Some great points here, thanks for sharing your thoughts.
We’re trying to balance the amount of Algo online (and the amount of nodes which we also want increased), with the cost of operating nodes for a large amount of stake.
The paper is exploratory, so these are not hard decisions, but the lower we make the cap, the more expensive we make global staking operations.
Ensuring we don’t have a stall is primary, and thankfully, as per point 4.5.1 in the paper, we have some protection:
The absenteeism mitigation approach effectively provides ”Garbage Collection” for consensus, permitting relatively significant portions of global stake (5% to 10%), within a reasonable rolling time period, to ungracefully exit consensus ad-infinitum without detriment to the network.