♟ THE GOK CHALLENGE

For years, ArnoHu's Game of Kings has been one of the benchmarks for what a Scratch chess engine can achieve.

It took roughly eight years of development.

I'm setting myself a challenge:

Before the end of 2026, can I take GoK itself, improve it, and produce an engine that can demonstrably beat the original?

⚔️ The goal isn't just to make GoK faster.

A higher NPS number doesn't automatically mean stronger chess.
A greater reported search depth doesn't automatically mean stronger chess.
And one lucky win certainly doesn't prove anything.

The goal is to create a version that can actually play stronger chess under fair testing conditions.

Deadline: 31 December 2026

This thread will document the challenge as it happens — discoveries, experiments, failed ideas, improvements, benchmarks and, eventually, GoK vs GoK Ascendant.


————————————————–


THE ORIGINAL: GAME OF KINGS

Scratch Chess AI — Game of Kings
Created by ArnoHu

▶ Official Scratch project

⚡ Play GoK on TurboWarp

GoK is already an extremely sophisticated Scratch chess project.

It supports multiple difficulty levels, full chess rules, FEN and PGN export, debugging information, search profiling, an opening book and neural-network evaluation.

On the original project, the strongest levels are time-based rather than fixed-depth:

• Medium — roughly 15 seconds of think time in the midgame
  
• Difficult — roughly 25 seconds in the midgame and 15 seconds in the endgame
  

The original project also contains dedicated debugging and profiling tools, including evaluation, search depth, NPS, MPS, function timing and transposition-table statistics.

That makes this challenge much more interesting.

I'm not starting with a weak engine.

I'm starting with something that has already been worked on for years.


————————————————–


MY VERSION: GOK ASCENDANT

Working name: Game of Kings: Ascendant
Short name: GoK Ascendant

GoK Ascendant is a modified version of Game of Kings.

The goal is not to replace the original project or erase its history.

The goal is to answer one question:

How much further can Game of Kings actually be pushed?

I want to preserve the spirit of the original project while improving the engine itself.

That means studying what GoK already does well, finding real weaknesses or bottlenecks, testing improvements individually, and rejecting ideas that do not actually help.


————————————————–


⚖️ WHAT COUNTS AS “BEATING GOK”?

This is probably the most important part of the challenge.

If I give GoK Ascendant five minutes to think and give the original GoK one second, then winning means almost nothing.

So I want the comparison to be as fair as reasonably possible.

The current comparison rules are:

• Scratch vs Scratch.
  
• TurboWarp vs TurboWarp.
  
• Same hardware where practical.
  
• Matched actual thinking time, not just matching a number in a setting.
  
• Same opening-book policy, or the opening book disabled for both sides during controlled comparisons.
  
• Both colours tested.
  
• Multiple starting positions or openings.
  
• No illegal-move or state-handling regressions accepted.
  
• Important games should be recorded in PGN where possible.
  
• One game is not enough to prove superiority.
  

Important:

A faster engine is not automatically a stronger engine.

A higher NPS result is useful.
A greater completed search depth is useful.
A faster evaluation function is useful.

But none of those alone proves stronger chess.

The final test is move quality and actual game results under comparable conditions.

The development plan already treats matched runtime, hardware and actual thinking time as essential to a valid strength comparison, rather than assuming that faster node processing automatically means stronger play. :chatgpt-content-reference{index=“0”}


————————————————–


WHAT AM I ACTUALLY CHANGING?

I am improving GoK itself.

I am not planning to hide another chess engine behind GoK's interface.

The final engine is intended to remain a real Scratch project.

Allowed inside the released engine:

• Native Scratch 3 blocks
  
• Scratch's built-in Pen extension where needed
  
• GoK's existing engine architecture, modified and improved
  
• New Scratch-side optimisations and search improvements
  

Not allowed as the actual engine:

• JavaScript chess engines
  
• WebAssembly chess engines
  
• Remote servers choosing moves
  
• Stockfish secretly running in the background
  
• Custom TurboWarp extensions doing the actual chess calculations
  

External tools can still be used during development for testing, verification and analysis.

For example, another chess program may be useful for checking whether a move is legal or verifying a test position.

But the released project itself must still calculate its own moves in Scratch.


————————————————–


HOW THE DEVELOPMENT WILL WORK

I don't want to throw twenty optimisations into the project at once and then have no idea which one helped.

The development process is deliberately experimental.

A typical improvement will go through something like this:

• Identify a possible weakness, bottleneck or strength limitation.
  
• Measure the original behaviour.
  
• Make one isolated change.
  
• Check chess correctness.
  
• Compare the candidate against the original.
  
• Keep the change only if the evidence supports it.
  

PROMOTED = evidence suggests the improvement should stay.

EXPERIMENTAL = still being tested.

REJECTED = it did not help enough, caused regressions, or produced unreliable results.

I will post important rejected experiments too.

A failed experiment is still useful if it teaches us something about the engine.


————————————————–


️ DEVELOPMENT ROADMAP

PHASE 1 — Understand GoK
STATUS: IN PROGRESS

Map how the engine works before making major changes.

