About AlgoZilla
Data-driven crypto trading signals built on institutional-grade methodology. No hype, no guarantees, just rigorous quantitative analysis.
AlgoZilla was born from a straightforward observation: the vast majority of crypto trading signals available to retail investors are fundamentally flawed. They are built on in-sample backtests that overfit to historical patterns, marketed with cherry-picked equity curves, and sold with unrealistic profit promises. We set out to build something different — a signal platform that applies the same walk-forward validation standards used by institutional quantitative hedge funds, making that edge accessible to individual traders at a fraction of the cost.
At the core of our platform is the Omniscius v6.2 (Pythia) model, a proprietary ensemble of LightGBM gradient-boosted decision trees trained on a candidate pool of 7,700+ features spanning momentum oscillators, trend-following metrics, volume profile analysis, multi-source sentiment data, market microstructure, candle-pattern signals, macro-cross interactions, and proprietary regime indicators. v6 introduces principle-driven feature engineering: per-coin permutation tests decide which features survive into the live classifier, instead of arbitrary correlation cutoffs and feature caps. This ensures the model adapts to evolving market conditions rather than memorizing stale correlations — and never relies on a feature unless statistical evidence demands it.
What truly sets AlgoZilla apart is our validation methodology. Every performance number we publish comes from walk-forward validated results: the model is trained on historical data and then tested on a future period it has never seen. We repeat this process across multiple expanding windows aligned with Bitcoin halving cycles to ensure consistency across different market regimes, from the deepest bear markets to the most euphoric bull runs. This is the gold standard for strategy validation in quantitative finance, and the only honest way to evaluate whether a trading system has genuine predictive edge.
We currently cover 85+ cryptocurrencies with regular per-coin retraining. Each coin has its own dedicated model because market dynamics differ significantly between large-cap assets like Bitcoin and mid-cap tokens like Polkadot or Cosmos. Our proprietary per-coin Pulse regime detection system classifies the current market state per coin per hour into INERT (capitulation), CHARGING (accumulation), AWAKENING (regime-pivot), RAMPAGE (bull run), or ATOMIC (distribution peak). Every signal parameter, including stop-losses, trailing stops, and entry thresholds, automatically adapts to the detected regime.
Transparency is not a marketing buzzword for us. We publish all key performance metrics including returns, Sharpe ratios, maximum drawdowns, win rates, trade counts, market exposure, and alpha over buy-and-hold on our live signals page. We report the bad months alongside the good ones. We run continuous paper trading to verify that walk-forward backtest results hold in real-time market conditions. And we will always be the first to tell you: past performance, including validated backtests, is not a guarantee of future results. Trading involves real risk, and you should only invest capital you can afford to lose.
Our Model: Omniscius v6.2 (Pythia)
Omniscius v6.2 (Pythia) is the principle-driven evolution of three years of Omniscius research. The classifier is an ensemble of LightGBM gradient-boosted decision trees, trained per coin with walk-forward validation. Every signal passes through a multi-stage filter pipeline (regime gating, calibrated entry/exit thresholds, signal-aware stop logic, false-peak detection) before reaching live execution — reducing false positives without relying on parallel-model voting.
The v6 generation introduces data-driven feature selection: per-coin permutation tests run across 16 prediction horizons (1h to 672h) decide which features survive into the live classifier. Features that show statistically significant predictive power get included; pure noise gets filtered out automatically. The model receives an evidence-curated input, not a hand-tuned one.
The model retrains per coin on a regular cadence, incorporating the most recent market data and adapting feature importance weights to current conditions. This is not a static system — feature selection, regime parameters, and entry thresholds all evolve as markets change. See the full Omniscius changelog for every model update from v1.0 to v6.2.
Walk-Forward Validation
Walk-forward validation is the gold standard for evaluating trading strategies. Instead of training and testing on overlapping data — which leads to overfitting — we use expanding windows with strict temporal separation. The model trains on all data up to a cutoff date, then generates predictions for the subsequent out-of-sample period. This process repeats across multiple folds aligned with Bitcoin halving cycles, covering bull markets, bear markets, and everything in between.
Every performance metric we report — Sharpe ratio, maximum drawdown, win rate, alpha over buy-and-hold — comes exclusively from these out-of-sample test periods. The model never saw the test data during training. This is the same methodology used by quantitative hedge funds and the only honest way to evaluate whether a system has genuine edge.
Why does this matter for you? Because most signal providers show in-sample results that look spectacular but collapse in live trading. Walk-forward validation is the closest proxy to live performance that backtesting can provide.
Real-world Execution Model
Every backtest and papertrade includes conservative real-world execution assumptions, so results reflect what an actual Bitvavo trader would experience:
- ✓ Entry timing: signal generated on bar X → entry filled at close of the first 15-minute sub-bar of bar X+1 (= ~15 minutes after signal). No look-ahead, no zero-latency idealization.
