MCEC anyone?

[amanjpro]

I was thinking, it will be probably nice to have a a "Medicore Chess Engine Championship", and being broadcasted on Twitch just like TCEC.
Probably for engines in the range of 2000-2900 or any other complimentary rating to TCEC cutoff. Cool thing about this competition is that, the games are way more accessible to our brain capability than say Stockfish and LC0. It might even be able to outcompete TCEC in popularity because of that.

It also encourages engine writers to make their engine better without the need to copy from Stockfish or any other top-engine (more originality?). Or, probably I am overthinking it.

[mvanthoor]

Actually, that isn't a bad idea at all. It could be a tournament with a few divisions:

under 1500
1500-1999
2000-2499
over 2500 (but under TCEC's cutoff Elo)

Then, an engine could participate in a tournament at each level of development.

I'd suggest:
- No derivatives
- No clones
- No NNUE

So, a tournament for new, classical engines written from scratch. (NNUE is excluded because, if you add it to a ~2000 engine, it will gain so much strength that it will be over 2900 already. Maksim's BBC has already proven this.)

I have no idea how to organize such a tournament though, without renting a big server somewhere.

[hgm]

A decade ago we had ChessWar, which was the best chess-engine competition the world has ever seen. There were live broadcasts, and many divisions. Virtually all existing original engines participated. Too bad Olivier discontinued it.

[amanjpro]

I have no idea how to organize such a tournament though, without renting a big server somewhere.

I don't think you need a big server, most of the engines at this level don't even have multi-core support. Maybe we can limit 4 cores/threads per engine, and one game at a time.

[hgm]

I used to broadcast tournaments like that for chess variants (Capablanca Chess, Knightmate, Shatranj, Spartan Chess, Xiangqi).

Through a page somewhat like this.

[amanjpro]

I used to broadcast tournaments like that for chess variants (Capablanca Chess, Knightmate, Shatranj, Spartan Chess, Xiangqi).

Through a page somewhat like this.

That is actually very cool...

[amanjpro]

So in practical terms how can we proceed with this. What are the setup, format, resources that we need.

At first we can start by a few engines representing different ratings and then add to it in later editions

[hgm]

It depends. The simplest way is to run the games on the web-server. That is what I did when I was broadcasting. Run a tournament under WinBoard, which then saves the ongoing game on a file, and at the same time run a web server there that opens the machine to the outside world. And contains the viewer page, which then access the game file. With many viewers the upload speed of your internet connection could become a limitin factor, however.

An alternative is to use a webserver somewhere in a data center, and upload the games you are playing on your PC at home continuously to that server.

[amanjpro]

I have a host with infinite bandwidth, that I can use to serve the website. If one of us has a spare desktop that he can use to run the tournaments, then we can start with that. The desktop runs the games, pushes the pgn to my host, we run pgn4web on the host and viola we will have a basic tourney

[niel5946]

I'd suggest:
- No derivatives
- No clones
- No NNUE

That list of rules seems good, but I have one issue with it. NNUE doesn't necessarily make an engine 3000+ elo, SF NNUE does that. Therefore I think you're excluding too many developers (like me) who are developing their own network implementations and training their own nets. I think an improvement could be:
- No derivatives
- No clones
- No SF NNUE evaluations. Neither the nets themselves nor the implementation.

Neural networks, or at least a hybrid between them and HCE's, are the future of evaluation functions, and I think that should at least to some extend be acknowledged.