This includes the move generator, board representation, search, evaluation, opening book, neural-network evaluation, timing system and existing optimisation techniques.


PHASE 2 — Establish the baseline
STATUS: NOT STARTED

Build a trustworthy baseline for correctness and performance.

The point is to know exactly what the original engine does before claiming that anything has improved.


PHASE 3 — Improve the engine
STATUS: NOT STARTED

Test individual improvements.

Possible areas may include search efficiency, move ordering, evaluation work, repeated calculations, list access, timing behaviour or other engine bottlenecks.

Nothing is assumed to be an improvement until it is measured.


PHASE 4 — Validate GoK Ascendant
STATUS: NOT STARTED

Make sure improvements have not broken:

• legal move handling
  
• castling
  
• en-passant
  
• promotion
  
• check/checkmate logic
  
• board state restoration
  
• imports and exports
  
• timing
  
• normal UI behaviour
  

PHASE 5 — GOK VS GOK ASCENDANT
STATUS: NOT STARTED

This is the part I'm most interested in.

Controlled engine matches.

Same environment.
Comparable time.
Both colours.
Multiple openings.
Recorded results.

This is where we find out whether the improvements actually translate into stronger chess.


PHASE 6 — RELEASE
STATUS: NOT STARTED

If the challenge succeeds, I will release the strongest tested version.

If the evidence turns out to be mixed, I'll say that too.


————————————————–


CURRENT STATUS

Challenge: ACTIVE

Deadline: 31 December 2026

Current phase: Phase 1 — Understanding GoK

Original engine: Game of Kings by ArnoHu

Candidate engine: GoK Ascendant

Public release: Not yet released

Strength claim: None yet — development in progress


Progress:

Phase 1 — Understanding GoK
Phase 2 — Baseline testing
Phase 3 — Improvements
Phase 4 — Validation
Phase 5 — Engine matches
Phase 6 — Release


————————————————–


UPDATE INDEX

I will link the important development updates here.

Update #1 — The GoK Challenge begins
Update #2 — Coming soon
Update #3 — Coming soon
Update #4 — Coming soon

I won't make a post for every tiny block change.

An update should contain something worth reading:

• a new discovery about GoK
  
• a completed experiment
  
• a benchmark
  
• a rejected idea
  
• a significant engine improvement
  
• a major bug fix
  
• a strength test
  
• a milestone
  

————————————————–


EXPERIMENT LOG

This section will eventually summarise the most important experiments.

Experiment #1
Pending

Experiment #2
Not started

Experiment #3
Not started


Example result labels:

✓ PROMOTED
The change passed testing and currently remains in GoK Ascendant.

~ EXPERIMENTAL
The result is interesting, but more testing is needed.

✗ REJECTED
The idea did not provide enough benefit or caused a regression.


————————————————–


❓ FAQ

Are you saying ArnoHu's GoK is bad?

No.

The opposite.

This challenge is interesting precisely because GoK is already an extremely ambitious Scratch chess engine.

Trying to improve something weak would be easy.

Trying to improve something that has already received years of work is much harder.


Have you already beaten GoK?

No.

There is currently no strength claim.

The point of this thread is to follow the attempt before the final version is released.


Why call it GoK Ascendant?

Because it is intended to be recognisably based on Game of Kings while representing an attempt to push the engine further.


Why not just increase the thinking time?

Because that would not be a fair comparison.

Serious strength tests will attempt to match actual thinking time.


What if GoK Ascendant reaches a higher depth?

That is promising, but it does not prove stronger chess by itself.

The same applies to NPS.

The important question is whether it chooses better moves and performs better in actual games.


Will TurboWarp count?

Yes, but separately.

Scratch results and TurboWarp results will not be mixed together.

GoK behaves very differently when given TurboWarp's much faster runtime, so each environment needs its own comparison.


Are you using Stockfish?

Not as the engine inside the project.

External chess software may be used during development as a testing or verification tool.

The final GoK Ascendant project must still calculate its own moves.


Can other people test GoK Ascendant?

Eventually, yes.

Later in development I may ask for test positions, unusual chess situations, games, bug reports or independent match testing.


Can I suggest an optimisation?

Absolutely.

Suggestions are welcome.

Just keep in mind that an idea still needs to be tested before it can be considered an improvement.


When will GoK Ascendant be released?

The target is before the end of 2026.

I would rather release a properly tested engine than rush out something that only looks stronger.


————————————————–


CREDITS

This project would not exist without the work behind the original Game of Kings.

Game of Kings
Created by ArnoHu

Official project

The original project credits the Smallbrain project for neural-network model data:

Smallbrain on GitHub

The supplied GoK credits also identify:

• ArnoHu — Scratch port and integration
  
• FastSkript — improved accumulator updates
  
• Grijfland — original opening book
  
• S_P_A_R_T — opening book data
  

These credits should remain clearly visible when GoK's NNUE work is reused. :chatgpt-content-reference{index=“1”}

Some original GoK artwork is separately licensed, including the AI bot image and piece SVGs, so those assets also need to be handled carefully in any eventual public release. :chatgpt-content-reference{index=“2”}

GoK Ascendant is an improvement project built on top of years of work by other people.

