Release IX-Sally v0.7.0: Lifelong Generalization

[BryceWDesign]

Summary

This PR upgrades IX-Sally from v0.1.0 to v0.7.0 and introduces the full CUC-1 through CUC-6 experimental cognition progression.

The release focuses on bounded generative cognition, lifelong learning, representation invention, meta-learning, cross-domain transfer, semantic formation, open-goal generation, knowledge restructuring, and recursive cognitive bootstrapping while preserving the project authority boundary:

AI proposes. Humans decide.

Major additions

• Generative open-choice construction beyond fixed action menus
• Hypothesis and primitive invention
• Bounded instrumental goal generation
• Semantic genesis from externally validated latent structure
• Open-ended goal generation within governed search bounds
• Recursive cognitive bootstrapping
• Long-horizon replanning and counterfactual reasoning
• Tool construction and validation
• Unknown-unknown detection
• Self-directed curriculum mechanisms
• Lifelong knowledge persistence and restructuring
• Context-sensitive knowledge splitting and retirement
• Compositional representation-program invention
• Persistent online meta-learning
• Relational cross-domain transfer
• Raw text-to-outcome grounding
• Multi-goal portfolio reasoning
• Snapshot/restore of lifelong and meta-learning state
• CUC-1 through CUC-6 experiments and validation documentation

CUC-6 lifelong generalization result

The v0.7.0 experiment demonstrates a measurable transition from fixed learning behavior toward experience-dependent learning strategy selection.

Early episodes explore multiple strategy families. Later episodes use accumulated meta-learning evidence to select a learning strategy more selectively while maintaining held-out task performance.

The compositional representation challenge improves from a best shallow representation accuracy of 81.25% to a synthesized representation achieving 100% training and held-out validation accuracy in the bounded experiment.

Validation

Local validation completed successfully on Python 3.13.2:

• 1,036 tests collected
• Full pytest suite completed successfully with exit code 0
• Installed wheel package smoke test passed
• Repository integrity: 312 source/test files, 0 violations
• Runtime dependency graph: 161 modules, 843 imports, 0 cycles
• Runtime architecture: 161 modules, 843 imports, 0 boundary violations
• git diff --cached --check passed

Claim boundary

This release does not claim AGI, consciousness, free will, or autonomous authority.

The implemented mechanisms are bounded experimental systems intended to investigate AGI-relevant properties such as generative representation, adaptive learning strategy, knowledge restructuring, transfer, and recursively reusable cognition.

Consequential external authority remains governed by human authorization.

Capability is not authority. AI proposes. Humans decide.