Suppose I have N ideal gas particles of unknown types, but they have the same mass. I then sample a negligible number of particles and conclude I have actually have 3 different colors of particles. That is, my detector measures the color of a random sample of particles (and nothing else) and by chernoff bound from a small sample size I conclude (with high probability) that I have some mixture $ p_{1} + p_{2} + p_{3} = 1 $ so that I have $ p_{1} N $ red particles, $ p_{2} N $ blue particles, and $ p_{3} N $ green particles. Then from Sackur-Tetrode: $$ S_{ \text{before sampling} } = k \, N \left( ln \left[ \frac{V}{N} \left( \frac{ 4 \pi m }{ 3 h^{2} } \frac{U}{N} \right)^{3/2} \right] + \frac{5}{2} \right) $$
$$ S_{ \text{after sampling} } = \sum_{i=1}^{3} p_{i} k \, N \left( ln \left[ \frac{V}{p_{i} N} \left( \frac{ 4 \pi m }{ 3 h^{2} } \frac{U}{ p_{i} N} \right)^{3/2} \right] + \frac{5}{2} \right) = S_{ \text{before sampling} } - \frac{5}{2} k \, N \left( \sum_{i=1}^{3} p_{i} ln \left( p_{i} \right) \right) $$
So recognizing some feature of my particles, regardless of how relevant to any of the physical properties of the particles, seems to change the entropy. If entropy has physical consequences, shouldn't there be a way to know if to regard or disregard properties of relevance or irrelevance in a system? Otherwise it would seem that inconsequential features may change my physical understanding.