AI Signal Metrics

Methodology

Blends volatility (annualised standard deviation of returns), liquidity depth (order-book slippage estimate), smart-contract exposure (audit score inversion), and market-cap concentration (Herfindahl index on holders). Each sub-factor is normalised 0–100 and weighted before aggregation. A score above 70 does not mean "don't buy" — it means the asset carries statistically higher drawdown risk and warrants tighter risk management.

Methodology

Rate-of-change is computed across three windows (1d, 7d, 30d) and blended with relative volume (today's volume vs 20-day average). Each window is exponentially weighted so near-term moves carry more influence. The result is normalised to 0–100 using a rolling percentile against the token's own history, making it meaningful even for low-cap assets.

Methodology

A fine-tuned FinBERT model scores recent news headlines, social-media posts, and community forum activity on a –1 to +1 scale. Scores are volume-weighted (viral posts count more) and time-decayed (events older than 72h lose weight). The raw score is mapped to 0–100. Extreme readings (> 85 or < 15) often precede short-term mean reversions due to sentiment exhaustion.

Methodology

Implements the classic Moskowitz–Ooi–Pedersen time-series momentum factor. Returns over 1, 3, 6, and 12-month look-backs are sign-weighted and normalised by realised volatility to give a volatility-adjusted momentum score. A positive TSMOM means the asset has been trending up persistently across most look-back windows — conditions historically associated with trend-continuation. A negative score suggests the opposite.

Methodology

Computed via rescaled range (R/S) analysis over a 90-day rolling window. The Hurst exponent measures the long-range dependence of the price series. Values significantly above 0.5 indicate that past returns are positively correlated with future returns (persistence) — a tailwind for trend-following. Values below 0.5 indicate negative autocorrelation — the price tends to snap back after moves, favouring mean-reversion strategies. This is a regime indicator, not a directional signal.

Methodology

Z = (today's volume − 30d mean) / 30d standard deviation. A high positive Z-score means today saw a volume spike relative to recent history — often the first signal of institutional activity, news impact, or whale accumulation/distribution. A negative Z-score means unusually quiet trading, which can precede a squeeze. Note: the direction of price impact depends on whether the volume is buy- or sell-driven; the Z-score alone is not directional.

Methodology

Average True Range (ATR) is computed as a 14-day rolling average of (high − low, |high − prev close|, |low − prev close|). The result is normalised to a 0–100 percentile over the past 90 trading days. Low percentile readings mean price has been unusually calm relative to its own recent history — a pattern that historically precedes sharp directional expansions. High percentile readings mean the asset is already in a volatility expansion phase.

Methodology

A weighted combination of three sub-signals: ATR Compression (is volatility coiled?), Volume Build-up (is volume quietly accumulating ahead of a move?), and TSMOM Alignment (is the trend direction confirmed?). Each sub-signal contributes 33% to the composite. The score is calibrated so that readings above 70 occur in the top ~15% of all historical observations for this asset — not a guarantee of a move, but a meaningful elevation in probability.

Methodology

A two-state Hidden Markov Model is trained on the 90-day rolling returns and volatility of the broad crypto market (BTC-weighted). The model outputs the posterior probability of being in the "bull" hidden state. High values (> 0.6) indicate macro tailwinds — trend-following strategies historically outperform in this regime. Low values (< 0.4) indicate macro headwinds — defensive positioning and shorter time horizons are typically rewarded. This is a market-wide signal, not token-specific.

Methodology

An Isolation Forest model is trained on a 60-day baseline of four on-chain metrics: daily transaction count, active address count, exchange inflow, and exchange outflow. Each day the model scores the current observation. A score ≥ 0.5 means today's on-chain fingerprint is an outlier relative to the trailing baseline. Anomalies can be bullish (e.g., unusual accumulation) or bearish (e.g., exchange inflow spike suggesting sell pressure) — always investigate what type of anomaly it is.

Methodology

Counts on-chain transfers of > $100,000 USD equivalent in the past 24 hours, normalised against a 30-day rolling average. The normalised score is then mapped to 0–1. High activity does not indicate direction — a whale moving tokens to an exchange may be preparing to sell, while a move to a cold wallet may signal long-term accumulation. Use alongside CVD (cumulative volume delta) from the trading bot signals for directional context.

Methodology

Computed as the Pearson correlation of daily log-returns between this token and BTC over the trailing 30 calendar days. A reading near +1 means the token rises and falls with Bitcoin — macro crypto sentiment drives price. A reading near 0 means the token has its own idiosyncratic price driver (product launches, protocol events, tokenomics). A negative reading is rare but indicates inverse price behaviour relative to BTC. This metric helps size position risk during broad market sell-offs.