Ztare

v0.1.0 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 1mo AI Agents & Assistants

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

adversarial-ml ai-alignment ai-governance ai-safety artificial-intelligence autoresearch

+8 more

formal-verification lean4 llm-evaluation multi-agent-systems neurosymbolic recursive-ai trustworthy-ai zero-trust

Summary

AI summary

Updates Highlights, lift-pending, and https://github.com/sparckix/ztare/blob/v0.1.0/CHANGELOG.md across a mixed release.

Full changelog

First tagged public release of ZTARE — a zero-trust research engine that keeps neural proposers strictly separate from deterministic verifiers, gates, and append-only ledgers, so the model cannot self-certify. The discipline is the product: it improves research rigor; it does not guarantee truth.

Highlights

leanmill — governed Lean proof-search engine: a governed DAG search over a typed move algebra behind a single anti-laundering kernel (axiom allowlist + statement-integrity + MNC) with composite decomposition→closure ratification. The solver proposes; governance ratifies. Environment lift is measured, not assumed (lift-pending).
Autoresearch law-discovery factory (discover → compress → prove) with the symbolic fitter, holdout tests, and Lean kernel as the only arbiters.
Shared common/ libraries: sandboxed execution, symbolic-witness / SymPy compute, constraint isomorphism, and more — reused across both lanes.
Primitive registry + atlas with an amnesia precheck to reuse capabilities before building new ones.
Deterministic publish-safety and anti-gaming gates, each dogfooded so it cannot silently rot.
Public analytics dashboard (GitHub Pages), leak-masked.