- ✓ Exit timing: same 15-minute lag for signal-driven exits. Stop-loss and trailing-stop exits use worst-fill semantics — the executed price is always the worst of market price or trigger price.
- ✓ Entry commission: 0.15% Bitvavo maker fee (limit orders).
- ✓ Exit commission: 0.25% Bitvavo taker fee (market & stop-loss orders).
- ✓ Additional slippage: 0.01% on every fill (conservative Bitvavo-spread assumption for top-mcap coins; small-cap altcoins effectively covered via the 15-minute lag).
- ✓ Net cost per round-trip: 0.15% + 0.25% + 0.02% slippage = ~0.42%. Higher AUM tiers (€100k+) reduce this to ~0.32%.
All published backtest and papertrade results are net of commissions and slippage.
External Data Sources
Pythia ingests live data far beyond OHLCV. The v6 release expanded the external data stack significantly:
- ✓ Macro markets: DXY, gold, silver, brent crude, S&P 500, Nasdaq 100, VIX, 10-year Treasury (TNX) — hourly resolution with full forward-leak audit on every lag computation.
- ✓ Sentiment: Fear & Greed Index (Alternative.me) at multi-timescale resolution (raw + 30d / 90d / 180d / monthly anchors).
- ✓ Crypto-market structure: BTC dominance, total market cap, alts vs BTC breadth — synthetic hourly series back to 2017.
- ✓ Derivatives signals: Coinalyze futures open interest + liquidations, Binance klines + funding rates — per-coin where available.
- ✓ DeFi & on-chain: DefiLlama TVL aggregates, Hyperliquid order-book metrics, CoinPaprika market structure.
- ✓ Synthetic Fear & Greed: a price-action-derived Fear & Greed proxy for coins where the external index lags or is missing.
7,700+ Candidate Features
Our feature pipeline computes thousands of indicators across every supported coin and timeframe.
Trend
- EMA crossovers
- Ichimoku
- ADX
- Supertrend
- Hilbert phase
- Aroon
Sentiment & Macro
- Fear & Greed
- DXY / gold / VIX
- Funding rates
- Open interest
- SynthFG
Microstructure
- Fair value gaps
- Order blocks
- Lorentzian distance
- 90+ candle patterns
- Cherenkov suite
Cross-Family NEW
- F&G × Pulse cross (42)
- Macro × Coin (37)
- Sentiment cross suite
- Pulse EMA10 ladder
- Monthly F&G anchors
- R21 macro-lag
85+ Coins, Real-Time Coverage
AlgoZilla covers 85+ cryptocurrencies across large-cap, mid-cap, and select small-cap assets. Each coin has its own dedicated model trained on coin-specific data because market dynamics, volatility patterns, and regime behavior differ significantly between assets. Bitcoin trades very differently from Solana, and a model trained on one should not blindly generate signals for the other.
Coverage includes major assets like BTC, ETH, SOL, XRP, ADA, DOT, and LTC, as well as emerging tokens that meet our liquidity and data-availability thresholds. Signals are generated hourly for all coins. View all supported markets →
Transparent Performance Reporting
We believe every subscriber deserves full visibility into how our signals perform. Our live performance dashboard reports walk-forward validated metrics for every supported coin, updated after each retraining cycle. You will find Sharpe ratios, cumulative returns, maximum drawdowns, win rates, trade counts, market exposure, buy-and-hold benchmarks, and alpha. We report every metric because Sharpe alone can mask problems that drawdown and exposure numbers reveal.
Paper trading results run continuously alongside live signals so you can compare theoretical edge with real-time execution. We do not hide bad months or selectively report favorable periods. If the model underperforms in a particular regime, you will see it. View live signal performance →
Pythia Exclusive — Coming Soon
Pythia v6.2 is our flagship Exclusive model, currently in final validation. In the coming weeks we’ll publish papertrades and backtests for the full Pythia stack, plus tier-specific commercial models that share the Pythia foundation with selected feature subsets — making institutional-grade quantitative signals accessible at every subscription level.
Updates will be announced via the Telegram channel and our Omniscius changelog.
The Team Behind AlgoZilla
AlgoZilla is built by a small team of quantitative researchers and software engineers with backgrounds in machine learning, financial mathematics, and algorithmic trading.
We operate independently — no venture capital, no external pressure to ship features before they’re ready. Every model update goes through rigorous walk-forward validation before it reaches production. We’d rather delay a release than ship something that hasn’t proven itself on unseen data.
Our philosophy is simple: transparency over hype. We publish our backtest methodology, document every model version in our changelog, and clearly state that past performance doesn’t guarantee future results. In an industry full of inflated claims, we let the numbers speak for themselves.
We’re continuously improving the model. AlgoZilla is not a finished product — it’s an evolving research platform. If you have feedback, questions, or want to discuss methodology, reach out at info@algozilla.io.
See It in Action
Explore our live signal performance or start trading today.
