Hi,

Cloud masks unfortunately are never 100% accurate. There is usually confusion between clouds and several land cover classes. For example, bright pixels in urban areas, snow and ice and bare rock can be easily confused with clouds by cloud detection algorithms. A quick search will show you there is a whole academic discipline looking into cloud detection algorithms in satellite imagery.

In addition, it’s important to note that the maxCloudCoverage filter sets the upper limit for cloud coverage in percent based on the precomputed cloud coverage estimate for each Sentinel-2 tile as present in the tile metadata. As this is set for each tile, the percentage might not be directly applicable to your chosen area of interest. For example, a tile may only be 20% covered by cloud but your area of interest maybe in the exact area where this cloud is and be 100% obscured.

Hope that this helps answer your query