Picard attempts to parse the track number from file name if it is not set. For this it uses a set of regular expressions, but if those fail it uses the lowest number it can find in the title.
This leads to many false positives as discussed in
Examples where numbers get extracted as track numbers falsely:
30,000 pounds of Bananas
Symphony no. 5 in D minor
We should look into improving the heuristics here. Maybe just remove the generic case of extracting any number and focus more on typical specific cases like it is already done with the existing regular expressions.