Trend Persistence Score

TradingView: https://de.tradingview.com/script/iVyNujT3/

This sits alongside regime_classifier as the graded-magnitude complement: regime_classifier produces a categorical Trend / Range / Chaos permission gate for the reversal pipeline, while TPS produces a continuous how-strong reading you can plot, threshold and alert on.

Features

• Four trend-efficiency sensors on one comparable 0–100 persistence axis
• Role-level sensor weights (R² 40 · ER 25 · ADX 20 · FDI 15 by default) plus per-sensor lines so you can see which sensor drives the verdict
• State read — Strong / Healthy / Transition / Weak / Dead — with one-band hysteresis so it does not flip on score noise
• ADX damping — ADX measures movement, not directional cleanliness; when both structure sensors (R², ER) are weak, its sub-score is damped so raw activity cannot rescue a dead-structure score
• Transition Risk readout — a derived inverse of TPS for “how likely is the trend to break down” (display + alert only, not a structure-break detector)
• Calibration anchors (ADX range/trend, FDI trend/range, damp threshold/factor) exposed as inputs — no hardcoded thresholds buried in the signal logic
• Two-dimensional oscillator read (default Direction × Strength): the line height is persistence, its colour is direction (green = up-efficiency, red = down, grey = no clear direction, from the regression sign), and its saturation scales with strength (vivid in a clean trend, pale in a dead market) — plus an optional heat column fill to zero. A strong green column = clean uptrend; a fading green column = uptrend losing power; a shift to red = down-structure forming. Direction stays visual only, never fed into the score. Switch to Persistence State mode for the classic band colours.
• Directional context (▲ Up / ▼ Down / • Flat / • Compress) also in the table — the flat read distinguishes measured Compression (Bollinger-width percentile rank) from plain Flat
• Compact mobile-friendly info label (default on) and a debug log on every state change (records which sensor drove the move)
• Weakening / Transition / Recovery / Transition-Risk-High alerts

Sensors

Sensor	Formula	Reads	Strong-trend reading
Regression R²	correlation(src, bar_index, n)²	fit of price to a straight line	high R²
Efficiency Ratio (Kaufman)	|Δsrc(n)| / Σ|Δsrc|	how directly price travelled	high ER
ADX strength + slope	ta.dmi ADX mapped + 3-bar slope	trend strength and whether it is building	high & rising ADX
Fractal Dimension	log(pathLength / range) / log(n) + 1	path roughness (≈1 clean … ≈2 noisy)	low FDI

Each sensor is mapped to a 0–100 sub-score via its own calibration anchors, then combined by the role-level weights into the composite. Because the four sensors measure correlated facets of the same property, the score is an intentional robustness ensemble — no single sensor can dominate or veto the reading.

Score and state

1. TPS = weighted mean of the four sub-scores, EMA-smoothed by Score Smoothing.
2. State is read from State Levels: ≥ Strong → Trend Strong; ≥ Healthy → Trend Healthy; ≥ Transition → Transition; ≥ Weak → Weak / Range; below → Trend Dead. Upgrades apply immediately; downgrades require a 3-point overshoot below the band edge (hysteresis).

How to use

• Trend-following filter: only take trend entries while TPS is in Healthy/Strong; stand aside in Transition/Dead.
• Reversal context: a high score rolling over (Strong → Transition) flags a maturing trend; a Dead reading marks range conditions where mean-reversion logic is more appropriate.
• Attribution: when the score disagrees with your read, the sub-score lines and the dashboard show exactly which sensor is responsible.

Notes

• Direction-agnostic on purpose — pair it with a Trend-direction module if you need a directional bias; the Bias cell is a convenience read only.
• Default Main Length 34 suits swing timeframes; shorten for intraday, lengthen for higher timeframes.