Project

Corrigibility

A framework for systems that can be corrected without being captured. One of several lines of independent work — not the whole job.

The framework is two companion papers and a versioned schema repository. Paper I (DPI, Rule of the Code) develops the invariant topology — five structural tests rooted in control theory and commons governance. Paper II (EPI, Rule of the Workflow) extends the same logic to learned and agentic systems and surfaces the failure modes specific to them: opacity, compute capture, governance denial of service.

The body of the framework — schemas, rules, measures — lives in the schema repository, versioned v1.7-spec. The papers stay focused on argument and proof.

The two papers

  • Paper I — DPI: Rule of the Code Five structural tests; ~50 pp.; SSRN DOI 10.2139/ssrn.6059075. v3.0, 2026
  • Paper II — EPI: Rule of the Workflow Learned and agentic public infrastructure; ~32 pp.; Zenodo DOI 10.5281/zenodo.19863649. v3.0, 2026

Read the long-form pages and download the PDFs →

Why this exists

Population-scale software runs on infrastructure operated by parties with their own incentives. Most governance debates argue about who should hold authority. The framework asks the prior question: what structural properties must hold for any holder of authority to remain correctable? Five of them, the framework argues, and they are testable.

Discussion

Recent essays on Layer 8 work through the framework's pieces in plainer language — start with thelayer8.substack.com. Citations, peer feedback, joint work: research@anivar.net.