Any feedback is welcome.
My real feedback: I personallh despise +/- voting systems to measure reputation, these systems can be gamed, you'll have to actively moderate fraud (if possible) and sybil attacks.
My suggestion, 1) just collect transactional and contextual telemetry and let clients decide how to measure reputation. 2) Similar to #1, require contextual inputs for transaction. Context about the user, the voter,the nature of the post or transaction (tags?) and output a network verdict but still allow clients to override network metrics. For example, a client can distrust or give a 10x multiplier for votes from a poster/user, or add a 10x multiplier on negative reps for certain types of transactions.
IMO, if you make reps a client only feature,you can adopt to threats and come up with better rep measurement by simply updating clients with a new version without worrying to much about how many people use a certain version.
> the initial reputation for the first author becomes negligible as time goes
curious, how do you handle automated author creation and reputation farming?