MIZAN

Mizan is a governance layer for AI trading agents built around one principle: AI may recommend a trade, but it should never have unchecked authority to execute one. Research, stock-selection, probability, and trader agents can generate TradeProposals, but every proposal must pass through Mizan’s deterministic risk engine, AI advisory layer, Portfolio Governor, authorization system, and execution gate before it can reach Alpaca Paper Trading. Mizan evaluates position size, portfolio exposure, concentration, volatility, drawdown, liquidity, buying power, option structure, and other policy constraints. The Governor can APPROVE, REDUCE, or REJECT a proposal, and advisory AI is never allowed to override deterministic safety rules or increase an authorized quantity. Every important decision is recorded through an explainable audit trail with policy evidence, authorization state, decision provenance, hashes, and replayable records. The interface clearly separates agent intelligence from governance authority, showing the full Requested → Authorized → Executed lifecycle. The project also includes kill-switch controls, duplicate-order protection, deterministic replay, reconciliation, PAPER-only enforcement, security checks, invariant tests, and explicit failure states rather than fabricated market or broker data. Mizan turns autonomous trading from “an AI that can place orders” into a controlled multi-agent system where intelligence can move quickly while authority remains measurable, deterministic, and auditable.