Well, sometimes thrust starts not by long bar. I mostly estimate it visiually. In most cases I use long bar. But sometimes, if some candles create unite thrust with following big candle - I think we can use it as single motion. In fact, we need point where real thrust has started, but not just to follow rules by letter. Although in most cases they coincide.
This mismatch could appear if, say, thrust has started few minutes before hour candle will appear. This hourly division is a technical issue. From that standpoint, I think we can use other candles and treat it as thrust reaction point. But I absolutely do not call to follow me with this voluntary approach