Methodology

How the lab gets to a decision.

These steps are intentionally narrow. AI interprets the setup, but it does not get to ignore rules, current exposure, or the option to do nothing.

01. Structured setup

Indicators and rules are calculated first.

The raw setup is built before any model reads it. Trend, regime, levels, and recent context enter the room before language does.

The system calculates RSI, Stochastic RSI, MACD, ADX, EMA-20, EMA-200, Bollinger Bands, VWAP, and relative volume from raw OHLCV bars. These indicators are structured into a snapshot that the AI reads — it never calculates them itself.

The regime (trending, ranging, volatile, or quiet) is determined from indicator agreement, not from the AI's interpretation. This means the model starts with a fact, not an opinion.

02. AI interpretation

The model reads a bounded market snapshot.

The AI explains the action inside a fixed structure instead of inventing its own playground. It is there to interpret, not to freestyle.

The prompt includes the indicator snapshot, the active strategy's rules, regime-specific behavior instructions, and the current portfolio state. The AI must respond with a structured decision: action, confidence, summary, analysis with bullish/bearish/neutral factors, key levels, and what would change the call.

The model cannot invent new indicators or ignore the ones provided. It works within the strategy's constraints.

03. Portfolio constraint

Cash, exposure, and limits stay visible.

Current holdings, available cash, and risk caps all constrain the action. The lab should be able to say HOLD as honestly as it says BUY or SELL.

Before any trade, the system checks: Is there enough cash for a new position? Does this coin already have a position? Is the portfolio too concentrated in one asset? These are hard rules, not suggestions.

The AI sees the full portfolio state — current value, cash reserve, open positions, and allocation percentages — so it can factor exposure into its reasoning.

04. Demo execution

Orders are paper trades on Binance demo.

The order path is real for the demo environment, which means positions and P&L can be inspected in public without pretending this is live capital.

When the AI decides BUY or SELL, the system submits the order to the Binance demo API. The fill price, quantity, and status are all real — the only difference is that no actual money moves.

This matters because paper trading with a real API means slippage, fill failures, and order rejections can all happen and be documented. It is not a backtest.

The indicators powering each decision.

Every decision in the lab starts from the same indicator set. Here is what gets calculated and why it matters.

Momentum

RSI

Relative Strength Index (14-period). Measures whether recent price action has been mostly buying or selling pressure. Above 70 is overbought, below 30 is oversold.

Stochastic RSI

RSI applied to itself. Faster signal for momentum shifts. Values near 0 suggest oversold momentum, near 1 suggests overbought.

MACD & Signal

Moving Average Convergence Divergence. Tracks the relationship between two EMAs. Crossovers between the MACD line and signal line flag potential trend changes.

Trend

ADX

Average Directional Index. Measures trend strength regardless of direction. Above 25 suggests a meaningful trend is forming; below 20 means the market is likely ranging.

EMA-20 & EMA-200

Exponential Moving Averages at 20 and 200 periods. EMA-20 tracks the short-term trend. EMA-200 is the long-term structural bias. Price above both is structurally bullish.

SMA-5 & SMA-10

Simple Moving Averages at 5 and 10 periods. Used for very short-term trend direction and crossover signals within the decision window.

Volatility & Volume

Bollinger Bands

Upper and lower bands around a 20-period SMA, set at 2 standard deviations. Price near the upper band may be overextended; near the lower band may be compressed. A squeeze (narrow bands) often precedes a breakout.

VWAP

Volume-Weighted Average Price. The fair price accounting for volume. Price above VWAP suggests buyers are in control for the session.

Relative Volume

Current volume compared to the average. Values above 1.5 suggest unusual interest; values below 0.5 suggest thin activity where signals carry less weight.

Context

Regime

A derived label (trending, ranging, volatile, quiet) based on indicator agreement. The AI receives the regime as a fact, not as something to determine.

Daily High & Low

The session's price range. Helps the AI understand where current price sits relative to the day's extremes.

Data Points & Timeframe

How many bars were used and at what interval. This tells the AI (and the reader) how much history backs the current snapshot.

The strategies that shape interpretation.

The lab currently runs 6 active strategies. Each one tells the AI how to read the same indicators differently.

ADX Trend Strength

Adx Trend Strength

Uses ADX as a primary filter to trade only when trend strength is confirmed. Combines ADX directional indicators (+DI/-DI) with EMA pullbacks to avoid false signals in weak or choppy markets.

Breakout

Trades volatility expansion when price breaks out of consolidation ranges. Uses Bollinger Band compression, ATR expansion, and volume spikes to identify and enter breakouts.

Composite

Combines multiple strategy signals — trend, momentum, structure, and market health — into a single weighted decision. Requires alignment across at least 2-3 sub-signals before acting. This is the default platform strategy.

Mean Reversion

Trades reversions to the mean when price overextends beyond Bollinger Bands or RSI extremes. Works best in sideways and range-bound markets. Avoids strong trends where momentum can persist.

Support/Resistance Flip

Sr Flip

Trades the classic S/R flip pattern where old resistance becomes new support after a breakout. Waits for price to retest the broken level with volume confirmation before entering.

Trend Following

Follows established trends by entering on pullbacks to key EMAs. Uses EMA alignment, RSI pullback signals, and MACD confirmation to ride trends while avoiding chasing pumps.

See all strategies in detail

Keep following the experiment

The point is to make the lab inspectable.

You should be able to watch the portfolio, see how and why decisions are made, learn from it, and understand when I got something wrong.

If you spot wrong data or want to suggest an improvement, send feedback here.

If having your own personal lab sounds useful, leave your email on the waitlist. If enough people are interested, it is worth building.

Join the Waitlist Back to the Homepage

This is a public paper portfolio. Past performance does not guarantee future results. Hugo's Trading Lab publishes rules-based analysis with AI-assisted decision summaries for informational purposes only, not financial advice.