alphadecay: AI options agent on Alpaca paper

alphadecay turns an options thesis into a reviewable protocol before an order can exist. The opening reason, structure, option legs, schedule, and risk rules stay frozen beside every later decision. A bounded model classifies only supplied evidence. It cannot choose sources, size a position, approve an order, or call the broker. On September 3, fixed policy approved one SPY call debit spread in the judged Alpaca paper account. The product submitted the multileg limit order at 10:45:02 AM ET, and the paper broker recorded a complete simulated fill. The spread expires October 9, 2026; quantity was one and debit paid was $175. Whole account reconciliation certified the entry and opened one managed position. A read only Alpaca check at render time supplies current equity, market value, unrealized P&L, DTE, delta, and theta without exposing strikes or identifiers. The full record includes every scheduled cycle, not only the fill. Since the fill, each scheduled check authorized no action, and the position stays open under its plan. The mandatory close is due September 4 at 9:45 AM ET. That is stated as intent; the record will show what happened. This is the safety behavior: missing facts mean no action instead of a guess. Alpaca's Trading API supplies paper account state, the spread order, fills, positions, and reconciliation. The official MCP server supplies read only research selected by the application. The CLI checks the paper endpoint outside the app. Public Replay is separate sample data and never represents a broker fill. The result so far is one paper fill authored by the product and an open managed position. A close and realized P&L are not yet part of the record. Paper fills are simulated. One paper outcome does not establish a repeatable edge, live performance, or lasting returns.