Mizan Constitution
Runtime governance for AI agent actions: Mandate, Council, Gate and Receipt.
PRODUCT MAP
One control stack. Clear product roles.
The public tree is organized by what the customer buys or uses, while internal engines remain engines. This avoids turning every experiment into a separate SKU.
Mizan Enterprise
Policy, shadow pilot, governance, teams and decision/audit operations in one workspace.
Mizan Rights / Hakkım
Facts, evidence gaps and official-route preparation for disputes. Not individualized legal advice.
CORE CONTROL ENGINES
Specialized controls beneath Constitution.
They can be exposed independently for pilots or integrations, but they belong to one architecture.
Mizan AgentGate / Mandate
Evaluates agent role, declared purpose, delegation scope and proposed tool actions before execution.
Mizan ActionGate
Policy, approval, evidence, authority and replay/duplicate checks for proposed business actions.
ProofGate / SourceAudit
Checks whether evidence or a cited source supports a claim. It does not certify that the source itself is true.
Human Seat / AlertRoute
Routes HUMAN_REVIEW cases by risk and operational priority. Routing cannot convert a hard VETO into ALLOW.
Receipt Ledger
Durable decision/action receipts and replay state across ephemeral runtime revisions.
Guardian Cloud
Operational surface for monitored Mizan cloud deployments and governance controls.
EVALUATION & LABS
Tools that strengthen the product without pretending to be separate flagship SKUs.
These surfaces support evaluation, policy quality, benchmarking and R&D.
Policy Stress
Inspect policy text for missing controls, contradictions and rollout risk.
Open tool ↗BenchmarkGate
Compare current system vs Mizan on calibration + untouched holdout with explicit false-safe / false-block reporting.
AutoBenchmark
Internal automation for repeatable evaluation runs and evidence packaging.
Cyber ActionGate
Security-focused policy/action evaluator feeding the broader ActionGate / Constitution control path.
Cognitive Gate
Behavioral feasibility research on how warnings and decision gates affect human choices.
FiFAR / OpenL2D research
R&D and benchmark evidence. Not positioned as a customer-facing product.