Huarizo AI - Autonomous Options Trading Agent

Huarizo AI is an autonomous options-only trading agent: a pipeline with hard gates, not a chatbot that suggests trades. PIPELINE live OCC chain -> contract rules (R1-R3, R5, R6) -> account gate -> idempotent reservation -> ledger -> one option order. Equity brackets fail closed (403). No LLM in the critical path. FIVE MANDATES Scout, Strategist, Pricer, Broker, Sentinel: one trader, five decision mandates, each able to veto the cycle. FIVE CONTRACT RULES R1 DTE 7-45, but ranked not just cut: 21+ preferred, 14-20 acceptable, <14 last resort. Without it, selection lands on 7 DTE: max theta, no time for the thesis. R2 rejects IV above 1.25x realised vol; abstains if RV is missing. R3 rejects theta burn above 3.0% of premium/day. R5 rejects the symbol when near/far IV exceeds 1.15 - an event is priced in. R6 caps concentration at one position per underlying, in-flight included. No R4: numbering grew organically and the logs cite it. ACCOUNT GATE 16 named refusals - daily-loss pause, duplicate in flight, gross exposure, position size, $5k notional cap, options buying power with a 10% buffer (a contract is 100 shares, so $3 is $300). Fail-closed: an unreadable snapshot is a rejection, never a pass. On the live book {SPY:1, QQQ:1, AAPL:1}, R6 refused a second SPY call: reason="underlying_concentration_limit". BORN FROM A REAL INCIDENT On 2026-08-31 an earlier build sent 180 orders in 18 seconds: idempotency was a derived key with no persisted reservation. Most guards here come from that night. WHAT IT REFUSES TO CLAIM Backtests are context and do NOT feed the confidence score. When validity conditions are unmet, win rate renders as "not measured", not a flattering number. Contract volume was 0 on all 3,648 SPY contracts on the free plan, so the UI says so. VERIFIABILITY 597 pytest tests pass; smoke_contest_flow.py checks the real account and MCP server.