I'm running a Meta model in production. I have it set up as a backtest over the past several months (it uses some indicators that require that much lookback). The trouble is that the backtest is not precisely repeatable. So one day I place some buy orders (ex. Buy AAPL 24 shares at market open), and the next day the backtest produces some sell orders that don't precisely match the open long positions (ex. Sell AAPL 22 shares at market open). I have several models running on the same universe of stocks, and it can be very confusing to figure out exactly which open positions to sell. In most cases it is not critical which gets sold, but it is inconvenient that my signals do not align with my open positions.
Request: either advice to help me find a way around this, or perhaps we could have a more deterministic backtesting mode, so that I can somehow get more accurate signals for closing positions that match the number of shares I have open.