Composite scoring

From reading to subscore

score_indicator(name, &output, &alerts) turns one indicator output and its alerts into a SubScore:

pub struct SubScore {
    pub indicator: String,
    pub score: f64,       // -1.0 ..= 1.0, negative is bearish
    pub raw_value: f64,
    pub reason: Option<String>,
}

Alerts such as a bullish cross or an oversold extreme push the score towards +1, their bearish counterparts towards -1. When an indicator raised no alert, some are scored from their level:

Indicator	Level scoring
rsi, stoch_rsi, mfi, williams_r, connors_rsi	above 70: -0.4, below 30: +0.4, above 55: +0.2, below 45: -0.2
macd	histogram sign: ±0.3
bollinger	%b above 1: -0.5, below 0: +0.5
vortex	(VI+ − VI−) × 2, if the gap exceeds 0.05
trend_quality	value / 100

The oscillators read as mean reversion: overbought is bearish. In a strong uptrend they therefore argue against the trend, and the regime vetoes the result. Combine indicators whose reading matches the signal you want to grade.

Aggregation

pub fn aggregate_subscores(
    subscores: Vec<SubScore>,
    weights: Option<&HashMap<String, f64>>,
    regime: MarketRegime,
    sr_zones: Vec<SupportResistanceZone>,
    latest_bar: Option<&Bar>,
    atr_val: f64,
) -> CompositeSignal

Step by step:

1. Clean up. Subscores with a non-finite score or raw value are dropped, the rest are clamped to -1..=1. With nothing left, the result is neutral, Hold and Veto.
2. Weigh. Each subscore gets the weight stored under its indicator name, default 1. A weight that is not finite or outside 0..=1000 falls back to 1. The score is the weighted mean, clamped to -1..=1.
3. Direction. score >= 0.20 is bullish, score <= -0.20 bearish, anything in between neutral.
4. Agreement. The share of subscores that agree with the direction: above 0.05 for bullish, below -0.05 for bearish, within ±0.20 for neutral. It is a share, not a win probability.
5. Heat score. |score| * 0.6 + agreement * 0.4, between 0 and 1.

Permission grade

The regime decides whether a direction may be traded at all:

Condition (checked in this order)	Grade
Neutral direction, or direction against the regime (bullish in bearish expansion …)	Veto
Regime is Transition	Caution
Heat >= 0.50 and direction matches the regime, or consolidation with agreement >= 0.75	ClearToTrade
Heat >= 0.30	Caution
Otherwise	Veto

Only ClearToTrade produces a trigger: Buy for bullish, Sell for bearish. Every other combination is Hold.

Risk plan

For a Buy or Sell with a valid latest bar, the signal carries a RiskPlan, measured in ATR from the close:

Level	Buy	Sell
Stop loss	close − 1.5 ATR	close + 1.5 ATR
Target 1	close + 1.0 ATR	close − 1.0 ATR
Target 2	close + 2.0 ATR	close − 2.0 ATR

atr_val is the absolute ATR in price units (the atr indicator puts it in extra["raw"]). If it is not finite and positive, one percent of the close is used instead. aggregate_subscores_with_instrument additionally rounds the plan to the instrument’s tick size.

Weight presets

WeightPreset::weights() returns a ready-made weight map: Balanced (all weights 1), TrendFollowing, MeanReversion and VolumeLocation.

use kestrel_chartkit::{aggregate_subscores, score_indicator, WeightPreset};

let weights = WeightPreset::TrendFollowing.weights();
let subscores = vec![score_indicator("macd", &macd_out, &macd.alerts())];
let signal = aggregate_subscores(subscores, Some(&weights), regime, Vec::new(), Some(&bar), atr);
println!("{:?} {:?} {:.2}", signal.direction, signal.permission, signal.heat_score);

CompositeSignal::explanation is a generated sentence summarising direction, heat, agreement, permission and regime. It is currently written in German; build your own text from the structured fields if you need another language.

The composite signal chart in the showcase marks every bar where a trigger starts.