CODA now has a released page.

[peter]

Same for me with 6men and this one path to them:
C:\Syzygy\dtz;C:\Syzygy\syzygy;C:\Syzygy\wdl
This one (for 5men) works:
C:\345\syzygy5\3-4-5

Then 4G max. isn't very much hash for analysis or LTC and MultiPV- mode would be nice to have too, regards

[Modern Times]

I don't think you can copyright algorithms? You can either keep them secret (=don't go open source) or try to issue a patent.

sadly nobody asked the authors - I believe AI training should've been opt-in, not opt-out (even here you have no guarantee),
but it's too late to fix that
what we're witnessing is normalization of IP theft - and this goes way beyond silly computer chess

that being said I'm done with open source

A long time ago pretty much everything was closed source. Then came the open source "revolution". Now, because of the use of AI in this way, we are likely to see a shift back towards closed source. And that is perfectly understandable. I would do the same.

Like many others I use AI to help me with my work - it can save me huge amounts of time trying to figure something out that is not working, or to learn a technique. Prior to AI it would have been case of googling the problem, working my way through web page hits and user forum posts to see if anyone else had asked something similar, and even posting in product forums for suggestions. AI cuts through all of that and you get the help way faster and more intelligently. But the help I get is targeted and a tiny part of the jigsaw puzzle. The work I produce is still absolutely my own. What has been done here with CODA is off the scale.

[jasper.sinclair]

But this is not a clean-room design process, clearly. It is like asking an Intern to shuffle up the code so that it still does the same thing, but looks different than the original.

Isn't this fairly common practice? Rewriting code differently to use an idea garnered from elsewhere?

You got it.

Said in other words, what Claude does openly is happening in silence big time for more than a decade.

And nobody cares.

It’s an eco system of silent complicity.

Exactly...
I asked AI:

There has been an explosion of open source engines in the last few years, many of them very close in strength.
IMO, it's akin to inbreeding...and it's hurting original chess engine development.

My request...please check: how many engines list Openbench, or another Openbench engine, or Andrew Grant (the creator of Openbench
...which appears to be derived from Fishtest?) as contributor or inspiration.

My suggestion here is that all share ideas/code in the closed private discord environment, and that's why we see so many in the top of the CCRL rating lists.

OpenAI Codex (5.6 Sol Extra High):

Yes—my GitHub audit strongly supports the existence of a closely connected development ecosystem.
It does not however, by itself, prove undisclosed or improper code sharing inside Discord.

# CCRL and OpenBench Public-Repository Audit

**Audit date:** July 19, 2026
**CCRL data date:** July 16, 2026
**Scope:** Complete CCRL 40/15 list, public OpenBench rosters, public GitHub repositories, and default-branch Git histories

## Executive summary

- All **17 of 17** engines identified in the OpenBench README as supported development projects appear in the complete CCRL 40/15 list.
- All **20 of 20** engines currently configured on the public OpenBench server appear in the complete CCRL 40/15 list.
- All 20 also have a current best-version entry in CCRL.
- Of the 20 public repositories:
- **14** explicitly reference or support OpenBench.
- **14** have a documented connection to Andrew Grant through acknowledgement, attributed component code, repository ownership, or commit history.
- **14** explicitly acknowledge another OpenBench engine as an influence, source, or provider of assistance.
- **10** have commits attributed to identifiable authors of another current OpenBench engine.
- **17** have either a written cross-engine acknowledgement or cross-project commit authorship.
- **19** meet at least one of the audit criteria above.
- **Seer** is the only conservative exception. Its vendored Syzygy code references a Stockfish pull request, but that was not counted as engine-level inspiration or collaboration.
- The public evidence strongly supports the existence of an interconnected development ecosystem and substantial technical convergence.
- It does **not** establish covert, unattributed, or improper source-code exchange in Discord.

