Alpaca AI Trading Agents Hackathon·paper account PA343VC6LL3T

Slippage Desk

The options agent that measures its own execution.

Why that matters. Every figure below is derived from the config and the 43 contracts actually filled:

a win banks
+$31.50
vs
its stop costs
$78.74
breakeven win rate
71.4%
vs
delta-implied OTM
76%

The entire edge is 4.6 percentage points, worth $5.04 per contract before costs. Crossing the bid/ask cost $1.38 of that, which is 27% of the edge. So on this horizon, how well you get filled matters as much as what you pick and this agent captured 98.3% of theoretical credit, giving up $60 to execution across every fill.

Delta-implied OTM is what a 0.24 short delta implies, not a measured win rate. Twenty spreads cannot establish one, and this page does not claim otherwise.

What this measures

Read this before the numbers below, or they are just numbers.

The problem
Execution

A credit spread's edge is thin enough that the price you actually get decides whether the edge exists at all.

Why it stays hidden
Fill vs mid

A fill only reads as good or bad beside the mid it was priced against. Without that comparison, execution cost never shows up as a line you can look at.

What this one does
Per-bucket capture

Scores every fill against the mid it was priced at, learns capture per (symbol, tenor, delta, time of day), and refuses buckets where it cannot get filled.

The honest part. Four sessions cannot establish that a strategy is profitable, for anyone in this hackathon. What four sessions can establish is that an agent measured its own execution and acted on it, and every number on this page is reconcilable against the paper account ID above, sourced from Alpaca's own activity log rather than our bookkeeping.

Live evidence

From the agent running on the paper account right now. Reload to refresh.

Considered
2368
candidates evaluated
Traded
36
674 cleared every gate
Credit captured
98%
across 36 fills, broker-verified
Lost to execution
$60
$0 open risk · 4 gates fired

Why the agent declined

Rejections by gate. A candidate can trip more than one.

  • sizing1175
  • symbol_concentration896
  • credit_to_width643
  • model_opinion112

Execution quality

What fraction of the theoretical credit each bucket actually captures.

BucketFillsCapture
SPY:dte2:d0.25:midday2/899%
SPY:dte1:d0.25:midday3/699%
QQQ:dte2:d0.25:midday4/18100%
QQQ:dte1:d0.25:midday4/7100%
QQQ:dte2:d0.25:close1/4100%
QQQ:dte1:d0.20:midday1/1101%

The brain

Every gate the agent ran, in order, and what it decided. Refusals included.

0 decisions

Waiting for the agent's first decision.

Alpaca surfaces

Three surfaces, deliberately separate jobs. The CLI executes and verifies the book. The SDK returns chains with greeks in one call. MCP supplies read-only research. Execution never routes through MCP: it is a stdio subprocess, and a hung subprocess must never sit between the agent and its stops.

15/15 responding
CLIThe agent's hands. Execution and book verification.
okaccount getequity, buying power, options levelequity $100,958.12135ms
okclockis the market openopen=False71ms
okposition listwhat the broker says we hold2 leg(s)112ms
okorder listresting orders0 open131ms
okaccount activity listthe broker's own fill record, used to verify capture54 fills81ms
okaccount portfolioequity curve for P&L35 points201ms
okcalendarreal session close, so a half day does not defeat the flatten1 session(s)94ms
okdata corporate-actionsex-dividend dates, which drive early assignment0 action(s)98ms
okdata latest-tradelive spot; the options feed lags the underlyingSPY 762.1580ms
SDKThe agent's eyes. Chains with greeks in one call.
okTradingClientatomic multi-leg order submissionaccount 1491780f...106ms
okOptionChainRequestchain snapshots with greeks and IV, needed by the delta gate1182 contracts, 304 with greeks1384ms
okOptionHistoricalDataClientoptions market dataOptionHistoricalDataClient0ms
MCPA second opinion. Read-only by construction.
okalpaca-mcp-serverread-only research handed to the advisor39 tools exposed3178ms
okALPACA_TOOLSETStrading toolset omitted, so no order tool exists to callaccount,stock-data,options-data,assets,news0ms
okget_stock_snapshotlive context for one underlyingreturned1821ms

Decision journal

Most recent 25 of 2368.

WhenSpreadCreditDeltaMax lossOutcome
Sep 3, 06:12 PM
QQQ 723/728
call credit
0.820.260$418model_opinion
Sep 3, 06:12 PM
IWM 293/285
put credit
0.320.220$768credit_to_width
Sep 3, 06:12 PM
QQQ 714/709
put credit
0.710.253$429symbol_concentration, sizing
Sep 3, 06:12 PM
QQQ 723/728
call credit
0.820.260$418traded 1x
Sep 3, 06:12 PM
SPY 770/765
put credit
0.520.229$448traded 2x
Sep 3, 06:12 PM
SPY 777/782
call credit
0.660.262$434traded 2x
Sep 3, 06:11 PM
SPY 777/783
call credit
0.710.267$529model_opinion
Sep 3, 06:11 PM
IWM 293/287
put credit
0.360.237$564credit_to_width
Sep 3, 06:11 PM
IWM 297/303
call credit
0.440.264$556credit_to_width
Sep 3, 06:11 PM
QQQ 714/709
put credit
0.640.246$436symbol_concentration, sizing
Sep 3, 06:11 PM
QQQ 723/728
call credit
0.790.263$421symbol_concentration, sizing
Sep 3, 06:11 PM
SPY 770/765
put credit
0.540.230$446traded 2x
Sep 3, 06:11 PM
SPY 777/783
call credit
0.710.267$529traded 1x
Sep 3, 06:10 PM
QQQ 723/728
call credit
0.800.262$421model_opinion
Sep 3, 06:10 PM
IWM 293/284
put credit
0.370.233$863credit_to_width
Sep 3, 06:10 PM
IWM 297.5/302.5
call credit
0.330.218$467credit_to_width
Sep 3, 06:10 PM
QQQ 714/709
put credit
0.680.252$432symbol_concentration, sizing
Sep 3, 06:10 PM
QQQ 723/728
call credit
0.800.262$421traded 1x
Sep 3, 06:10 PM
SPY 770/765
put credit
0.510.228$449traded 2x
Sep 3, 06:10 PM
SPY 777/782
call credit
0.690.260$431traded 2x
Sep 3, 06:08 PM
QQQ 723/728
call credit
0.800.259$420model_opinion
Sep 3, 06:08 PM
IWM 297/303
call credit
0.420.261$558credit_to_width
Sep 3, 06:08 PM
IWM 293/288
put credit
0.360.238$464credit_to_width
Sep 3, 06:08 PM
QQQ 714/709
put credit
0.680.251$433symbol_concentration, sizing
Sep 3, 06:08 PM
QQQ 723/728
call credit
0.800.259$420traded 1x