The appropriate answer is Analyze and report both analysis, full and trimmed data set.

Trimming data is defined as selecting data to make results look better. Cooking data is defined as creating a set of observations that will produce a known result, so this experiment appears to be a case of trimming data. The accepted research practice in this case calls for the researcher to present both sets of analyses, the full data set and the trimmed data set. This is especially called for if there is some known or suspected reason calling into question the measurement validity of the trimmed observations. A clear and full statement describing the procedures and justification for deleting the observations should be included, more prominently displayed than in a footnote.