A very high trade-count backtest with a 96.8% win rate and PF 2.54, but with repetitive same-timestamp multi-ticket clusters and large recurring stop-out losses that require out-of-sample and forward validation before any confidence.

数据透明度

核心指标

增长

以导入报告的起始资金归一化为 100。该图按已保存的逐笔结果采样，不等同于经纪商实时权益曲线。

详细计算结果

收益率 = 净利润 ÷ 初始资金 × 100%；PF = 总盈利 ÷ 总亏损绝对值；单笔期望 = 净利润 ÷ 已平仓交易数；平均盈亏比 = 平均盈利单 ÷ 平均亏损单绝对值；最大回撤采用报告输出的最大回撤值。若报告含入金或出金，收益率应结合现金流另行解释。

策略安全评分依据

• Verified positives: 2413 closed trades, PF 2.54, return 317.4% on a 1000 base, max drawdown 34.0%, win rate 96.8%; profit is not concentrated (top 1% of trades = 6.3% of gross profit, top 5% = 21.3%; remaining PF stays 2.0-2.38).
• Repeatable multi-ticket clusters: many identical size 0.01 tickets open and close at the same minute and price, e.g. rows 1157-1162 all +3.19, rows 25021-25026 all -25.07/-50.14, rows 213123-213129 all about -50, indicating several positions taken as one basket at one timestamp.
• Clustered exits produce large simultaneous losses: -25.07 x5 plus -50.14 and -24.06 on 2024.05.09; about -24.85 x6 plus -23.84 on 2024.06.07; about -25.10 x5 plus -50.20 on 2025.03.05; about -50.30/-75.45 x3 plus -48.96 on 2026.02.19; the same-minute sequences are repeated multiple times across the record.
• Loss magnification: same direction, same time, and the 0.02 tickets have exactly double the loss of the 0.01 tickets (for example -50.20 versus -25.10; -50.30 versus -25.15), so losses scale additively across the cluster at a single event.
• Same-bar/same-minute opens and closes appear throughout (open equals close timestamp on almost every sample row), an execution assumption to verify, not by itself proof of invalid data.
• Equity swings are uneven: a run from about 1358 to about 1609 is followed by a drop to about 915, and later from about 3453 to about 2254, so gains are not monotonic and basket stop-outs materially reset the curve.
• Hard evidence of tester quality warning: 24 mismatched charts errors are reported for this run.
• No commission, swap, or fee fields are present in the supplied context; cost sensitivity is therefore an open question.
• Exit type labeling is mostly s/l while many of those exits show positive profit, characteristic of a trailing-stop/breakeven exit style rather than open losses.

关键风险信号

• Clustered same-timestamp basket entries with correlated losses: the 2024.05.09 sequence opens around 17:18 with about five -25.07 tickets plus -50.14 and -24.06, removing roughly 6-7% of account value in one minute before the next recovery.
• On 2024.06.07 a similar cluster of about six -24.85 tickets plus -23.84 takes the balance from about 1088 to about 916 in one minute, a drop of roughly 16%.
• On 2025.03.05 approximately four -25.09 tickets, one -50.18 ticket, and one -24.42 ticket drain the balance from about 2104 to about 1904, roughly 10% in one minute.
• On 2026.02.19 three large same-minute tickets of about -50.30 and -75.45 plus -48.96 remove roughly 9-10% of account value in one minute.
• If cluster direction is correlated, simultaneous stop-outs scale linearly with ticket count rather than being independent; the 0.02 ticket magnitude being exactly twice the 0.01 loss confirms additive exposure within the cluster.
• Test-quality flag of 24 mismatched charts errors means the reported metrics could differ on the intended symbol and data set.

现有数据无法判断

• Exact instrument, broker, and chart symbol are not provided.
• Commission, swap, spread, and fee treatment are absent, so net-of-cost behavior is unknown.
• Order logic, cluster sizing rules, stop distance, and whether the 0.02 tickets are a scaling grid or part of a basket cannot be confirmed without the full backtest settings.
• Minimum holding time, stop level, and tick model are not shown.
• Whether the mismatched charts errors materially changed the symbol, timeframe, or bar data is not stated.

下一步行动

• Re-run the test with the exact intended symbol and a clean chart history to resolve the 24 mismatched charts errors, then compare summary metrics.
• Export the backtest settings and add commission, swap, and spread assumptions to test whether the clustered multiple-ticket losses remain survivable after costs.
• Analyze the full trade list for same-timestamp ticket clusters to measure the worst simultaneous loss and confirm whether cluster direction is correlated, which would clarify the exposure risk.
• Run a forward or demo sample over the same instrument and conditions to compare live vs backtest fill quality and cluster behavior.
• Stress-test position sizing and cluster rules under degraded conditions, such as wider spreads or slippage on the cluster exit timestamps.

常见问题