Why a Top Analyst Handed $80,000 Altcoin Portfolio to Claude AI

Crypto analyst Michaël van de Poppe watched his altcoin portfolio shrink from $160,000 to $80,000 over 18 months. He attributes the losses to emotional decision-making during volatile cycles. This week he made a structural change: he gave trade authority to Claude AI, a custom framework built around sigma deviations, RSI triggers, and Bitcoin correlation analysis.

The simple read is that van de Poppe outsourced trading to a chatbot. The better market read is that the framework is designed to enforce discipline by removing the behavioral gap between analysis and execution. The most revealing test came when van de Poppe wanted to sell a position and Claude told him to hold.

The Framework: Sigma, RSI, and Bitcoin Correlation

Van de Poppe built his system to process three inputs for every decision:

Sigma deviation: how far price has moved from its rolling mean. Extreme extensions trigger sell signals.
Daily RSI: overbought readings above 70 prompt exit orders when combined with sigma extension.
Bitcoin correlation: if BTC is pulling back, altcoin long signals are downgraded regardless of the asset's own readings.

The system outputs specific buy and sell levels, not general market commentary. Van de Poppe's role is now limited to executing the signals.

The First Real Test: Claude Forces a Hold

Van de Poppe wanted to sell a position that had grown to 41% of his portfolio. Claude analysed the framework and concluded the asset had not reached the exit zone. The AI's reasoning on the NEAR position was specific:

Van de Poppe admitted it felt uncomfortable sitting on a 41% allocation into a single asset while the AI said do nothing. That discomfort is the exact emotion the framework is meant to override.

Three Trades That Reveal the Model's Strengths and Limits

Van de Poppe executed several trades during the integration period. The outcomes show where the framework works and where it breaks down.

PEAK Exit: Clean Signal Execution

Van de Poppe sold $500 of PEAK after the sigma extension and RSI hit overbought levels on the daily chart. The same technical setup preceded a 30% correction in three days during a prior cycle. He followed the signal, took profit, and avoided the drawdown.

Renzo to Wormhole: When the Trader Overrides the Model

The Behavioral Gap: Why Van de Poppe Handed Over Control

Van de Poppe said his 70-80% drawdowns over the past year and a half were driven by emotional reactions to volatility. The framework is designed to remove the behavioral gap between analysis and execution. The key insight is that most trading mistakes are emotional. AI has no emotions.

Trusting the Model Through Discomfort

The transition is not seamless. Van de Poppe admitted the discomfort of holding a 41% allocation while the AI told him to do nothing. That discomfort is exactly what the framework is meant to override. It exposes a deeper tension: the trader must trust the model even when the model appears to be doing nothing.

Bottom line for traders: handovers to AI do not eliminate losses. They redirect the axis of failure from emotion to model calibration. A poorly calibrated sigma trigger or a stale correlation assumption can cause the same drawdowns as panic selling.

What Confirms the Setup Is Working

The framework works when van de Poppe follows its signals on every trade, not just the ones that match his bias. The NEAR position will be the first true test of sustained compliance. If he holds through volatility until the exit zone, the discipline is real. If he overrides the model again, the AI integration is window dressing.

The Upcoming Telegram Dashboard and Execution Risk

Van de Poppe is building a Telegram-connected dashboard that will send automated buy and sell signals when the framework triggers. The goal is to remove manual intervention entirely. The dashboard will execute the trade or send a signal to van de Poppe's exchange API.

This is a natural extension of the framework. It also introduces execution risk. If the Telegram bot or API fails during a volatile move, the trade slips or misses entirely. Van de Poppe will need a fallback mechanism for both connection failure and human override in case of model failure.

For traders considering a similar approach, the practical rules from van de Poppe's experience are clear:

Define exit zones in advance using sigma deviations and RSI thresholds. Do not adjust them in real time.

Use Bitcoin correlation as a gate, not just a secondary indicator. If BTC is in a downtrend, altcoin longs should be downgraded.

Test the framework on a single position before full deployment. The Wormhole mistake happened because van de Poppe ignored the correlation filter.

Plan for infrastructure failure. A Telegram dashboard is only as reliable as the API connection and the exchange's uptime.

The framework is not a magic fix. It is a set of rules that forces the trader to sit through discomfort. The value is in the enforcement, not the prediction. Van de Poppe's portfolio will prove whether discipline beats insight.