We are the Computer Chess Community. We control ourselves.

[AndrewGrant]

The idea that a few unhappy people can bully rating organizations into supporting their (sometimes extreme) views is really funny. 'Boycotting CCRL', (just to name an example), how would you do that? What does it even mean if you vow to 'ignore it' yourself? Would you make a fool of yourself by claiming a strength for your engine that most others would see to be completely off? How would anyone care if you didn't look at a certain rating list? And in particular, why would those who create such lists care? It is not like you pay them a fee per look. You can be sure that any noob who bumped into a chess engine and wants to know its rating is directed by Google to CCRL...

Its not bully, its abandoning. I won't provide a version to testers who don't meet some standard I define for myself. Obviously no individual has the power to compel a group to do something. The point of the post was that as a collective, we do have that power. And if we reach a consensus that stolen engines should not be rated, then that can be reflected in the rating lists.

CCRL is the gold standard, not because they are doing anything right, but because there was a collective idea that CCRL had the best data to offer. That can easily change. XYZ can become that new standard. Maybe they do it by meeting the things I would like to see. Maybe they do it by providing greater games, longer time controls, better hardware, better data discovery, etc.

You see this as some war against an individual entity, or set of them; but really its a very basic statement: Don't support people or groups that do not support you.

[towforce]

Fast food is for fat people who die soon.

And die happy? Want my brussel sprouts?

This is OT, so VERY briefly: if you drink nothing but water, you'll find, to your surprise, that you'll start enjoying drinking water.

Oh, and it's largely about the genetics.

No. It's easy to show that lifestyle makes a big difference to health and longevity in most cases.

[gaard]

The idea that a few unhappy people can bully rating organizations into supporting their (sometimes extreme) views is really funny. 'Boycotting CCRL', (just to name an example), how would you do that? What does it even mean if you vow to 'ignore it' yourself? Would you make a fool of yourself by claiming a strength for your engine that most others would see to be completely off? How would anyone care if you didn't look at a certain rating list? And in particular, why would those who create such lists care? It is not like you pay them a fee per look. You can be sure that any noob who bumped into a chess engine and wants to know its rating is directed by Google to CCRL...

Its not bully, its abandoning. I won't provide a version to testers who don't meet some standard I define for myself. Obviously no individual has the power to compel a group to do something. The point of the post was that as a collective, we do have that power. And if we reach a consensus that stolen engines should not be rated, then that can be reflected in the rating lists.

CCRL is the gold standard, not because they are doing anything right, but because there was a collective idea that CCRL had the best data to offer. That can easily change. XYZ can become that new standard. Maybe they do it by meeting the things I would like to see. Maybe they do it by providing greater games, longer time controls, better hardware, better data discovery, etc.

You see this as some war against an individual entity, or set of them; but really its a very basic statement: Don't support people or groups that do not support you.

Suppose someone were to start a new rating list that did meet your standard... what would their rationale look like for the inclusion or exclusion of an engine? Do you think a ratings group should test NNs separately, for example? What if my SF net was weaker than the default, but I optimized (say) 10 parameters within the search that made up for that, for example? I'm sure we'd agree to 95% of the details, but it's the remainder that are hard to quantify and find agreement on.

[towforce]

In the 1980s and 1990s, the "Selective Search" computer chess magazine used to have its own proprietary list, and that used to be the one that I respected the most (the magazine also published selected parts of the SSDF list IIRC). I never heard of any disputes between SSDF and this proprietary list.

Doing something like this sounds time consuming, but once you've set up your Google Spreadsheet (in which you get to choose which pages are published), it won't really take long. If it were me, I'd have a master list, and for each program on the list, a selector to determine which list it would appear on, and a bit of script to do the copy/paste operations.

Billboard magazine recognised that music needed many categories of chart (link), and the sheer number of chess programs means that the same now applies to us.

[AndrewGrant]

Suppose someone were to start a new rating list that did meet your standard... what would their rationale look like for the inclusion or exclusion of an engine? Do you think a ratings group should test NNs separately, for example? What if my SF net was weaker than the default, but I optimized (say) 10 parameters within the search that made up for that, for example? I'm sure we'd agree to 95% of the details, but it's the remainder that are hard to quantify and find agreement on.

1. "Pure" lists, meaning that only the most recent (not best!) version of a given engine is used for computing elo values. Namely, suppose your list has SF11, SF12, and SF13, all with a thousand games against Ethereal 13.00. Only the SF13 vs Ethereal 13.00 games should be used in the elo calculation. Otherwise you are skewing the results towards an engine pool where Stockfish is over represented. CCRL used to have this option, minus the (not best!) caveat.

