Giulio giulio@anche.no writes:
If I assume that the majority of witnesses are honest and take the median time (as rgdd suggested), that should give me a reasonably accurate timestamp, ensuring that the logging operation couldn't have happened before then. In practice, for our use case, we just need to make sure that content cannot be future-dated, and it seems this approach gives us exactly the properties we want.
Duplicates may also be relevant. Entries could always be submitted to additional logs at a later time.
Entries cannot be resubmitted to the same log, provided we remain strict with duplicate detection. There have been discussion for other logs that duplicate detection could be more of a best-effort property, and I think we need to keep this usecase in mind if we consider relaxing that for Sigsum.
Regards, /Niels