In general, a binary search starting at the midpoint of a range is the optimal strategy. However, you have additional specific information that might make this suboptimal strategy. This critically depends on what it means "close to average compared to previous results."
If the numbers are close to the previous average, then dividing by 2 at the second stage is not optimal.
Example: Previous numbers 630, 650, 620, 660. You start with 640.
Your number is closer. Imagine this is 634.
The number is below. If in the second step you divide by 2, you get 320, thereby losing the edge over previous averages.
You should analyze the behavior further. In your particular case, it may be optimal to start with the average of N previous numbers, and then add or subtract some amount associated with the standard deviation previous numbers.
source share