How did you make your data more error free then if you collected the data for yourself ? What about latency when you collected themself, because a few 100ms could change the game on 1 second bars. And how much data with how much time span did you collect ? Just to be sure you can get a meaningful backtest.
Exchanges report trades with a nanosecond timestamp. Latency is not an issue when collecting the history of trade data.
If you only work with limit orders how do you assure of being filled with the size you want when doing the backtests ? How could you do that without level 2 data and orderbook as past history ? I mean your assumptions could fail in reality or did you test your strategy in real so far ? If yes, how long ? what have been your results here (in comparison to backtests) ?
This kind of trading is extremely tricky to backtest accurately. You did not specify how you simulate fills but the claimed returns are so large I would tend to think your assumptions are probably way too optimistic. Orderbook data was not mentioned, but if you are using limits then queue position is very relevant for liquid stocks. Additionally, 1 second update speed is way too slow to avoid adverse selection. You said that you are doing 2-3% return from $250K and 65-70 million volume. Let's break that down. Assuming the volume is for both sides (the most charitable interpretation for you), 70 million dollars is 140 round trips of $250K, which means your average return is about 1.8bp per round. That works out to less than 1 cent on a $50 stock. Unless you used conservative fill assumptions, your whole profit could be in the margin of error.