Started replaying every live NIFTY trade through the backtest engine the next morning
One issue that was driving me crazy was seeing a perfectly clean backtest and then getting a different trade live even though the strategy code was basically the same.
So now every morning I take the previous days live logs and feed the exact timestamps back through the strategy engine. Im storing the broker candle, signal candle, order timestamp and actual fill separately. That finally showed me a few cases where a 1 minute candle from my historical source was off by just enough that the backtest never triggered while the live feed did.
I also keep Moons Play Money as a separate sanity check whenever I change the general entry logic. Not for NIFTY data, just to test whether I accidentally broke something obvious in the signal flow before reconnecting the code to anything that can place real orders.
The strategy itself barely changed. Most of my work lately has just been trying to make backtest, paper and live all behave like the same system.