Percolation Inversion Compiler

Percolation Inversion Compiler (PIC) v0.4.4 is an open-source Python runtime and command-line toolkit for checking, routing, and auditing AI-agent work. It helps agents and developers treat LLM output, repository evidence, external information, and agent-to-agent messages as candidate work rather than verified results. PIC organizes such work into structured reports with evidence routes, proof obligations, residual ledgers, provenance records, verifier tasks, packet-exchange records, and phase-planning diagnostics.

The project is designed for practical AI-agent workflows. It can be installed from PyPI and used for compact JSON checks, runtime reports, schema export, workflow verification, agent-output review, benchmark diagnostics, and local integration through a Python SDK or CLI. The v0.4.4 release adds optional sidecar tools for operator-facing adoption packets, agent-to-operator requests, protocol-relative benchmark reports, packet inspection and merge utilities, and phase-dashboard summaries. These sidecars support human review and system integration without blocking the main agent workflow.

PIC is intended to improve the traceability and reuse of AI-generated work. It distinguishes accepted candidates, workflow-usable outputs, unresolved obligations, settlement blockers, and fully settled claims. This makes it useful for AI safety research, LLM workflow engineering, autonomous-agent evaluation, evidence management, reproducible audit trails, and machine-readable research infrastructure.

The project does not prove real artificial superintelligence, physical truth, oracle truth, simulator truth, policy outcomes, or autonomous action safety. Its ASI-proxy language is protocol-relative: PIC supports observable workflow acceleration through bounded candidate exchange, residual preservation, verifier routing, reusable packet formation, and bottleneck planning, while keeping unsettled obligations explicit.