including items in final version of new instrument

I have 2 questions I would like advice on. They relate to a paper I'm reviewing for a journal.

When developing a new instrument, if you do a factor analysis and find a factor that explains 45% of the variance, can you then include in your final version of the instrument items that didn't contribute to that factor, i.e. that didn't meet the loading threshold? I thought you would just inspect the scree plot and drop those items that didn't contribute.

Secondly, if you have an item in the instrument you are testing, and it doesn't load, can you justify, theoretically, including it in the final version of the instrument? I thought you couldn't, or shouldn't, but in this case the author says the item is "known to be important" (in this case to self efficacy).