​
Signal Architecture: TRIGGER and FILTER

MangroveAI provides 136 active signals (66 TRIGGER, 70 FILTER) classified into two fundamental types. This classification enforces a strict architectural constraint that ensures strategies have precisely one trigger event and one filter condition for entry.

​
Signal Types

​
TRIGGER Signals (66 active)

Event-based signals that detect specific market events or pattern completions. Characteristics:
  • Detect crossovers, breakouts, or pattern completions
  • Return True when the event occurs at a specific moment
  • Used for timing entry/exit decisions
Examples: sma_cross_up, ema_cross_up, macd_bullish_cross, bb_upper_breakout

​
FILTER Signals (70 active)

State-based signals that check current market conditions. Characteristics:
  • Evaluate current market state
  • Return True when the condition is met (can persist for extended periods)
  • Used to validate or confirm trading decisions
Examples: rsi_oversold, rsi_overbought, is_above_sma, obv_bullish, adx_strong_trend

​
The 1+1 Constraint

​
Entry Rules

Every strategy entry requires:
  • Exactly 1 TRIGGER — the event that fires the entry
  • Exactly 1 FILTER — the condition that validates the market state
Both must evaluate to True for an entry to occur. 
{
  "entry": [
    {
      "name": "sma_cross_up",
      "signal_type": "TRIGGER",
      "params": {"window_fast": 10, "window_slow": 50}
    },
    {
      "name": "rsi_oversold",
      "signal_type": "FILTER",
      "params": {"window": 14, "threshold": 30}
    }
  ]
}

​
Exit Rules

  • Exactly 1 TRIGGER — the exit event
  • 0 or 1 FILTER — optional exit condition
Multiple TRIGGERs or multiple FILTERs in entry rules will be rejected. The system enforces this at the Strategy class level.

​
Why This Architecture

  1. Simplicity — Strategies are easy to understand and explain
  2. Reduced Overfitting — Fewer degrees of freedom means less curve-fitting
  3. Clear Intent — TRIGGER answers “when to act”, FILTER answers “is the market ready”
  4. Better Backtesting — Easier to analyze why a strategy entered or did not
  5. Explainability — AI Copilot can clearly explain trading decisions

​
Signal Classification Rules

​
TRIGGER Classification

Use for signals that detect discrete events:
  • Crossovers: *_cross_up, *_cross_down, *_crossover
  • Breakouts: *_breakout
  • Reversals: *_reversal
  • Squeezes: *_squeeze

​
FILTER Classification

Use for signals that check ongoing conditions:
  • State checks: is_above_*, is_below_*
  • Overbought/oversold: *_overbought, *_oversold
  • Strength: *_strong_trend, *_weak
  • Direction: *_bullish, *_bearish (when checking state, not events)
Direction words like “bullish” do not determine signal type. A bullish_cross is a TRIGGER (discrete event), while bullish_di is a FILTER (current state).

​
Validation Points

The 1+1 constraint is validated at multiple stages:
  1. Strategy Class — Validates signal_type fields and enforces count constraints
  2. Plan Signals State — AI Copilot validates TRIGGER presence before assembly
  3. Assemble Strategy State — Uses docstring metadata as source of truth for signal types