1. the paper only considered the effect of decimal places. It it intuitive to image that the rounding to whole number will lead to worse results in the estimation of theta. But I guess there would be larger bias while experts rate score to two (or even three) decimal places than just considering the effect of using how many decimal places. It may come from the rating inconsistency.
2. It is unclear how many items are used to conform a item cluster in this study? I think it does matter.
3. If item parameters are not assumed known, it will be confused when we are trying to separate the rounding effect from the raw cut scores. Because the error of item parameters is varying with the rounding effect, they may compensate to each other while estimating them.