Teaching AI to Compromise: Dynamic Elastic Thresholds for Crypto Trading Bots

Описание и применение

Бот для криптовалютной торговли, использующий статический порог уверенности 85, оказался парализован, отвергая 354 последовательных сделки. Внедрение динамического эластичного порога, снижающего требуемую уверенность до 70 после 300 непрерывных отказов, позволило боту адаптироваться к волатильному рынку и продолжить прибыльную торговлю, сохраняя жесткий минимум 60.

Задача

The AI crypto trading bot became paralyzed, vetoing 354 consecutive trade setups and remaining inactive during a volatile market regime.

Результат

The dynamic elastic threshold enabled the bot to adapt to a choppy market, executing trades that previously would have been vetoed, maintaining profitability while avoiding low‑probability risks.

Рабочий процесс

• Monitor consecutive veto count in the scoring engine
• When veto count exceeds 300, trigger degradation protocol
• Decrease required confidence threshold from 85 toward 70 following a predefined curve
• Maintain a hard floor of 60 to prevent reckless trades
• Continue enforcing other risk rules (e.g., F-160)
• Execute trades that meet the new elastic threshold

Доказательства

The bot wasn’t blowing up; it was doing something arguably worse. It was completely paralyzed.

Источник

The previous iteration of the scoring engine operated on a strict, static confidence threshold. To execute a trade, the aggregated multi-factor score had to exceed an 85.

Источник

If the AI says "no" 300 times in a row, the market regime has fundamentally shifted, and the AI must change with it.

Источник

The required confidence threshold dynamically steps down. In our implementation, it scales down from 85 to 70.

Источник

The system continued to enforce other iron laws, such as the F-160 rule, ensuring that manual positions or unrecorded reconciled positions were safely isolated and not subjected to automated TP/SL interference.

Источник