you bring some very good concerns about ranking and reputation. We had the same questions in mind when we designed these into Knoks platform. We try to look at ranking and reputation as two parameters and how they should affect the signal provider and their signals. One defines the min and max price for signal in a certain group (rank) while the other can affect where within that rank the price will be.
I will call
the person who shares the signal an
OracleThe first idea that comes to mind is... what if... you add a third parameter, that measured the rate on which an Oracle receives ratings, that is basically hidden.
We decrease the scale of the original example. This would work something like this:
The platform has 100 users when the
Oracle joins. And after the first signal is provided by the
Oracle, if 10 more people join the network, and those 10 new people give the
Oracle, a 5 star rating on it's first signal, the system keep that
Oracle in a database dedicated for red-flagged
Oracles.
Obviously, this can't be implemented when the platform has only 20-30 users for obvious reasons.
The database keeps in store a total of, lets say, 3 red-flags. After the 3rd red flag, it's obviously an issue. This way, even if it's a pure coincidence that the first 5 people who joined immediately after the
Oracle, this can't happened 3 times in a row.
Obviously, this implies a very strict emphasis on statistics, so the system doesn't automatically activate at any scale (especially in the beginning).
Again, the red flag database should be hidden, because the scammer should not know he is being watched... that would make him bail before he faces the consequences.
Your idea is quite interesting. We had a lot of thoughts about that topic, and we were thinking to apply something similar and also a dynamic weight of the users feedback. Another flag is if users (especially new users) purchase signals only from one signal provider (we call them knoksers).
High level speaking, we have this general approach that the less we know about someone, the less weight they'll have. I am talking
only about pure statistical data such as success rate, number of purchases, etc. So eventually, once we'll have enough users activity, we can start applying these models.