One thing that appears to be true is the ranking of the outputs are rather similar.
While they are not identical, they are within a reasonable margin of error.
This could be due to a number of causes, my implementation could not be calculating the cosine similarity the same way as Lucene.
However, the fact that the outputrs are relatively similar is reassuring.
Furthermore, Lucene could be weighing certain terms differently than I am.
Lucene could know that "the" for example, can be dropped from the query or removed from the results entirely.
In my implementation, this is not done.
Another important thing to note is that in QueryID 4, there are plurals and singular versions of the same words:
"bulb" - "bulbs" and "alternative" - "alternatives"

The stem of these words could be obtained and the outputs would be much different.