## Sources

- [Complete CCRL 40/15 list—all engines and versions](https://computerchess.org.uk/4040/rating_list_all.html)
- [CCRL 40/15 best-version rating list](https://computerchess.org.uk/4040/index.html)
- [Live public OpenBench server](https://chess.grantnet.us/)
- [Official OpenBench repository and supported-engine list](https://github.com/AndyGrant/OpenBench)

## Methodology

The audit used the following rules:

1. Engine families were counted once, irrespective of how many versions appear in CCRL.
2. The complete CCRL 40/15 page was searched—not the shortened text captured from the original link.
3. The OpenBench roster was measured in two ways:
- The **17 supported development projects** named in the OpenBench README.
- The **20 configured engines** displayed by the live public server.
4. Current repository trees were searched for:
- `OpenBench`, `chess.grantnet.us`, and associated build targets.
- Andrew Grant’s name, GitHub identity, and attributed Pyrrhic/Fathom code.
- Named OpenBench engines and explicit descriptions such as *inspired by*, *influence*, *based on*, *help*, *credit*, or *thanks*.
5. Full default-branch Git histories were inspected for authors identifiable with another current OpenBench engine.
6. The following were **not** treated as proof of collaboration:
- Similar playing strength.
- Similar algorithms without acknowledgement or provenance.
- Merely playing another engine in a test.
- Generic references to Stockfish or other engines for comparison.
- Vendored tablebase comments that do not establish engine-level influence.

## Corrected CCRL overlap

The complete CCRL list materially changes the earlier result:

| OpenBench definition | CCRL overlap |
|---|---:|
| README-listed supported development engines | **17/17** |
| Live-server configured engines | **20/20** |

The earlier figures of 14/17 and 16/20 were incorrect because the captured text did not include the lower-ranked portion of the complete list.

## Position within CCRL

All 20 live-server engines have a best-version entry in CCRL:

| Cumulative ranking | Number of OpenBench engines |
|---|---:|
| Top 20 | **3** |
| Top 50 | **7** |
| Top 100 | **12** |
| Top 150 | **16** |
| Complete ranked list | **20** |

### Top-50 OpenBench engines

- Stockfish — rank 1
- Berserk — rank 10
- Halogen — rank 13
- RubiChess — rank 26–28
- Ethereal — rank 30
- Seer — rank 35
- Igel — rank 40–41

### Remaining engines

- Koivisto — rank 52
- Black Marlin — rank 64
- Winter — rank 72
- Equisetum — rank 81
- Stash — rank 94
- Weiss — rank 114–115
- Zahak — rank 116
- Laser — rank 136
- Drofa — rank 143
- Demolito — rank 210
- FabChess — rank 213–214
- Bit-Genie — rank 223
- 4ku — rank 242–243

This shows a substantial top-level presence, but not domination of the very top: only three of the current OpenBench engines are in CCRL’s top 20.

## Public GitHub evidence

The categories overlap; an engine may appear in several.

| Publicly documented connection | Repositories |
|---|---:|
| Explicit OpenBench reference or support | **14/20** |
| Andrew Grant acknowledgement, component, ownership, or commit | **14/20** |
| Another OpenBench engine acknowledged as influence, assistance, or code provenance | **14/20** |
| Commits attributed to authors of another OpenBench engine | **10/20** |
| Written cross-engine acknowledgement or cross-project authorship | **17/20** |
| At least one requested criterion | **19/20** |

Under the narrower 17-engine list in the OpenBench README, **16 of 17** meet at least one criterion. Seer remains the only conservative exception.

## Cross-project Git authorship

The following default-branch histories contain commits attributed to identifiable authors of another current OpenBench engine:

| Repository | Cross-project authors recorded in history |
|---|---|
| [4ku](https://github.com/kz04px/4ku) | Morgan Houppin/Stash: 1 |
| [Drofa](https://github.com/justNo4b/Drofa) | Terje Kirstihagen/Weiss: 5 |
| [Ethereal](https://github.com/AndyGrant/Ethereal) | Lucas Braesch/Demolito: 17; Terje/Weiss: 7; Morgan/Stash: 4; Jay Honnold/Berserk: 1 |
| [FabChess](https://github.com/fabianvdW/FabChess) | Terje/Weiss: 7 |
| [Halogen](https://github.com/KierenP/Halogen) | Andrew Grant: 5; Aryan Parekh/Bit-Genie: 3; Luecx/Koivisto: 1 |
| [Koivisto](https://github.com/Luecx/Koivisto) | Jay/Berserk: 31; kz04px/4ku: 4; Aryan/Bit-Genie: 3; Morgan/Stash: 3; Doruk/Black Marlin: 2 |
| [Laser](https://github.com/jeffreyan11/laser-chess-engine) | Andrew Grant: 4 |
| [Stash](https://github.com/mhouppin/stash-bot) | kz04px/4ku: 1 |
| [Weiss](https://github.com/TerjeKir/weiss) | justNo4b/Drofa: 11; Andrew Grant: 1 |
| [Winter](https://github.com/rosenthj/Winter) | justNo4b/Drofa: 1 |

These are Git author-attribution counts, not measurements of contribution size. A recorded commit can result from a pull request, merge, cherry-pick, or another normal Git workflow.

