Skip to content

Ztare

v1.2.0 Feature

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

✓ 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 LeanMill and AxiomPack, Verification, and Project Workbench across a mixed release.

Changes in this release

Security Medium

Scrub maintainer‑specific paths from public artifacts.

Scrub maintainer‑specific paths from public artifacts.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Feature Medium

Adds explicit public-project allowlist and one-origin production server for React app and Python API.

Adds explicit public-project allowlist and one-origin production server for React app and Python API.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Refuses unlisted project reads, file previews, and writes before dispatch.

Refuses unlisted project reads, file previews, and writes before dispatch.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Adds evidence-fetch receipts, plugin management, model briefs, saved decision visits, report readiness, project scenarios, and broader review/history flows.

Adds evidence-fetch receipts, plugin management, model briefs, saved decision visits, report readiness, project scenarios, and broader review/history flows.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Refreshes Verdict view and small-screen layouts, including the LeanMill workspace.

Refreshes Verdict view and small-screen layouts, including the LeanMill workspace.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Adds release runbook, CI boundary job, Docker build context, interaction smoke, and public release smoke.

Adds release runbook, CI boundary job, Docker build context, interaction smoke, and public release smoke.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Adds campaign closure, formal-task boundaries, external-science admission, language advancement, compound implication sieves, resumable VPS actions, and kernel-checked Lean ratification.

Adds campaign closure, formal-task boundaries, external-science admission, language advancement, compound implication sieves, resumable VPS actions, and kernel-checked Lean ratification.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Publishes campaign receipts and new AxiomPack Lean artifacts for differential-mode, finalist, orbit-action, witness, and reconstruction boundaries.

Publishes campaign receipts and new AxiomPack Lean artifacts for differential-mode, finalist, orbit-action, witness, and reconstruction boundaries.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Medium

Adds typed transition identity, observation charts, equivariance certificates, factored search, schema routes, compiled planning, and deterministic candidate production to ARC-AGI world-model harness.

Adds typed transition identity, observation charts, equivariance certificates, factored search, schema routes, compiled planning, and deterministic candidate production to ARC-AGI world-model harness.

Source: llm_adapter@2026-07-16

Confidence: high

Feature Low

Treats agent‑authored theory program and its lineage as governed campaign object; compact axiom packs remain one calibrated profile.

Treats agent‑authored theory program and its lineage as governed campaign object; compact axiom packs remain one calibrated profile.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Feature Low

Includes exact bounded order‑7 coprime‑cycle result and finite database audit over published orders 8–10 with finite‑scope and source‑completeness boundaries.

Includes exact bounded order‑7 coprime‑cycle result and finite database audit over published orders 8–10 with finite‑scope and source‑completeness boundaries.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Feature Low

Expands validator, orchestration, replay, and identity‑boundary coverage in ARC‑AGI world‑model harness.

Expands validator, orchestration, replay, and identity‑boundary coverage in ARC‑AGI world‑model harness.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Feature Low

Sets package and citation metadata to 1.2.0.

Sets package and citation metadata to 1.2.0.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Bugfix Low

Repairs changelog history for v1.0.0 and v1.1.0.

Repairs changelog history for v1.0.0 and v1.1.0.

Source: llm_adapter@2026-07-16

Confidence: high

Refactor Low

Consolidates superseded world-model adapters and duplicate planning/compiler surfaces into shared primitives.

Consolidates superseded world-model adapters and duplicate planning/compiler surfaces into shared primitives.

Source: llm_adapter@2026-07-16

Confidence: high

Refactor Low

Moves provider SDKs and research libraries into optional extras, keeping base install offline‑public path.

Moves provider SDKs and research libraries into optional extras, keeping base install offline‑public path.

Source: granite4.1:30b@2026-07-16-audit

Confidence: low

Full changelog

ZTARE v1.2.0 deepens the local Project Workbench, advances AxiomPack into governed theory-program campaigns, and publishes the current ARC-AGI/world-model harness as work in progress.

Project Workbench

  • Adds an explicit public-project allowlist and a one-origin production server for the React app and Python API.
  • Refuses unlisted project reads, file previews, and writes before dispatch.
  • Adds evidence-fetch receipts, plugin management, model briefs, saved decision visits, report readiness, project scenarios, and broader review/history flows.
  • Refreshes the Verdict view and small-screen layouts, including the LeanMill workspace.
  • Adds a release runbook, CI boundary job, Docker build context, interaction smoke, and public release smoke.

Run locally:

make forensic-workbench-live

Or build and serve the one-port release image:

docker compose --profile workbench up --build workbench

LeanMill and AxiomPack

  • Treats an agent-authored theory program and its lineage as the governed campaign object; compact axiom packs remain one calibrated profile.
  • Adds campaign closure, formal-task boundaries, external-science admission, language advancement, compound implication sieves, resumable VPS actions, and kernel-checked Lean ratification.
  • Publishes campaign receipts and new AxiomPack Lean artifacts for differential-mode, finalist, orbit-action, witness, and reconstruction boundaries.
  • Includes an exact bounded order-7 coprime-cycle result and a finite database audit over published orders 8–10. These artifacts retain their finite-scope and source-completeness boundaries.

ARC-AGI / world-model harness

ARC-AGI remains WIP in this release.

  • Adds typed transition identity, observation charts, equivariance certificates, factored search, schema routes, compiled planning, and deterministic candidate production.
  • Consolidates superseded world-model adapters and duplicate planning/compiler surfaces into shared primitives.
  • Expands validator, orchestration, replay, and identity-boundary coverage.

This release makes no ARC-AGI benchmark, solve-rate, or leaderboard claim.

Packaging and release discipline

  • Sets package and citation metadata to 1.2.0.
  • Moves provider SDKs and research libraries into optional extras so the base install retains the offline public path.
  • Repairs changelog history for v1.0.0 and v1.1.0.
  • Scrubs maintainer-specific paths from public artifacts and refreshes the public roadmap and claim register.

Verification

  • make gates
  • make docs-check — 99 Markdown files checked
  • make forensic-workbench-release-check — 13/13 boundary checks
  • Workbench interaction smoke — 21/21 actions
  • Canonical route render smoke — 22/22 routes
  • Production Vite build and Workbench Docker build
  • Final parser/LeanMill invariant suite — 140 passed

Boundaries

  • The Project Workbench remains local-first and filesystem-backed; this is not a hosted multi-user service.
  • AxiomPack campaign and Lean artifacts demonstrate governed execution and bounded formal results; they do not establish general theorem-discovery lift.
  • Local ARC environment data, raw review transcripts, caches, backups, and active PutnamBench scratch were not included.

Full changes: https://github.com/sparckix/ztare/compare/v1.1.0...v1.2.0

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track Ztare

Get notified when new releases ship.

Sign up free

About Ztare

All releases →

Related context

Beta — feedback welcome: [email protected]