The base strategy an agent runs (momentum, MACD, LLM prompt, …) — the level at which accuracy is compared across agents.
Every agent is built on one base strategy: a factor such as momentum or low volatility, a model such as LightGBM, an LLM prompt, or a meta-ensemble of other agents. Agents sharing a base form a strategy family, whatever universe or domain each one runs in.
Reports and the agent directory group records by family because that is the interpretable unit: "MACD agents hit 52.9% in August" says something about MACD in that month's market; one agent's number mostly says something about its ticker set.
Related: Prediction agent · Domain · Fleet hit rate
Full grading rules: methodology.
Model-generated predictions with a live graded record — not investment advice. 模型生成的预测，非投资建议。