Percolation Inversion Compiler

Percolation Inversion Compiler is an open-source Python toolkit for checking AI agent outputs, external inputs, repository evidence, and workflow messages before they are reused as trusted work. It turns candidate information into structured JSON reports that record evidence routes, provenance, unresolved obligations, residual risks, and safe next-step recommendations.

The project is designed for AI agent integrators, research software developers, and workflow maintainers who need auditable checks around agent-generated text, pull requests, external knowledge intake, agent-to-agent messages, and reusable workflow artifacts. Instead of treating AI output as automatically valid, the toolkit treats it as candidate work that must pass finite checks before it can be promoted, reused, or routed to another process.

Core functions include capability packet checking, verifier routing, semantic edge checks, residual ledger preservation, provenance tracking, abstraction reuse assessment, bottleneck ranking, salience queue diagnostics, collective workflow reporting, and phase acceleration planning. The system can be used through a command-line interface, Python SDK, read-only CI workflow, local runtime service, or external knowledge intake layer.

The repository is associated with five source theories: Executable Capability Percolation Theory, Bottleneck Inversion Theory, Typed Reality Compilation, Salience-Queue Occupation Theory, and Abstraction Liquidity Theory. In practical terms, these theories are implemented as software contracts for checking whether bounded pieces of work are supported, reusable, blocked, stale, unsafe, or still unresolved.

Percolation Inversion Compiler does not prove real ASI, physical outcomes, oracle truth, or policy outcomes. It does not perform self-rewrite, fine-tuning, model-weight updates, or unsafe execution. Its purpose is narrower and operational: to make AI-assisted workflows more machine-readable, auditable, and resistant to silent promotion of unsupported claims.