2. Only one version of a given engine is on the "main" list. Obviously engines update over time, and its okay to track that data forever. However, those "old" results are not apart of the elo computation. Additionally, lawful derivatives like Sugar, Shashchess can be rated at your discretion (although I suggest against it, or not having them ever as the "best" version), but only one should be shown. Unlawful derivatives like Fire and Houdini should make no appearance on the rating list.

3. You get into a grey area when its not entire clear what is a derivative. IMO, the _only_ engine in a grey area is Allie. All other projects that "Used Leela as a starting point", or "Used Leela's net infrastructure", or "Used Stockfish's search" are cheap clones. You cannot make everyone happy, but the metric for what constitutes an original, although inherited work, should likely include the amount of effort required to produce such an entity.

4. The list should be run on the same hardware. It used to be the case that you could simply scale the time control and get good results (CCRL, CEGT, etc), however now that we have GPU engines, as well as engines which require AVX, AVX2, or higher instruction sets, in order to play optimally, its best that a given list is tied to a particular setup. IE, "Here is my rating list. All CPU games are played on a Ryzen XYZ, and the GPU is an Nvidia ABCD". You always have the downside of an upgrade tossing out the data, but that is how it has to be.

5. Testing conditions should be clear and expressed. Including all the engine options, the book options, adjudication settings, etc. Additionally, one might want to make note of the origin of an engine binary. Ideally a list is run on a Linux machine, which builds open-source engines according to author's instructions, producing faster binaries than what could be distributed. So one should cite whether the binary is from a download at a given URL, compiled on their own by invoking a certain command, etc.

[towforce]

In the 1980s and 1990s, the "Selective Search" computer chess magazine used to have its own proprietary list, and that used to be the one that I respected the most (the magazine also published selected parts of the SSDF list IIRC). I never heard of any disputes between SSDF and this proprietary list.

Doing something like this sounds time consuming, but once you've set up your Google Spreadsheet (in which you get to choose which pages are published), it won't really take long. If it were me, I'd have a master list, and for each program on the list, a selector to determine which list it would appear on, and a bit of script to do the copy/paste operations.

Billboard magazine recognised that music needed many categories of chart (link), and the sheer number of chess programs means that the same now applies to us.

A better idea: offer the complete list with filters so that people could build their own preferred categories.

The extra work here would be the extra columns in a spreadsheet for the information to be able to filter on: Google sheets already has filtering capability (link).

[gaard]

Suppose someone were to start a new rating list that did meet your standard... what would their rationale look like for the inclusion or exclusion of an engine? Do you think a ratings group should test NNs separately, for example? What if my SF net was weaker than the default, but I optimized (say) 10 parameters within the search that made up for that, for example? I'm sure we'd agree to 95% of the details, but it's the remainder that are hard to quantify and find agreement on.

1. "Pure" lists, meaning that only the most recent (not best!) version of a given engine is used for computing elo values. Namely, suppose your list has SF11, SF12, and SF13, all with a thousand games against Ethereal 13.00. Only the SF13 vs Ethereal 13.00 games should be used in the elo calculation. Otherwise you are skewing the results towards an engine pool where Stockfish is over represented. CCRL used to have this option, minus the (not best!) caveat.

2. Only one version of a given engine is on the "main" list. Obviously engines update over time, and its okay to track that data forever. However, those "old" results are not apart of the elo computation. Additionally, lawful derivatives like Sugar, Shashchess can be rated at your discretion (although I suggest against it, or not having them ever as the "best" version), but only one should be shown. Unlawful derivatives like Fire and Houdini should make no appearance on the rating list.

3. You get into a grey area when its not entire clear what is a derivative. IMO, the _only_ engine in a grey area is Allie. All other projects that "Used Leela as a starting point", or "Used Leela's net infrastructure", or "Used Stockfish's search" are cheap clones. You cannot make everyone happy, but the metric for what constitutes an original, although inherited work, should likely include the amount of effort required to produce such an entity.

4. The list should be run on the same hardware. It used to be the case that you could simply scale the time control and get good results (CCRL, CEGT, etc), however now that we have GPU engines, as well as engines which require AVX, AVX2, or higher instruction sets, in order to play optimally, its best that a given list is tied to a particular setup. IE, "Here is my rating list. All CPU games are played on a Ryzen XYZ, and the GPU is an Nvidia ABCD". You always have the downside of an upgrade tossing out the data, but that is how it has to be.

