Probability of Profit - POP Alpha Desk

POP Alpha Agent is an unattended paper options desk for the Alpaca × lablab Options Alpha Agents hackathon. The model is allowed to have opinions; it is not allowed to touch the broker. An optional LLM returns a small thesis JSON—underlying, structure, bias, and one sentence. No quantity, no limit, no OCC symbols. If that JSON is missing or nonsense, the scan continues: the governor does not need the model to work. Everything that matters lives in a deterministic governor we wrote for the window. Each loop builds a liquid tape from Alpaca’s most-active names and movers, then enumerates defined-risk templates: bull and bear call/put verticals, plus iron condors and butterflies early in the week. Candidates are scored with a Monte Carlo hold map aimed at the contest mark—Friday 4 Sep 9:30 a.m. ET—not at far-away expiration. Probability of profit, mean mark P&L, and the chance of reaching 25% of max profit by that manage-by day are hard gates. Packages that clear are ranked; the winner becomes a proposal with qty, join-NBBO limit, and an MCP-shaped multi-leg payload. Risk is fail-closed and paper-only. Size is floor(1% of equity ÷ |max loss|), with an unattended lot cap so a bug cannot fire a forty-five lot. Concurrent book risk steps up through the week with matching equity floors. One package per name. No new risk in the last fifteen minutes of the cash session, and none on Friday. Opens and closes join NBBO as DAY orders; take profit and stop sit at 50% of defined max profit or risk. The Alpha Desk is GET-only—it never places. Orders leave through the same shape as Alpaca MCP place_option_order, either pasted by a human or posted when LOOP_SEND=true. Live keys are never constructed. Official book: fresh $100k paper account PA3K9GXACTM7, judged on total equity. What stands out is the separation of concerns: creative AI optional, hard risk always on.