I would say the life cycle is correct. Though I never stop 3 and 4 if possible, I just add 5. It helps to know the differences and try to eliminate them via adjusting slippage or fees.
The docs describe QC's paper trading as:
"See how your algorithm would have performed with our paper trading feature. We use real live-data feeds but a virtual brokerage to execute your trades. Each project is allocated $100,000 virtual currency to track how you've performed."
This QC paper step is very important. Moving from the backtester to a live server/data may have some unexpected surprises to debug. One example: I didn't realize a compute process would take 5+ minutes due to a slight difference with History speed. So my on-open orders were over 5 minutes late due to my own bad planning (fixed by simply moving compute stuff to pre open). You want to perform a hardcore review to ensure your live-data algos match the intentions of the backtest and how it was undertaken.