zhangsensen

Use when analyzing A-share or index OHLCV data with multi-timeframe Chan-style structure, support/pressure ladders, MACD auxiliary context, current-state observations, or deterministic condition branches without action language.

Deep, structured analysis for backtest ledgers, trade logs, daily PnL series, scorecards, and A/B comparison packs. Use when the user provides concrete parquet/csv/json backtest artifacts and wants a decision such as promote, watch, or reject based on the data itself. Covers return, risk, stability, and trade quality, prevents single-metric conclusions, and prevents total PnL from dominating the verdict. Do not use for open-ended root-cause diagnosis without a concrete dataset, or for implementation-vs-research consistency audits.