Research note

Composite & lens system

How six weighted lenses become one −100 to +100 regime score: the formula, the weights, the Direction conventions, and hysteresis.

2 min read

The composite score is a weighted sum of six lens scores, scaled to [-100, +100]:

composite = Σ (lens_score × lens_weight) × 100

Default lens weights

LensWeight
macro20%
credit20%
breadth15%
volatility20%
crossasset10%
sentiment15%

The weights sum to 1.0.

Each lens score is itself a weighted average of its indicators' regime contributions in [-1, +1]. At the shipped setting, that blend winsorizes them at their mean ± 2.5σ when at least nine weighted contributions are present, so a print beyond that band cannot dominate the lens; a print inside ±2.5σ is left alone. The per-indicator contributions themselves stay raw, so a clipped lens score can differ from them when a clip binds (docs/DATA_QUALITY.md). An indicator's regime contribution is its percentile rank within a 3-year lookback, mapped to [-1, +1] via its Direction:

  • HIGHER_IS_BULLISH — high percentile → +1 (bullish); e.g. high HY OAS, high VIX (after our May 2026 polarity flip).
  • LOWER_IS_BULLISH — low percentile → +1; e.g. falling DXY.
  • HIGHER_IS_BEARISH_CONTRARIAN — high percentile → −1; e.g. AAII bull-bear at extreme highs (crowded long is bearish input).
  • LOWER_IS_BEARISH_CONTRARIAN — low → −1; capitulation = bullish input (the framework already inverts contrarian indicators internally so "positive contribution = bullish" is always the convention).

Regime classification thresholds

CompositeLabel
>= +60Strong Bull
>= +20Bull
>= -20Neutral
>= -60Bear
< -60Strong Bear
plumbing tripDefensive — Plumbing Stress (overrides all of above)

The headline label is then run through hysteresis: a flip only sticks if the new label has held for at least 3 consecutive raw classifications. This stops the regime label from oscillating on noise. Plumbing trips bypass hysteresis (they're step-changes that demand action immediately).


This article is also a chapter of the Education page.

Not financial advice

Built for education. Backtests are historical; they do not predict future returns and are not a recommendation to buy or sell any security.

connecting…