Skip to content

AnimusLab/Canon

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Canon 🏛️

Deterministic governance knowledge integrity engine for AI systems.

Canon is a lightweight, tamper-evident governance synchronization engine designed to fetch, verify, diff, and audit external regulatory and organizational policies. It acts as the structured source-of-truth pipeline for AI constitutional guardrails.


Key Features

  • Source Adapters: Extensible adapters for crawling structured authorities (e.g., OWASP LLM Top 10, NIST AI Risk Management Framework, FINOS AI Governance).
  • Deterministic State Hashing: Cryptographically hashes governance rulesets to detect changes instantly and reliably.
  • Evidence Packages: Generates comprehensive diff packages detailing added, modified, deprecated, or removed rules, including severity deltas.
  • Tamper-Evident Ledger: Logs all transitions to an append-only, hash-chained ledger (ledger.jsonl) to ensure auditability.
  • Human-in-the-Loop Approvals: Approval gate that secures state transitions behind individually-attributed cryptographic approval records.
  • Microsecond Latency: Built for high-frequency runtime checks with execution overhead under a millisecond.

Installation & Setup

  1. Activate your environment and install dependencies:
    conda activate canon
    pip install -e .
  2. Configure your GitHub token (optional, but recommended to avoid rate limits):
    set GITHUB_TOKEN=ghp_your_token_here

CLI Usage

1. Fetch Latest Policies

Fetch state from all tracked governance sources, compute hashes, and compare them against local baselines.

canon fetch

2. Review Pending Decisions

Interactively inspect and approve/reject pending evidence packages.

canon review

3. Verify Ledger Integrity

Verify the tamper-evident signature chain of all recorded approval records.

canon verify

4. Check Governance Status

Display the active version, rule count, last fetch time, and state hashes of all tracked sources.

canon status

Empirical Performance (ICAIF Benchmarks)

Under benchmark conditions (compiled on Python 3.14), Canon operates at microsecond scales, presenting negligible overhead for enterprise deployment.

Operation Mean Latency Median Latency P95 Latency P99 Latency Throughput
Source State Hash (100 rules) 135.36 µs 129.00 µs 155.10 µs 253.40 µs 7,387.9 /s
Diff Engine (100 rules, 5 changes) 88.12 µs 83.30 µs 104.10 µs 176.00 µs 11,347.8 /s
Diff Engine (500 rules, 20 changes) 398.27 µs 366.70 µs 548.10 µs 834.10 µs 2,510.8 /s
Evidence Hash Generation 12.27 µs 11.20 µs 11.50 µs 18.70 µs 81,496.9 /s
Ledger Chain Hashing 1.34 µs 1.20 µs 1.20 µs 1.40 µs 744,990.0 /s
Approval Record Hashing 3.34 µs 3.30 µs 3.40 µs 3.70 µs 299,043.3 /s
End-to-End Pipeline (Diff + Hash + Ledger) 490.10 µs 450.05 µs 734.20 µs 1,112.60 µs 2,040.4 /s

Next Steps & Integrations

For a deep dive into how Canon integrates with the broader ecosystem, please read the Architecture Document.

About

Deterministic governance integrity engine for monitoring, versioning, and verifying regulatory and policy sources.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors

Languages