## Explicit written acknowledgements

Several repositories describe their relationships unusually clearly:

- [Berserk](https://github.com/jhonnold/berserk#credit)
- Has a section called **Engine Influences**.
- Lists Stockfish, Ethereal, Koivisto, Weiss, and others.
- Lists OpenBench among its development resources.

- [Black Marlin](https://github.com/jnlt3/blackmarlin)
- States that its low-depth singular-extension idea was inspired by Koivisto.

- [Drofa](https://github.com/justNo4b/Drofa#origins)
- States that it uses Weiss’s LMP base-reduction formulas.
- Describes its tuning implementation as closely following Weiss’s logic and effectively being a C++ rewrite.
- Credits Ethereal, Stockfish, Berserk, Koivisto, Andrew Grant, and the OpenBench community.
- Says the OpenBench community helped find bugs, taught programming practices, and provided technical discussion.

- [Equisetum](https://github.com/justNo4b/Equisetum)
- Describes itself as a continuation of Drofa.
- Repeats many of Drofa’s acknowledgements and technical influences.

- [Halogen](https://github.com/KierenP/Halogen)
- States that its neural-network architecture was inspired by Stockfish NNUE.
- Uses OpenBench for development.
- Contains Andrew Grant-attributed commits and Pyrrhic code.

- [Igel](https://github.com/vshcherbyna/igel#acknowledgements)
- Says Andrew Grant’s invitation to OpenBench accelerated its strength development.
- States that Ethereal, Stockfish, OpenBench, and the listed projects influenced its development.

- [Koivisto](https://github.com/Luecx/Koivisto#acknowledgements)
- Lists contributors associated with Berserk, Bit-Genie, Stash, 4ku, and Black Marlin.
- Credits Andrew Grant for invaluable help and advice.

- [Laser](https://github.com/jeffreyan11/laser-chess-engine)
- Credits Stockfish and other open-source engines for inspiration and ideas.
- Has commits attributed to Andrew Grant.

- [RubiChess](https://github.com/Matthies/RubiChess)
- States that its author examined Ethereal, Stockfish, Laser, Koivisto, Berserk, and other source code.
- Thanks Andrew Grant and the OpenBench testing infrastructure.

- [Zahak](https://github.com/amanjpro/zahak#acknowledgement)
- Credits Bit-Genie assistance.
- Lists Weiss, Ethereal, Berserk, and other engines.
- Explicitly thanks the OpenBench Discord community.

## OpenBench and Fishtest

The OpenBench README describes OpenBench as **heavily inspired by Fishtest**.

- OpenBench is not identified by GitHub as a Fishtest fork.
- The public evidence therefore supports the description **Fishtest-inspired reimplementation**.
- Describing it as source-derived from Fishtest would require a separate code-history comparison.

## Interpretation

### What the evidence supports

The public record supports the existence of a highly interconnected development ecosystem involving:

- Shared search and pruning concepts.
- Reused or rewritten reduction formulas.
- Common NNUE architectures and training methods.
- Shared trainers, datasets, and testing practices.
- Attributed Pyrrhic/Fathom tablebase components.
- Direct advice between engine authors.
- Public cross-project commits.
- A common distributed SPRT framework that rapidly validates incremental changes.

This provides a credible public explanation for technical convergence among many modern engines.

### What the evidence does not support

The audit does not establish:

- Covert source exchange.
- Unattributed copying.
- License violations.
- Coordinated manipulation of rating lists.
- That private Discord conversations are the primary cause of the convergence.

OpenBench publishes a public Discord invitation. Its discussions are not indexed and are difficult to audit externally, but that opacity is not itself evidence of improper conduct.

## Effect on original engine development

The concern about reduced originality is technically plausible:

- Distributed SPRT rewards small, measurable, low-risk gains.
- Successful patches and ideas can be examined and retested quickly by other authors.
- Common NNUE and alpha-beta designs reduce architectural diversity.
- Novel approaches may initially regress and therefore fail short-term testing despite longer-term potential.
- New authors can reach high strength quickly by following an established collection of successful techniques.

