Cloud mask of 'manual run' differs from Theia cloud mask

[octoflar]

This is rather a question than an issue. We have processed about a single month of data (tile 32UME, 03 March / 07 April 2020, 10 L1C products in total from both S-2A and S-2B) with MAJA using your start-maja script and everything has gone fine.

However, when comparing the cloud masks in the outputs of the manual run with the cloud masks in corresponding Theia products, we see that the cloud masks differ.

The attached images shows a tile for 28 March, one for the 'manual MAJA' output (top image), the other from obtained from Theia (bottom image). Black = any cloud mask bit set. Could you explain the reasons why the cloud masks can be different? We have used Version 3.3.2 without the CAMS option.