Principles¶

G6 uses six design principles as load-bearing constraints. They explain why every component is a harness, why every result carries evidence, and why uncertainty is surfaced instead of hidden.

The practical goal is agent self-training, harness engineering, and agent reliability: a bridge from today's AI systems to economically useful AI-enhanced work that is safe, compliant, and quality assured.

Principles¶

• P1: Every component is a harness
• P2: Each harness is stateful
• P3: Each harness mitigates a specific LLM failure mode
• P4: Each harness has its own failure modes
• P5: T3 evolves capability from examples and experiences
• P6: Each MCP is extensible with new skills by T3

Contract Gates¶

• C1: Block contracts are mandatory
• C2: Mitigation declarations are mandatory
• C3: State surfaces are declared
• C4: Extension points are explicit
• C5: Verification is harness-owned

Reliability Surfaces¶

• GUI, TUI/CLI, REST, and MCP reliability surfaces