5. Testing conditions should be clear and expressed. Including all the engine options, the book options, adjudication settings, etc. Additionally, one might want to make note of the origin of an engine binary. Ideally a list is run on a Linux machine, which builds open-source engines according to author's instructions, producing faster binaries than what could be distributed. So one should cite whether the binary is from a download at a given URL, compiled on their own by invoking a certain command, etc.

The only ambiguities I can find are in 3, re cheap clones. I am going to be pedantic and possibly contentious, but how many hours must I train a network and producing how many Elo over the default, before a clone is no longer a "cheap" clone? 50 Elo over default definitely seems like a lot of work, but what if by some accident it took me 10 minutes?

Again, I only want to be pedantic to a point. In another way, I am considering starting my own rating list and would like to know what would give the greatest value to the greatest number of users.

[AndrewGrant]

3. You get into a grey area when its not entire clear what is a derivative. IMO, the _only_ engine in a grey area is Allie. All other projects that "Used Leela as a starting point", or "Used Leela's net infrastructure", or "Used Stockfish's search" are cheap clones. You cannot make everyone happy, but the metric for what constitutes an original, although inherited work, should likely include the amount of effort required to produce such an entity.

The only ambiguities I can find are in 3, re cheap clones. I am going to be pedantic and possibly contentious, but how many hours must I train a network and producing how many Elo over the default, before a clone is no longer a "cheap" clone? 50 Elo over default definitely seems like a lot of work, but what if by some accident it took me 10 minutes?

Again, I only want to be pedantic to a point. In another way, I am considering starting my own rating list and would like to know what would give the greatest value to the greatest number of users.

Training a new network using existing tools requires very little, if almost no effort. Reproducing a net for Leela using their pipeline, or for Stockfish using their pipeline, or even for Ethereal's pipline at this point, requires zero domain knowledge of computer chess programming. It does not constitute an original work, nor does it imply that the creator has any particular skillset in relation to chess programming, nor does it require any innovative or original thought.

This means that everyone using a "Stockfish trained" or "Stockfish implemented" Network would not qualify. That goes for Fat Fritz, down to the following if they are still using an Shogi pipeline trained Network: Rubichess, Igel, Nemorino, and the other weaker ones that do not come to mind. Komodo does fall under that catagory as well as far as I can tell, so I would personally not rate Komodo NNUE in its current form, unless the authors can make a statement affirming that Dkappe did not simply run the Stockfish pipeline

[mclane]

The facts remain. Why should an idiot pay 99 euro for something he can dowload in the internet for free without chessbase logo ?!

Because albert silver or frederic friedel said in a marketing campaign that fat fritz II “will change your life” ??

The uneducated mass maybe buys it.

I remember when frederic told me in the 80ies that “chessbase will change my life”.

Now its 2021 and the campaigns are still the same.
And they are plein wrong as they lied in the 80ies.

It is not that long ago you were suggesting Chessbase would be the perfect place to host CTF and that they were your friends:

As I said, if universities do not follow our demands, we can asks companies like
Schach Niggemann or my friends at chessbase in Hamburg.

I could ask Frederic and matthias.

No i was suggesting they host CCC !

And you obviously did not get the irony in my posting.

[emadsen]

Its time that people stand up and become the change that this community so badly needs, and that so many of us would like to see. That means not using ChessBase products.

I stopped using ChessBase software years ago when it became apparent they'd never fix the flicker problem that afflicts their GUIs. Every menu selection causes the screen to repaint itself 17 times before it stabilizes. That alone was reason enough for me to abandon them. It suggests they're building off a 20 year old codebase they don't understand. How can you look at that flicker and think, yeah, the customer wants this. No one in their right mind would say that's a good feature. Which leads me to conclude they don't understand why it flickers.

So their appropriation of Stockfish seems in line with their general cluelessness and their disdain for their customers. Screen flicker? The customer is too stupid to realize this is shoddy, too dumb to realize quality software doesn't behave like this. They'll just think it's part of the chess experience. Engine origins? What does a computer chess customer know about computer chess engines? Nothing. If we call it our own creation, they're not going to know. Just proclaim we're at the vanguard, we've built the strongest chess entity ever created. And sell the next version of Fritz... with appropriated chess brains and a 20 year old screen painting bug. Value.