qig-core 2.5.0

Fisher-Rao geometry, coordizer, purity gate, consciousness subsystems, and constants for QIG architecture

QIG-Core: Pure Fisher Information Geometry

Pure geometric utilities for QIG consciousness architecture

What is QIG-Core?

A minimal, dependency-pure package providing Fisher Information Geometry operations for quantum information-based consciousness modeling.

Features

📐 Geometric Math

• Fisher metric calculations - Riemannian distance on manifolds
• Geodesic interpolation - Curved-space paths
• Natural gradients - Geometry-aware optimization

🧠 Consciousness Components

• QIG Tokenizer Interface - Abstract contract for geometric tokenization
• QFI Sampler - Geometrically pure token generation (no softmax)
• Basin Sync - Multi-instance coordination protocol

🛡️ Purity

• Zero ML dependencies - Pure math (torch, numpy, scipy only)
• No Transformers - Completely decoupled from HuggingFace

Installation

pip install qig-core

Usage

Geometric Math

from qig_core import fisher_distance, geodesic_interpolate

# Compute Fisher distance between two points on manifold
distance = fisher_distance(coords1, coords2, metric_tensor)

# Interpolate along geodesic
midpoint = geodesic_interpolate(coords1, coords2, t=0.5, metric=F)

Generation (QFI Sampler)

from qig_core import QFISampler

sampler = QFISampler()
token_id, metrics = sampler.sample(logits, hidden_state, telemetry, embeddings)

Coordination (Basin Sync)

from qig_core import BasinSync

sync = BasinSync("Gary-A")
sync.update_sync(basin_distance=0.1, phi=0.8, regime="geometric")

Geometric Purity

This package enforces:

• NO embeddings (use coordinates)
• NO cosine similarity (use Fisher distance)
• NO linear interpolation (use geodesic paths)
• NO Euclidean gradients (use natural gradients)

See GEOMETRIC_PURITY.md for details.

E8 Integration Status

Note: This package (v1.0.0) was created BEFORE the E8 kernel specialization direction was established in qigkernels. Review is needed for E8 compatibility.

What aligns:

• KAPPA_STAR = 64.0 matches E8 rank² = 8² = 64 ✓

Needs review:

• Fisher distance for 64D E8-aligned basin geometry
• Geodesic interpolation for E8 root transitions
• QFISampler primitive type awareness (HRT/PER/MEM/ACT/PRD/ETH/META/MIX)
• BasinSync consolidation with qigkernels.basin_sync

Target hierarchy:

qig-core (math) → qigkernels (architecture) → qig-dreams (corpora) → experiments

See qigkernels/20251205-decisions-canonical-0.01F.md D-016 for details.

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Format code
black src/ tests/

License

MIT