- please see my reply to this post. A lot of the thoughts here are incorrect, my apologies.
The white paper states
You can see this cap in the actual contract code
I'm trying to interpret your printing risk comment and see three possibilities (in order of probability):
1. You are referring to numerai printing new NMR after burning it during a previous tournament
This seems likely to me. Burning NMR every tournament could make it scarce. I think nothing prevents numerai from printing up to 21m NMR. If it is profitable to do so, why wouldn't they? Does this create a perverse incentive for numerai to steal NMR from participants by submitting false scores to the smart contract.
2. You think there is no real cap on the NMR.
If this is the case then please see the smart contract code indicates a hard cap at 21m.
3. You think a new numerai will somehow circumvent the smart contract cap.
I'm not sure how they would do this. Maybe print a new coin or something. This seems unlikely to me
When it comes to the value of NMR I think this is proportional to the prize pool and the probability of submitting a winning algorithm. A few scenarios which I haven't spent much time thinking about (maybe you can improve them)
trivial success P(success) = 1.0
If it is trivial to submit an algorithm with logloss < ln(0.5) then all players submit an infinite confidence and stake all their numeraire. In this case the payout to each player is 0 (assuming multiple participants).
trivial failure P(success) = 0.0
In this case no players ever submit an entry and numeraire value is 0.
uniform random success P(success) = U(0,1)
I take this to mean that probability of success is uniformly distributed and not dependent on prior testing. In this case the tournament becomes a lottery where each submitter has an equal chance of winning. This is where it gets interesting and I don't have time to work out the math.
The white paper addresses this question of value by considering confidence separately for each participant. They say that for a player with probability of success = 1 the value of numeraire is the net present value of all future stake payouts (confidence / stake). I think this ignores the competition aspect of the tournament but I'm not sure (i.e. with 1 participant who has P=1 this statement would be the net present value of all future tournament payouts, but with 2 omniscient participants there is a tragedy of the commons problem where each player must set confidence to infinite.)
I think data scientists are also making NMR during tournaments? This wasn't clear to me in the white paper, but some emails made it seem so.