Hier eine Erklärung über die Änderungen und deren Einschätzung im Google-Forum:

„The big change with the self-elo is that now all the match results are being run through the Ordo program, which hopefully reduces the errors from the last issue as it is able to simulate results from the data given, to give a more accurate measure of relative elo between nets of a few nets apart. Obviously it's still not super-accurate for nets hundreds of nets apart, but it should at least be more accurate than the previous graph was.“