The most accurate description is therefore **development monoculture** or **convergent evolution**, rather than proof of covert copying.

## Conclusion

The corrected audit finds that:

1. Every current public OpenBench engine appears in CCRL.
2. Nineteen of the twenty public repositories have an explicit, publicly verifiable OpenBench, Andrew Grant, or cross-engine connection.
3. Ten repositories contain commits attributed to identifiable authors of another current OpenBench engine.
4. The public evidence is already sufficient to explain substantial technical convergence without assuming secret code sharing.
5. The concern that this ecosystem can discourage genuinely different engine architectures is reasonable, even though misconduct is not established.

[jasper.sinclair]

For those of you, reluctant to read and digest this whole tecnical report...
there are a couple important take-aways:

"Yes—my GitHub audit strongly supports the existence of a closely connected development ecosystem."
"The public record supports the existence of a highly interconnected development ecosystem"

[AndrewGrant]

My request...please check: how many engines list Openbench, or another Openbench engine, or Andrew Grant (the creator of Openbench
...which appears to be derived from Fishtest?) as contributor or inspiration.

Your brilliant AI agent missed the following:

- OpenBench is not derived from Fishtest
- grantnet.us is one of dozens of "instances" of the OpenBench software
- Nearly -every- engine in the top-25 is using an OpenBench instance. Not just a few as you suggested.
- Of the top engines, only Stockfish, Dragon, Fritz, are not. Even Leela has an OpenBench instance.
- OpenBench is just a distributed way to run Cutechess/Fastchess.
- There are dozens of free-to-join Discord servers that talk about chess, just in the same way as this forum used to

This is also WILDLY off-topic from the original thread, and the following conversations about Coda's license violations.

I've asked the moderators to move your post to a new thread. And this response should follow as well.

[jasper.sinclair]

My request...please check: how many engines list Openbench, or another Openbench engine, or Andrew Grant (the creator of Openbench
...which appears to be derived from Fishtest?) as contributor or inspiration.

Your brilliant AI agent missed the following:

- OpenBench is not derived from Fishtest
- grantnet.us is one of dozens of "instances" of the OpenBench software
- Nearly -every- engine in the top-25 is using an OpenBench instance. Not just a few as you suggested.
- Of the top engines, only Stockfish, Dragon, Fritz, are not. Even Leela has an OpenBench instance.
- OpenBench is just a distributed way to run Cutechess/Fastchess.
- There are dozens of free-to-join Discord servers that talk about chess, just in the same way as this forum used to

This is also WILDLY off-topic from the original thread, and the following conversations about Coda's license violations.

I've asked the moderators to move your post to a new thread. And this response should follow as well.

I don't claim Ai to be brilliant, but I do thinks it's factual.

I'm simply repsonding and ageeing with Chris...and using AI (slop! ) to point out where and how I believe it's happening.
If you have a problem with my post, please take it up with OpenAI.

But this is not a clean-room design process, clearly. It is like asking an Intern to shuffle up the code so that it still does the same thing, but looks different than the original.

Isn't this fairly common practice? Rewriting code differently to use an idea garnered from elsewhere?

You got it.

Said in other words, what Claude does openly is happening in silence big time for more than a decade.

And nobody cares.

It’s an eco system of silent complicity.

[AndrewGrant]

I don't claim Ai to be brilliant, but I do thinks it's factual.

Ideally, no one will post any AI slop, because it is not worth reading . . .
But if they must, then they ought to due their due diligence first in confirming the results . . .

We already saw in this very thread that the AI is perfectly capable of producing non-factual results, such as when it claimed that Fathom/Pyrrhic are under a GPLv3 license when they are in fact under an MIT license. Who knows what else it gets wrong. I recently had Gemini confuse AMD CPUs for Intel CPUs, and provide me fake citations that proved itself wrong.

[Rebel]

This is also WILDLY off-topic from the original thread, and the following conversations about Coda's license violations.

This is true, post is off-topic. But the question arises, when Claude fixes the license issues, will programmers be held to the same standard ?

[Rebel]

https://edition.cnn.com/2026/07/24/busi ... ement-vrtc

Interesting court ruling regarding Anthropic (Claude).

[adamtwiss]

I've just released coda v0.9.3. Headlines include a new net (+15 Elo at LTC) and a significant EGTB-probing speed-up (end-game NPS now about 4x higher). It also adds MultiPV support, and options for bigger hash table sizes that were requested earlier in this thread.