Four gates between a chart and a trade
Every signal runs a gauntlet. A trend has to exist, the market has to cooperate, the news has to not be a red flag, and the risk has to fit. If any gate says no, Mercurio does nothing — and doing nothing is often the most profitable decision it makes.
Generate the signal
Trend following on 1-hour bars. No signal, no trade.
Check the regime
Longs only when the broad market confirms an uptrend.
Filter with intelligence
An AI layer can veto a trade — never amplify it.
Size, protect, execute
Strict risk limits and a stop on every entry.
Generate the signal
Mercurio trades a single edge — trend following on 1-hour bars. The goal is not to predict the next tick, but to recognize when a durable trend is underway and ride it.
Triple EMA stacks
Three exponential moving-average pairs (12/48, 24/72, 48/144) must agree on direction. One pair can whipsaw; three aligned pairs describe a real trend across timeframes.
ADX trend filter
An Average Directional Index floor near 22 confirms the trend has strength. A stack can align in a flat, listless market — ADX keeps the engine from mistaking noise for a move.
Deterministic by design
Given the same bars, the strategy always produces the same decision. There is no randomness to hide behind, which is what makes the backtest trustworthy.
The rule
The S&P 500 must close above both its 50-day and 200-day moving averages — computed from genuine daily bars — for ten consecutive sessions before any long position is allowed.
Long-only, on purpose
Backtesting showed the short side of this strategy was a structural loser in bear markets. So Mercurio does not short. In bear, choppy, or unconfirmed regimes, it holds cash.
Two different 'regimes'
This directional filter (which gates trade direction) is separate from the AI regime classifier (which informs risk sizing and the dashboard). They are independent systems with different jobs.
Check the market regime
A great signal in a bad market is still a bad trade. The regime filter keeps Mercurio out of the conditions where trend following bleeds — and the discipline to wait in cash is the whole point.
Filter with intelligence
On top of the technical edge, an AI layer screens each candidate for reasons to stand down. Crucially, it is a brake, never an accelerator.
Sentiment & events
The layer reviews news sentiment, scheduled events, and earnings proximity. A strong technical signal into an earnings print is exactly the kind of trade it can quietly veto.
Reduce, never increase
Any conviction adjustment can only lower position size or skip the trade. Size multipliers are capped at 1.0 — the risk calculation sets the maximum, and nothing can push past it.
Graceful degradation
If the AI layer is unavailable, the engine does not stop — it trades on the technical strategy alone. The edge must work without the filter; the filter only adds a margin of safety.
Size the position, protect it, and execute
Only after a signal clears the first three gates does the risk manager decide how much to trade — and how the position will be defended from the moment it opens.
Sizing & protection
- Maximum 1.5% of capital at risk on any single trade
- Every entry is a bracket order with a stop attached — no naked positions
- Size scales down as more positions open and during losing streaks
- Circuit breakers halt trading at 5% daily, 7% weekly, and 15% drawdown limits
How trades exit
- Chandelier trailing stop — 3.0x ATR below the 48-bar high — rides winners
- Stop moves to breakeven once the trade is up one unit of risk (1.0R)
- A wider ATR stop (5.5x) and a catastrophe stop (7.0x) bound the worst case
- A 60-day maximum hold closes trades that stop trending
From a price tick to a monitored position
The full path a decision travels through the engine. Each stage can stop the flow; only trades that survive all of them reach the market.
Mercurio runs in paper-trading mode only, on a single trend-following strategy in a 12-month validation. Nothing here is financial advice. The engine's edge is real in sustained bull markets and weak over a full cycle — see the backtesting page for the honest numbers.
See the gauntlet pay off
The same pipeline, replayed over two years of real market data.