Please change the way of playing test matches

If devs can change the way of playing test matches, it can make the lives of testers much easier.

1. Let us call a given latest network A1, and appoint this as the benchmark network. Then, let all the following candidate networks (A2, A3, A4, etc.) play test matches against the benchmark, not against the one immediately preceding the candidate.

2. Only when a candidate defeats the benchmark with an elo difference of at least, say, 35, the former will become the new benchmark, called B1. Then, all the newer candidates (B2, B3, B4, etc.) would play matches against B1. And so on.

3. If they wish, they can throw in extra matches between different benchmarks, e.g., between A1 and E1, to make the self-play elo more realistic.

A few lines of server-side code suffices to implement this, but there are very clear benefits. With this implementation, not only would the "rock-paper-scissor" issue be made much less severe, but also would the work of testers be much easier, given that they need to concentrate on the benchmarks only, and afford to ignore others.

(I have said the same thing on Discord, the forum, but no one has given a serious thought. But I hope a server admin may notice if I say it here.)

No file or test is named. Start by locating the server-side test-match scheduling entry point and compare the current pairing flow with the proposed benchmark progression. Done means candidates use the latest qualifying benchmark, with the proposed 35-Elo threshold and optional cross-benchmark matches covered by tests.