How We Built the Tradewink Trading Loop

The first version of our trading loop answered a deceptively simple question: how does a market idea become an order, and how does that order become a managed position? In February 2026, Tradewink was still called Hoodwink. It already had analysis and broker components, but autonomous day trading needed one coordinator to connect them.

On February 13, we added that coordinator: DayTradeManager. Its original job was to screen stocks, ask the strategy engine to evaluate candidates, size a position, pass the result through risk checks, submit an order, and keep watching for an exit. The first version also had a separate scan loop and monitor loop in the main agent. That separation mattered: finding a new setup and caring for an open trade run on different clocks.

Start with a complete lifecycle

The initial pipeline was intentionally sequential. The screener found candidates. Momentum, mean-reversion, and breakout analysis gave each one a score. The position sizer turned a trade idea into a share count. The executor and risk manager decided whether an order could be placed. After entry, the monitor checked stops and targets and could flatten positions near the close.

A high score never meant “buy” by itself. A candidate could fail because the setup was weak, the size did not fit the account, a broker was unavailable, or a risk rule blocked it. We built these as explicit steps because the explanation for waiting is as useful as the explanation for trading.

The first design also made room for paper operation and broker rejection. An automated system must treat a submitted order and a filled order as different states. That lesson continued to shape later reconciliation and exit work.

Expand the evidence before expanding execution

The next day, the screener grew beyond a fixed list. It incorporated dynamic movers, volume, support and resistance, and intraday price context. Another wave added market-regime checks, quiet-market filtering, smarter order handling, and signal-quality tools. These were not extra votes for a trade. They were ways to ask whether a promising chart still made sense in the current market and account.

On February 15, micro-account support and day-trade rule checks arrived. Fractional sizing made smaller accounts usable, while explicit limits kept those accounts from receiving an order they could not support. On February 16 and 17, sessions and broker resolution became per-user. The loop could no longer assume that one global account, one broker, or one set of settings applied to everyone. Each user's scan, sizing, execution, and monitoring had to stay attached to that user's account context.

That change is part of the loop's identity today. The agent can discover a market candidate broadly, but an actual trade must be evaluated against a specific account, broker, preferences, and risk budget.

Treat monitoring as part of trading

Execution is the midpoint, not the finish line. The monitor handles stops, targets, time limits, changing conditions, and end-of-day flattening. Later iterations added broker stop synchronization, fill reconciliation, and recorded outcomes. Those additions came from the same principle as the original two-loop design: a position needs ongoing care after the scan has moved on.

The scheduler also gained a fast path for market events. A meaningful price move can queue a ticker for evaluation without waiting for the next ordinary scan. The event still enters the same evaluation and risk path. Speed does not skip the checks.

What the loop is now

Today, the trading loop is one execution pathway inside Tradewink's broader autonomous agent. The surrounding agent watches earnings, news, momentum, portfolio health, and other events. The trading pathway has a narrower responsibility: screen, evaluate, size, check risk, route an order, and monitor the resulting position.

The most important design choice has held from the beginning: the loop can decide to do nothing. Missing data, a weak setup, an account restriction, or an uncertain broker state should be visible reasons to wait. Automation earns trust by making those decisions inspectable, not by maximizing the number of trades.

The companion story is how we built the Signals loop. It shares some market intelligence with Autopilot, but its output is a researched signal for review rather than a broker order.

Trading involves risk of loss. This article describes product architecture and is not investment advice.