Nostradamus

Overview

Nostradamus is an autonomous trading intelligence platform that detects hidden liquidity patterns, validates them through rigorous statistical backtesting, and monitors markets in real time. The system scans crypto, equities, forex, and metals across up to 21 timeframes per asset class - from monthly candles down to one-hour intervals, including non-standard institutional timeframes like 23H, 18H, 5D, and 3W that most retail tools ignore entirely.

At its core, the detection engine identifies three categories of hidden liquidity: HOB (Hidden Order Blocks fully concealed behind fair value gaps), PHOB (partially mitigated variants), and BB (Breaker Blocks with wick intrusion). Each level receives a composite power score based on timeframe weight, order block size, structural position at swing highs and lows, and distribution evidence - the presence of a fair value gap after the order block, indicating institutional movement. A Volume Spread Analysis module then classifies directional intent using a three-signal majority vote across volume distribution, POC position, and absorption bar detection.

The Challenge

Manual technical analysis across multiple assets and timeframes is slow, error-prone, and impossible to scale. A single trader monitoring even seven crypto assets across 21 timeframes faces 147 charts to review - and that analysis is outdated the moment a new candle closes. Opportunities at hidden liquidity levels appear and expire within minutes, often during off-hours when no one is watching.

Beyond detection, the harder problem is trust. Most trading signals lack statistical validation. Without walk-forward testing on out-of-sample data, there is no way to distinguish a real edge from curve-fitted noise. The client needed a system that could not only find patterns autonomously but prove they worked on data the model had never seen.

The Solution

We built an end-to-end platform with four tightly integrated layers. The detection engine scans every symbol across its full timeframe set, clustering nearby levels into zones and ranking them by power score. A backtesting engine then replays every detected zone against historical OHLCV data - determining entry at the order block midpoint, stop loss at the zone edge with buffer, and take-profit at configurable R:R multiples. Same-bar ambiguity is resolved conservatively, always assuming the stop is hit first.

On top of backtesting sits a walk-forward validation engine that splits history into five expanding folds and runs a 162-configuration grid search on each in-sample window. The winning parameters are evaluated on the out-of-sample fold, measuring degradation between IS and OOS performance. Bootstrap confidence intervals at the 90% level and permutation tests (p<0.001) confirm the edge is statistically significant, not random.

The live system runs on Railway as a dual-monitor architecture inside a single container - one process for crypto (BTC, ETH, SOL, DOGE, XRP, BNB, ADA) and one for equities (AAPL, MSFT, NVDA, AMZN, GOOGL, META, TSLA, SPY, QQQ, AMD). Each process polls prices every five seconds, rescans levels every fifteen minutes, and advances alerts through a five-state lifecycle: Detected, Approaching, At Level, Triggered, and Filled. When a zone triggers, the paper trading engine opens a position with partial exits - 25% at TP1, 20% at TP2, and a 55% runner with a trailing stop - all tracked in SQLite and reported via Telegram bot in real time.

Tech Stack

Results