Will be latencies measured once again, or the submission latency is the final one? I ask, because we observed that the reported times can vary significantly (~0.6 ms) between the submissions of the same model, and it seems, based on the metrics, submission latencies and provided final score...