How about the first 7 matches an algo plays are for calculating your elo, those matches are played from places very high and very low and then after those 7 matches you get your elo rating. Here is an idea of how the calculating system could work:
first, you play a match from an extremely low algo and an extremely high one. then based on how many rounds it took, whether you won, your health after winning or their health after losing, you determine whether it is closer to the high elo or the low one, giving a very crude estimate. you then play 2 matches, one 250+ elo and one 250- elo, rinse and repeat for 7 matches, the final one being very close to your estamiated elo to check if it is close or not.
I do not see this being any better than the current system.
You could just be unlucky and thus take way more time to climb up the ladder (being unlucky as described here) or get a false demotivation from your crude estimate.
I feel like the current system is totally fine because winning all your matches will bring you to the top no matter what and everything else can take its time and just be fine because the ranking does not really matter (you can watch your matches to see how you are doing).
In addition, from what I gather this system stops calculating your elo after those first matches. If I assume incorrectly let me know. I do not feel this would accurately evaluate a span of matches, primarily when competing against future algos that are uploaded. The newer algo would be ranked, but the older ones have no opportunity to do better/worse, only maintain their position and essentially defend their title.
I personally feel this gives older algos a disadvantage. But really, I think the biggest problem is that there aren’t enough matches in this model. I strongly agree with this:
In addition to what others have mentioned, the ONLY factor the Elo system should look at is who won and who lost. If we gave more Elo to players who won with more health, defensive strategies would rise even with lower winrates. If we gave Elo to players who won quickly, we would encourage ‘aggressive’ strategies. Looking at factors other than winrate allows users to game the system by trying to optimize secondary factors, even at the cost of maximising winrates.
We are happy with how quickly our Elo system moves Algos. We had an event and the winner of the event reached the top 15 within an hour of being uploaded, though movement slows significantly at high ranks. We are planning to investigate matchmaking to ensure high ranked players are not being matched against players ranked too low to give them any Elo for a win
But lets say, hypothetically, the best win rate strategy involved passing on turn one as discussed in other threads. In this strategy, you would be letting yourself take 5 extra damage, in the hope that the flexibility in how you spend your first cores makes the trade off worth it.
This strategy would often win in the same number of turns as strategy B, but with 5 more health. Because you are losing health using strategy A, people realize it is causing them to lose Elo, and use the weaker strategy B - and are displayed as ‘better’ because of that choice.
No algorithm or combination of stat tracking will ever more accurately predict success than monitoring win rate, and only win rate, directly.