The performance I posted is of the author of the strategy through their own backtest with only 2 years of data from polygon.

I would encourage you to run my code with Zorro's M1 data, and you will see vastly different results, besides the obvious fact that you would be testing different periods (Zorro 2011-2023 vs 2022-2024) I am not sure if my code is running with the same logic proposed by the author, specifically the VWAP calculation which Zorro uses, in the paper it looks like their VWAP is calculated with the daily open to close while Zorro's uses a fixed period (I.e. the last 5 bars)

I would really appreciate if someone could look over my code and point out any flaws regarding the logic.