
StratProof 交易策略回測 | Backtest & Overfit Check
每個回測工具都會給你一條漂亮的權益曲線。StratProof 會替它打分數。
Every backtest tool shows you an equity curve. StratProof grades it.
用白話描述你的策略就好——TradingView、YouTube、ICT/SMC 或你自己的筆記都可以,不需要會寫程式。StratProof 會把規則改寫成精確條件、寫成 Python 策略、用真實的手續費、交易稅與滑價回測,然後想辦法把它「打破」。
Describe your setup in plain words — from TradingView, YouTube, ICT/SMC or your own notes. No coding needed. StratProof restates your rule in exact terms, writes it in Python, backtests it with realistic fees, taxes and slippage, then tries to break it.
【你會得到 / What you get】
- report.html — 一頁式報告(報酬、最大回撤、勝率、獲利因子、權益曲線),中英文皆可 / one-page report in Chinese or English
- strategy.py — 可執行的策略程式,歸你所有 / the runnable strategy, yours to keep
- trades.csv — 每一筆交易紀錄 / every trade, entry to exit
- 判定 Verdict — 穩健 ROBUST / 可能有效但脆弱 FRAGILE / 很可能過度擬合 OVERFIT / 扣除成本後沒有優勢 NO EDGE,附理由 / with the reasons
【如何確認優勢是真的 / How it checks the edge is real】
- 樣本外檢驗 Out-of-sample — 在沒用來調參數的資料上還有效嗎? / Does it work on data it was not tuned on?
- 參數敏感度 Parameter sensitivity — 參數稍微改一下就失效嗎? / Does a small change kill it?
- 蒙地卡羅回撤 Monte Carlo — 連續虧損最糟會到哪裡? / How bad can the losing streak get?
答案是「不行」時,它會直接說「不行」——在你投入真金白銀之前。
It is built to say "no" when the answer is no — before you risk real money.
【台股成本照實計算 / Taiwan costs, counted honestly】
手續費 0.1425%、賣出證交稅 0.3%(ETF 0.1%)全部計入。很多在美股能賺的短線策略,套上台股成本後就不賺了。
The 0.1425% fee and 0.3% sell tax (ETFs 0.1%) are included. Many short-term strategies that look profitable on US stocks stop working once Taiwan costs apply.
【支援市場 / Markets】
台股上市櫃(2330.TW、0050.TW)、美股與 ETF、外匯、幣安現貨與合約,或你自己的 CSV。
Taiwan listed/OTC stocks, US stocks & ETFs, FX, Binance spot & futures, or your own CSV.
【使用方式 / How to use】
需要 Claude Code 與 Python 3.9+。把資料夾放進 ~/.claude/skills/,直接描述交易想法即可,第一次執行會自動安裝套件。使用你自己的 Claude 帳號執行。
Requires Claude Code and Python 3.9+. Put the folder in ~/.claude/skills/ and describe a trading idea; the first run installs the packages. Runs on your own Claude account.
本工具僅供研究,不構成投資建議。回測績效不代表未來結果。
Research tooling, not financial advice. Past backtest performance does not guarantee future results.


