It’s even more complicate. The satellite cannot audit every single piece (do you remember how much time it takes just for filewalker executed by the local process not requested remotely piece by piece?), so it uses a probabilistic model: pieces are audited randomly, and if the certain amount of audits are failed, it will assume that it cannot trust this node anymore. Just this number of failed audits roughly translated to an amount of lost data, see details there:
2 Likes