Trebuchet - how it was built and tested
Trebuchet started as a study of a working four-hourly channel-breakout approach for gold. The work had two goals: understand exactly what makes that kind of system work, and then build our own version of it with fewer give-backs and smaller dips.
1. What the rules are
Every four hours the EA reads a price channel - the highest high and lowest low of a fixed number of past bars - and rests a buy stop exactly on the channel high and a sell stop exactly on the channel low. One breakout is traded by seven legs that share the entry and differ in their target (0.4 to 5 percent of price) and stop loss (0.1 to 0.9 percent of price plus the spread). An order is placed only when the channel edge is far enough from price, untriggered orders expire after three and a half to ten and a half days, and a trade that is a little ahead gets a break-even stop.
These rules were measured, not guessed: our first version placed its orders where the approach we studied did for most legs on two brokers, and its net result landed within a few percent of it (backtest, MetaTrader 5 Strategy Tester, real ticks, Vantage, 2025-07-04 to 2026-10-13).
2. The give-back problem
On a client's chart a breakout system like this shows a familiar pattern: equity climbs above the balance while trades run toward their targets, then falls back as many of those trades return to their break-even stop. Measured trade by trade, about half of the trades that reached 40 percent of the way to their target ended at break-even or worse, and trades that came within ten percent of their target and then missed kept only a few percent of it (backtest, MetaTrader 5 Strategy Tester, real ticks, Vantage, 2025-07-04 to 2026-10-07).
Three kinds of protective exit were compared on three separate test slices, and the setting was chosen by the WEAKEST slice, not the strongest one:
- a lock that follows the open profit on every trade from early on cut the far-target trades that carry the results; it was dropped;
- a give-back limit sized to recent volatility (the kind used on short-target systems) proved too tight for targets this far away;
- two fixed locks scaled to each trade's own target worked: a burst lock for trades that cover 40 percent of the way within three minutes of the fill (fast spikes often snap back), keeping 30 percent of the target, and a near-target lock at 75 percent of the way, keeping 20 percent. Neighbouring settings gave similar results, and a setting picked on two slices held on the third.
Together they cost no net profit on any of the three brokers tested and cut the trades that gave everything back from 54 to 38 percent (backtest, MetaTrader 5 Strategy Tester, real ticks, default settings, BlackBull 2022-11 to 2026-10, Vantage 2025-07 to 2026-10, Pepperstone 2025). An optional trailing stop over the last stretch keeps most of a near miss for a small cost in net profit.
3. When it works and when it does not
The edge of a channel breakout is the follow-through of a trending market. In these tests:
- the profit came mainly from the trending gold market of 2024 to 2026; in the quiet market of 2023 the balance went sideways for most of the year (backtest, MetaTrader 5 Strategy Tester, real ticks, BlackBull, 2022-11 to 2026-10);
- inside a trending period, ranging days earned about as much per trade as trending days - short ranges were not the problem;
- busy markets were: when gold's recent daily range was in the busiest third of its own past year, the fixed stops sat inside the noise and the breaks were weaker. Selective mode stands aside then;
- a long sideways market is the real weakness: on gold prices from 2013 to 2015, tested at today's price level, the same rules lost money over two years, and Selective mode only softened it.
4. Living next to other gold breakout systems
Breakout systems trade the same kind of moment. In the tests about half of Trebuchet's entries fell within fifteen minutes of a same-side entry of Kingfisher, and about half of a Captain Karat-Smash entry. Because the exits differ, the daily results moved little with Kingfisher's (correlation 0.03) and moderately with Captain Karat-Smash's (0.35) (backtest, MetaTrader 5 Strategy Tester, real ticks, Vantage, 2025-08-08 to 2026-09-25, default settings). Trebuchet is a different exit design on similar moments, not a different market idea.
5. What is still open
- Balance-based position sizing (the tests use a fixed 0.01 lot per order).
- One of the seven legs tracks the approach we studied less closely than the others.
- Peak equity drawdown on two of the brokers is a little above the studied approach's, while net profit and profit factor are above it on all four brokers tested.