Skip to content

Repository files navigation

Fail-Closed AI

Research Surface Map: lalaSkye.github.io/inspection-surface — full index, provenance, and cross-links

Public disclosure boundary

This repository is a public inspection surface, not full architecture disclosure.

It shows a bounded claim, public language for that claim, and the claim limit.

See PUBLIC_DISCLOSURE_BOUNDARY.md.

What this repo is

A documentation-first public proof surface for one governance pattern:

Invalid consequence-producing paths should fail closed rather than silently proceed.

The core question is:

Where does the system physically stop?

Minimal proof shape

A useful public proof surface should show:

  1. what action was attempted
  2. what proof was required
  3. which check failed
  4. whether consequence occurred
  5. what receipt was written

What this does not prove

This repository does not prove adoption, certification, standardisation, production readiness, runtime enforcement, path-universal deployment coverage, or complete AI governance.

It documents a bounded pattern and points readers to public proof objects only where each object states its own claim boundary.

Design principles

  • Stop is a first-class primitive.
  • Authority is checked before mutation.
  • Ambiguity fails closed.
  • Receipts are written for refusal, not just success.
  • Claims must be no larger than the evidence can carry.

Boundary rule

This repository must not be read as an architecture map, orchestration model, runtime substrate, deployment model, or protected system design.

It is a public claim-surface and language anchor only.

Status

v0.1 — documentation-first public surface.

Licence

MIT.

About

Public artefacts for fail-closed AI governance, refusal receipts, and runtime consequence control.

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors