
• ✅ Observe context (read measurements)
• ✅ Analyze dependencies (find related concepts)
• ✅ Confidence ceiling enforcement
• ✅ Evidence-based confidence boost
• ✅ Observer attribution

Stubbed (future):

• ⏳ Wave function fitting (currently just averages confidence)
• ⏳ Interference calculations (|ψ(A)·ψ(B)|²)
• ⏳ Frame transformations (different observer perspectives)
• ⏳ Confidence decay over time
• ⏳ Orthogonality detection (independent concepts)

Files

• quantum_context/core.py - Implementation (~200 lines)
• quantum_context/models.py - Pydantic types with extra="allow"
• skill.md - How Claude uses this skill
• RESPONSIBLE_USE.md - Ethics and safety guidelines
• README.md - This file

Example: Self-Modeling

A Claude instance can model itself:

act_record("claude-instance-2026-02-05", "can", "write-code",
           confidence=0.7, confirm=True)
act_record("claude-instance-2026-02-05", "uncertain-about",
           "shor-equivalence", confidence=0.7, confirm=True)
act_record("claude-instance-2026-02-05", "tends-to",
           "over-engineer", confidence=0.6, confirm=True)

Future instances can read this and know the previous instance's capabilities, uncertainties, and biases.

Status

Alpha (0.1.0) - Minimal working implementation. Wave function math is stubbed. Use for experimentation only.

License

MIT - See LICENSE file

Authors

Hallie Larsson, Unity Environmental University With Claude Sonnet 4.5

---

Measure. Compress. Interfere. Understand.

The structure teaches. The friction creates meaning. The waves preserve truth.