Some change to the curve and posting/curation split may help matters, but it should not penalise small accounts too much. We know the orcas and whales do well from curation. Then again the system should discourage creating masses of small accounts. It's a tricky one to get right and you will never please everyone. Maybe it could be a variable system that adapts to deal whatever ways people try to game it, but then there will be lots of strategies going on.
One thing that should change is that rep should go up slower if all votes come from the same few accounts. We have people with really high rep just from buying votes, so it is meaningless.
Every measure i can think of has some negative side effect that goes up exponentially as you increase the degree of that measure.
Curation - less $ for authors, assuming honest voting
Free downvotes - toxicity, bullying, possible collusion at high levels
Coverging Linear - punishes smaller votes
The idea is to use as little of these measures as possible while still being sufficient to bring about honest voting behavior from most stakers.
If you can think of other measures that can bring about this change we desperately need with fewer negative side effects, I'm definitely open to them.
Might help to run some actual simulations with historic data. People can explain why they think these measures might work and others like me might explain why we feel these measures will likely backfire, but in the end simulations and causal modelling can answer many questions that pure reasoning can't. And if these give ambiguous results, incremental changes that can be rolled back can help to test the model without being too disruptive. Let's try to avoid an other HF20 incident.
Yes, I'm not against running simulations, of course when the idea is to change behavior it's difficult to get a clear idea of what the new behavior would be under different economic incentives if we just model it on historical behavior, but it'll definitely help avoid something like using the wrong multiplier by a factor of 10 like with RC.
I guess this would help the covergent linear curve the most to help us arrive to some sensible constants in the n^2(n/c+1) curve.
It's obviously possible to not get close to the right numbers with curation and free downvotes too, but it's harder to see how historical behavior can help too much here.
I'dd personally like to see the simulated differences in outcomes between a set of those curves for different c, with a set of something like:
Where V is effective vests and S is a scaling factor somewhere in the 1.01 .. 1.05 range.
We have some benefits to down votes with SteemFlagRewards, which has manual checks to prevent abuse. Any automatic system can probably be gamed.
I don't claim to have the answers. I do okay from rewards, but I see many getting frustrated whilst they see others raking it in (by whatever means). Reducing the benefits of self votes is okay unless people create proxy accounts.
The freedom of Steem is a great feature, but we see it being abused all the time. We can create as many accounts as we want (RCs permitting) so there could also be a landgrab of good names.