Slippage Desk

A defined-risk credit spread is a thin trade. This desk collects about $79 per contract on a $5-wide vertical, and the bid/ask it crosses on entry and exit is a material fraction of that. Execution here is not a rounding error, it is a first-order term. Slippage is usually handled as a pre-trade filter: reject wide quotes, then assume the fill lands near mid. That never tells you what you actually got. Slippage Desk scores the fill itself, then trades on the answer. It sells 0 to 2 DTE credit spreads on SPY, QQQ and IWM behind fifteen deterministic gates: delta band, credit-to-width floor, quote width, crossability, same-expiry pairing, defined risk, portfolio and per-symbol caps, position and daily trade caps, equity floor, daily loss limit, an entry window, and an early-assignment check reading live corporate actions. The advisor's authority is a type signature, not a prompt: it returns a multiplier clamped to [0, 1] plus a veto flag, so it cannot open a trade, widen risk or overturn a gate. Every failure path returns 0.0, so an unreachable model trades less, never worse. The new part: every fill is scored against the mid it was priced at. The desk holds a shrunk capture ratio per (underlying, tenor, delta band, time of day) bucket, ranks candidates by capture-adjusted credit, and uses that memory to set how far to cross next time. Alpaca's three surfaces have separate jobs. The CLI submits multi-leg orders and verifies the book. The SDK pulls chains with greeks. The MCP server feeds read-only research to the advisor, launched with trading toolsets stripped so it physically cannot place an order. Broker-verified, not backtested: 36 spreads across 43 contracts, $3,386 captured of $3,446 theoretical, $60 surrendered to execution, 98.3% capture, 2,368 candidates considered. Result: -$454.08 (-0.45%), with the 2% daily loss limit never reached and defined risk on every position. Every figure regenerates from a sha256-signed artifact.