---
name: backtest-review
description: Review a supplied backtest report for timing, execution, cost and validation gaps. Use to assess evidence before interpreting performance; this skill does not connect to a broker or run trades.
---

# Backtest review / 回测结果核查

Review the report the user supplied. Preserve their research question and distinguish a calculated result from a claim about a strategy.

- Identify the instrument, currency, data period, frequency, data version and the time each input could have been known. Missing fields remain unknown.
- Reconcile the signal timestamp with order submission and executable prices. A close-dependent signal needs an explicit assumption before a same-close fill can be accepted. Inspect partial and rejected orders when available.
- Identify fees, spread, slippage, financing, leverage, cash flows and sizing. Do not compare returns with different capital or cost definitions as equivalent.
- Separate development, validation and final holdout evidence. Repeatedly consulted test periods are no longer untouched. Do not consume a protected holdout, rerun live trading or change risk controls merely to review a report.
- Inspect neighbouring parameters, time segments, adverse conditions and failed attempts when supplied. A selected winner or a smooth curve alone cannot establish robustness.
- Report what is directly supported, what is inferred, what is missing and the next test that would resolve a specific uncertainty. Include sample size and limitations. Do not invent metrics or certify profitability.

For synthetic examples, state that the data is synthetic and restrict conclusions to calculation mechanics. For account statements, do not describe a self-calculation as an independent audit or GIPS verification.

Output a concise review tied to the supplied evidence. A useful format is: claim, supporting record, gap, consequence, next check. Adapt the detail to the actual question.

Source: WeiMind, https://weimind.hk/zh/ai-solutions/research-tools/ . Version 1.0, 2026-10-20. Free public example; personal research and learning use. No resale or paid copy-trading redistribution. No system connection, paid entitlement or investment-service authorisation is included.