Derivative is a computational reasoning substrate, and Forge is its software-building pipeline. This system transforms natural-language requirements into executable Python software. It runs the generated code in isolation, validates it independently, and only packages builds that pass all validation gates. The process involves turning a requirement into a typed contract, then a grounded plan, followed by code generation, isolated execution, and independent validation, ultimately leading to a verified package or explicit failure evidence.
Automated Software Generation: Builds software from natural-language requirements.
Isolated Execution: Runs generated code in a secure, isolated environment (Docker sandbox) to ensure safety and reproducibility.
Independent Validation: Implements multiple layers of validation to ensure the generated code meets specified contracts and quality standards.
Fail-Closed Architecture: Designed to fail securely, preventing the packaging of unverified or invalid builds.
Traceability: Maintains traceability of requirements throughout the build process, from source text to final artifact.
Quality Contracts: Enforces executable obligations for security, persistence, rate limiting, auditability, and observability.
Bounded Repair: Implements a mechanism for targeted repair of failures, requiring material artifact changes before revalidation.
Extensibility: Supports various reasoning capabilities through optional installation profiles, including symbolic, probabilistic, causal, quantum, and physical reasoning.
Evidence Generation: Produces typed evidence for each stage of the build process, including build specifications, plans, code artifacts, and validation results.
The system is built with Python and leverages Docker for its sandboxing capabilities. It is designed for developers seeking a robust and trustworthy way to automate software creation and validation.
Built with