Skip to content

About

Formal Verification | @AdaptiveIntelligenceCircle

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

1 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

AIC-Formal

Formal models (TLA+ / Apalache) for Adaptive Intelligence Circle reference properties.

Status: pre-Covenant, experimental, under-claim.

These models are design aids and property checks for Parallel TestNet components.

They do not constitute a verified production system, a Covenant declaration, or any form of certification.

Purpose

AIC-Formal provides machine-checkable models of selected invariants that appear in the Ethical Kernel, governance evaluation, SSI operations, recovery paths, and parallel-net isolation.

The models exist so that:

  • contributors can state properties precisely,
  • regressions in design intent can be detected early,
  • the gap between “we intend fail-closed” and “we can check a small model of fail-closed” is reduced.

They are not a substitute for testing, code review, or operational judgment.

Scope (explicit limits)

In scope Out of scope
Small-state models of Allow / Deny / NeedHuman Full protocol verification
Fail-closed and human-gate properties Performance or liveness of real deployments
SSI bind / revoke / rotate invariants (simplified) Cryptographic correctness proofs
Recovery orchestration sketches Mainnet or Covenant claims
Parallel-net isolation (default-deny bridge) Token or economic models

Layout

AIC-Formal/
├── README.md
├── LICENSE
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md
├── SECURITY.md
├── docs/
│   ├── overview.md
│   ├── properties.md
│   ├── modeling-guide.md
│   └── limitations.md
├── specs/
│   ├── common/           # shared constants, types, helpers
│   ├── ethical_kernel/   # core evaluate decision
│   ├── governance/       # governance evaluate & succession sketches
│   ├── ssi/              # bind / revoke / rotate
│   ├── recovery/         # recovery orchestrator sketches
│   └── parallel_nets/    # isolation & default-deny bridge
├── tools/                # helper scripts (Apalache / TLC)
└── examples/             # small runnable configurations

Quick start

  1. Install TLA+ Toolbox or use command-line TLC / Apalache.
  2. Read docs/overview.md and docs/limitations.md.
  3. Open a module under specs/ and its corresponding .cfg.
  4. Run model checking on the small configurations provided in examples/.

Apalache is preferred for inductive invariants; TLC is sufficient for small exhaustive checks.

Relationship to other AIC repositories

  • AIC-TestNet / Parallel TestNet — these models abstract selected behaviors that the C++ reference nets implement.
  • AIC-Covenant — criteria and checklists may reference properties stated here; models do not declare Covenant readiness.
  • AIC-TransparencyDashboard — formal results (when produced) can be logged as audit events; no automatic pipeline is assumed.
  • AIC-Beginners / MyVision — philosophical and onboarding context; formal models stay technical and bounded.

Principles observed

  • Under-claim: models are partial, finite-state, and explicitly limited.
  • Fail-closed: default behavior in models is Deny when conditions are not met.
  • No token rule-power: no economic variables appear in any module.
  • Entity ≠ immunity: formal results do not grant any organization or person special status.
  • Pre-Covenant: nothing in this repository advances a mainnet or Covenant declaration.

License

GPL-3.0-or-later (see LICENSE).
Documentation may be dual-licensed under CC-BY-4.0 where noted.

Maintenance note

During periods of reduced maintainer availability the repository remains public.
Contributions that preserve under-claim language and do not expand scope beyond the documented limits are welcome.

About

Formal Verification | @AdaptiveIntelligenceCircle

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages