AUTONOMOUS OPTIONS DESK

A desk that hires
its own strategies.

Aperture researches bounded options strategies, promotes only those that survive a chronological holdout, and lets a deterministic Risk Warden decide whether any order reaches Alpaca.

Watch the desk think Read the method
ALPACA execution + market data AI AGENTS research + decisions PYTHON final risk authority
ACCOUNT EQUITY — From a $100k paper account
TOTAL RETURN — Since desk inception
TODAY — Marked to market
DRAWDOWN — From high-water mark
DEFINED RISK OPEN — — structures
WARDEN VETOES — Unsafe proposals stopped
THE AUTONOMY PROOF

Not a signal bot.
A capital-allocation loop.

The model never places an order. It helps the desk form testable hypotheses and explain results. Promotion, sizing, vetoes, fills, and P&L attribution stay deterministic and auditable.

01
RESEARCH LAB

Generate. Test. Hire.

An agent selects one-knob mutations from a bounded strategy family without seeing outcomes. Python tests them against real option history and an untouched holdout.

—tested tonight
—cumulative trials
↗
LATEST OUTCOMEAwaiting research
02
RISK WARDEN

Language proposes.
Code disposes.

Every structure is reduced to expiry payoff geometry, then checked against liquidity, concentration, quote freshness, loss limits, and the tournament clock.

✓ DEFINED RISK ONLY ✓ QUOTES ≤ 90 SECONDS ✓ MAX 4% PER TRADE ✓ ATOMIC MULTI-LEG ORDERS
03
CAPITAL MARKET

Performance is the vote.

Realized return on risk earns allocation. Repeated vetoes count as evidence of miscalibration. New hires start at 5%; persistent failure means no new capital.

PORTFOLIO EQUITY

One-week mark

— No observations
Account equity over the last week The chart will update when a snapshot is loaded.
Account equity Alpaca portfolio history · hourly marks, sessions spaced evenly
LATEST LAB RUN

Awaiting close

IDLE
SELECTION WINDOW —
35D
EMBARGO
UNTOUCHED HOLDOUT —

The nightly lab runs only after a completed market session.

F
HYPOTHESIS MODELNot configured
F
EXPLANATION MODELNot configured
THE ROSTER

Strategies are employees.

Each gets a budget, a record, and a reason to remain on the desk.

LIVE BOOK

Open structures

0
Open options structures and their maximum loss
StrategyUnderlyingStructureMax lossStatus
SCOREBOARD

Strategy attribution

fill-derived
Per-strategy realized performance and risk
StrategyClosedWin rateRealizedRisk at work
THE NON-NEGOTIABLES

The Warden
does not negotiate.

These limits are code, not prompts. The reasoning model cannot raise them, skip them, or reinterpret them after seeing a trade it likes.

FINAL
AUTHORITYRISK
WARDEN
4%
MAX LOSS / TRADEWorst-case expiry payoff, not premium paid
8%
MAX / UNDERLYINGStops correlated structures piling into one symbol
25%
PORTFOLIO RISK CAPSum of defined maximum losses across the book
−3%
DAILY ENTRY HALTExisting exits continue; new entries stop
−8%
DRAWDOWN BREAKERActive kill switch and atomic flattening
16:00
THURSDAY MEASUREMENT · ETOfficial EOD equity snapshot; no forced liquidation
APPEND-ONLY AUDIT

Every decision
leaves a receipt.

Proposals, veto reasons, fills, hires, fires, and research outcomes are recorded as events. The dashboard exposes the public-safe subset.

    DAILY SHAREHOLDER LETTER

    Awaiting first close

    WRITTEN BY THE DESK
    After each session, the reasoning layer explains a deterministic fact block. The publisher rejects invented or rounded numbers.
    THE APERTURE DESKFACTS LOCKED BY CODE