# LogiSignals Engine & Methodology

> Updated: 2026-06-20

**TLDR:** LogiSignals generates crypto trade signals with a proprietary engine, not a chart scanner. Every candidate setup passes through a **7-layer scoring architecture** and **15 rejection gates** — and only about **3% of analyzed setups** are published (roughly 3–5 signals/day from hundreds of scanned USDT pairs). The engine enforces a **minimum 1.2–1.8x risk-to-reward** per timeframe (break-even win rate ~40%), draws take-profit/stop-loss from real support/resistance levels, and learns continuously from tracked outcomes via online machine-learning calibration. This page describes the architecture for transparency; it is not financial advice, and past performance does not guarantee future results.

## Key numbers

- 7 scoring layers, 15 rejection gates.
- ~3% pass rate (≈3–5 signals/day from hundreds of scanned pairs).
- Minimum risk-to-reward 1.2–1.8x per timeframe (break-even ~40% win rate).
- Targets set from proprietary support/resistance levels with touch-count scoring.
- Every setup tracked at 24h / 48h / 7d (MFE + MAE); outcomes published on the public Track Record.
- Separate structural geopolitical model covering 250 countries (modified PITF, walk-forward validated, holdout Brier ≈ 0.059).

## How it differs from a chart scanner

A typical scanner checks RSI/MACD/moving-averages and fixed Fibonacci or pivot levels on a single chart, with the same parameters everywhere and no awareness of what happens outside the chart. The LogiSignals engine adds:

- Proprietary support/resistance with touch-count scoring (not Fibonacci).
- 3-layer trend (short/medium/long term) with cross-alignment enforcement.
- Per-timeframe parameters calibrated from actual outcomes.
- Real-time derivatives positioning (funding rate, open interest, long/short ratio).
- Cross-asset regime context (BTC structure × DXY × VIX × Fear & Greed).
- Per-coin rolling beta vs BTC.
- 14+ macro indicators, geopolitical risk, and crypto event catalysts.
- An adaptive circuit breaker that tightens filters during losing streaks.
- Online ML calibration from an outcome feedback loop.

## The 7-layer scoring architecture

Technical analysis sets the base confidence; the remaining layers adjust it up or down, or reject the setup entirely.

1. **Technical core** — proprietary support/resistance, 3-layer trend, RSI gates, volume confirmation, pattern matching. Base confidence (35–65) plus 5 rejection gates.
2. **Derivatives positioning** — funding rate, open interest, long/short ratio from perpetual futures. Can hard-gate a signal or apply up to ±12 combined confidence adjustment.
3. **Cross-asset regime** — BULL/NEUTRAL/BEAR classification from BTC structure vs its 200-day average, volatility and sentiment gauges, lagged DXY correlation, and per-coin beta. Up to ±12 (largest single layer).
4. **Macro & geopolitical risk** — 14+ indicators (Treasury yields, Fed funds, DXY, VIX, S&P 500, gold, oil, Fear & Greed, CPI, FX) plus GDELT news tone and sanctions monitoring, compressed into a directional risk score. Up to ±8.
5. **Crypto event catalysts** — token unlocks, halvings, listings, hard forks, airdrops, applied directionally per coin. Up to ±8.
6. **Timing optimization** — session liquidity, day-of-week, and proximity to high-impact macro releases (CPI, FOMC). Up to ±5.
7. **ML calibration** — an online Bayesian logistic regression (8 features, SGD with L2 regularization) trained on proxy and real outcomes. Up to ±8 (capped near extremes).

## Quantitative methods in production

These run live, not in a whitepaper: online Bayesian logistic regression, Newton-Raphson maximum-likelihood estimation, walk-forward cross-validation, Page-Hinkley change detection (CUSUM) for drift, Bayesian Beta-decay reliability scoring for analysts, rolling cross-asset correlation, shadow-model promotion (A/B for prediction models), graph-based spatial neighborhood features, Kelly criterion position sizing (quarter-Kelly), an adaptive drawdown circuit breaker, liquidation-cascade detection, and isotonic (PAV) probability calibration. The methods are disclosed; the coefficients, thresholds, and trained weights are proprietary.

## Signal rejection pipeline

Scan hundreds of USDT pairs → exclude stablecoins/wrapped/delisted → fetch candles → run gates (ATR cap, support/resistance presence, pattern/breakout, risk-reward minimum, trend alignment, stop-loss risk, funding-rate extreme) → score the 7 layers → final-confidence gate → take-profit-cap gate → de-duplicate → publish as PENDING (tracked from that moment). A setup must survive every gate; one failure means rejection.

## Self-improvement

Every analysis — accepted or rejected — is stored as a full snapshot with all indicators. Outcomes are tracked automatically (MFE/MAE at 24h/48h/7d), proxy outcomes label snapshots without a linked signal, Page-Hinkley drift detection watches model performance, shadow models are promoted only when they beat the active model on enough common predictions, and every parameter change is stored in versioned parameter sets with a full audit trail.

## Primary data sources

The engine's macro, geopolitical, and derivatives layers are built on public, primary-source data:

- Macro indicators (Treasury yields, Fed funds rate, CPI, Fed balance sheet): [FRED — Federal Reserve Bank of St. Louis](https://fred.stlouisfed.org/).
- Geopolitical news tone across thousands of global sources: [the GDELT Project](https://www.gdeltproject.org/).
- Structural country-risk ground truth: [UCDP — Uppsala Conflict Data Program](https://ucdp.uu.se/) fatality data, [V-Dem](https://v-dem.net/) democracy indices, and [World Bank](https://data.worldbank.org/) macro series.
- Derivatives positioning (funding rate, open interest, long/short ratio, liquidation cascades): perpetual-futures venues such as [Bybit](https://www.bybit.com/) and [CoinGlass](https://www.coinglass.com/).

The data feeds are disclosed; the coefficients, thresholds, and trained weights are proprietary.

## Disclaimer

This page describes the architecture of the LogiSignals engine for transparency and audit purposes. **Nothing here is financial advice. Past signal performance does not guarantee future results.** Trading cryptocurrencies carries substantial risk. Engine parameters, thresholds, scoring formulas, and trained model weights are proprietary intellectual property.

Canonical page: https://logisignals.com/engine