[Kungfoo]

So much for my plan to come out of nowhere to blitz-game my way to HoF dominance.

Couple of issues tend to crop up in ELO settings - 'sitting' on higher ranks (as you mentioned); increased stratification between 'tiers'; and the propensity to avoid matchups that you can lose. These problems tend to be exacerbated when the sample size is small, which makes it probably not the best fit for dom3 (way fewer matches between way fewer players than something like chess or starcraft).

I think you need to be careful about how you go about punishing people for losses. I mean, you want to create an incentive for people to play more games, not fewer. Probably better to go about giving more bonus points for a win, or especially a win against strong competition, than to start removing points for losses - the latter option would seem to disincentivize participation in more competitive games, which is probably not what you want. Adding a metric to track win% and number of games played is probably a great start, and if penalties are to be awarded, let them be for dropped/abandoned games, stales, and stealth-AIs - as those are the behaviors you really want to avoid.

As for how to encourage/reward playing larger games? Well, maybe take some sort of formula 1-style points system wherein the top half of finishers receive some credit, with the overall winner getting a disproportionately-large share? In an 8-player game this could be something like 1pt-4th, 2pts-3rd, 3pts-2nd, 7pts-1st? and a 16-player game would be 1pt-8th through 7pts-2nd, with 15pts for first.

You could then probably do something like aggregate points score/# of total games played to arrive at sort of a general 'strength' score, and then compare your win % against the avg. strength score of your combined opponents to get a fairly decent shot at how you stack up overall.

Couple of problems that jump out immediately are that this doesn't address or account for 'cliques'; ie, if there are no common opponents between two players then the rankings are devalued somewhat... but that's a difficult problem to avoid without adding weightings in at the start of any rankings system, which kinda impugns the fairness aspect (not that a wraithlord or thedemon isn't deserving of a higher ranking off the bat, but that should be evident from the stats alone). A better solution might be to have a round-robin style tournament or two with participation from both the vets and some of the newer members of the community, providing a decent set of common opponents and cross-game samples, as well as some fantastic trash-talking if anyone's able to pull an upset.