risk-management

An automatically generated risk-management reference for a trading agent, built from analyzing profitable versus losing trades across competition outcomes. It is intended to be consulted when sizing positions or setting stop-losses. The document is produced and updated by an Observer Agent and carries metadata about its own freshness, reporting a last update timestamp, a count of active patterns, a total sample count in the thousands, and a confidence threshold, so readers can gauge how current and well-supported the rules are.

The core of the skill is a large table of learned rules, each with a plain-language description, a success rate, a sample count, a confidence percentage, and how many times the pattern was seen. Rules span themes such as trade frequency correlating inversely with performance in bullish markets, halting SHORT entries when all tracked assets are broadly positive, position sizing at 2 percent equity risk with a 2 to 1 reward ratio under aligned market direction, capping trades per 24 hours, multi-timeframe bearish alignment, and behavior in mixed or low-volatility regimes. A Top Risk Rules section expands the highest-signal entries with their success rate, observation count, confidence, and first-identified date. Because entries range from very high to very low success rates, the table effectively records both what worked and what failed.

General guidelines distill the patterns into simple defaults: never risk more than 2 percent of equity on a single trade, use stop-losses on every position, reduce position size in high-volatility regimes, and do not add to losing positions. A confidence guide maps percentage bands to interpretation, from 90 percent and above as high confidence with strong historical support down to below 60 percent as experimental and needing more data. Because it is auto-generated from historical samples of variable size and confidence, the rules are best treated as data-driven inputs weighed alongside other signals rather than guarantees.