The purpose of this challenge is not to erase that work.

It is to see how much further it can go.


————————————————–


⚔️ THE CHALLENGE

Eight years of development.

One of Scratch's most ambitious chess engines.

A programming environment never designed for high-performance chess search.

And a deadline at the end of 2026.

Can Game of Kings be pushed further?

I don't know yet.

That's the point.

The challenge starts now.

♟️

update

update

update

update

UPDATE #2 — THE FIRST ASCENDANT BASELINE IS COMPLETE

A lot has happened since the challenge started.

The first full improvement cycle is now complete.

After mapping Game of Kings, establishing a controlled baseline, testing individual optimisation ideas and validating the resulting engine, I now have the first version of GoK Ascendant that I am comfortable using as the new development baseline.

Performance result:

In a matched fixed-depth benchmark, the current Ascendant engine completed the same search work in approximately 5.72% less total time than the baseline.

That does NOT automatically mean 5.72% stronger chess.

The important part is that the candidate also passed the bounded correctness tests used during this development cycle.

Engine-v-engine testing:

I also built an automated match system that:

• uses the same Scratch runtime
  
• matches actual wall-clock thinking time
  
• uses paired openings
  
• reverses colours
  
• uses the same opening-book policy
  
• checks moves with an independent chess arbiter
  
• records games and results
  

The long match became much more expensive to run than expected because of Scratch/headless-runtime behaviour, so I eventually stopped it rather than spending several more days testing.

The usable completed sample currently stands at:

GoK Ascendant: 7 wins
Baseline GoK: 4 wins

There was also one unfinished very long game and one baseline timeout/fault.

Important:

I am NOT claiming a +Elo value or statistically proven superiority from 11 completed games.

The result is encouraging, but it is still a relatively small and incomplete sample.

What I can currently claim is:

✓ the first Ascendant optimisation cycle passed its correctness checks
✓ the current engine is measurably faster in the controlled benchmark
✓ limited fair match evidence is currently favourable
~ statistically significant playing-strength superiority is still unproven

Because of that, I have promoted this engine to become the new GoK Ascendant development baseline.

The next stage is NOT release yet.

I'm starting a second improvement cycle to see whether the engine can be pushed further before doing the final strength test.

The challenge continues.

update

update

update

update

update

Good luck for the challenge!

austin_intel001jrahs wrote:

UPDATE #2 — THE FIRST ASCENDANT BASELINE IS COMPLETE

A lot has happened since the challenge started.

The first full improvement cycle is now complete.

After mapping Game of Kings, establishing a controlled baseline, testing individual optimisation ideas and validating the resulting engine, I now have the first version of GoK Ascendant that I am comfortable using as the new development baseline.

Performance result:

In a matched fixed-depth benchmark, the current Ascendant engine completed the same search work in approximately 5.72% less total time than the baseline.

That does NOT automatically mean 5.72% stronger chess.

The important part is that the candidate also passed the bounded correctness tests used during this development cycle.

Engine-v-engine testing:

I also built an automated match system that:

• uses the same Scratch runtime
  
• matches actual wall-clock thinking time
  
• uses paired openings
  
• reverses colours
  
• uses the same opening-book policy
  
• checks moves with an independent chess arbiter
  
• records games and results
  

The long match became much more expensive to run than expected because of Scratch/headless-runtime behaviour, so I eventually stopped it rather than spending several more days testing.

The usable completed sample currently stands at:

GoK Ascendant: 7 wins
Baseline GoK: 4 wins

There was also one unfinished very long game and one baseline timeout/fault.

Important:

I am NOT claiming a +Elo value or statistically proven superiority from 11 completed games.

The result is encouraging, but it is still a relatively small and incomplete sample.

What I can currently claim is:

✓ the first Ascendant optimisation cycle passed its correctness checks
✓ the current engine is measurably faster in the controlled benchmark
✓ limited fair match evidence is currently favourable
~ statistically significant playing-strength superiority is still unproven

Because of that, I have promoted this engine to become the new GoK Ascendant development baseline.

The next stage is NOT release yet.

I'm starting a second improvement cycle to see whether the engine can be pushed further before doing the final strength test.

The challenge continues.

From the time of your original posting I am estimating your baseline GoK NNUE might be around version 7.3383.2318 (you can verify by pressing ‘v’). I played a short 20-games series against that baseline on TurboWarp (1sec), and current GoK NNUE won 12.5 - 7.5. Doesn't matter so much which version it was, what I am saying is the project underwent improvements since, although I will slow it down now and focus on GoK Classic.

ArnoHu wrote:

austin_intel001jrahs wrote:

UPDATE #2 — THE FIRST ASCENDANT BASELINE IS COMPLETE

A lot has happened since the challenge started.

The first full improvement cycle is now complete.

After mapping Game of Kings, establishing a controlled baseline, testing individual optimisation ideas and validating the resulting engine, I now have the first version of GoK Ascendant that I am comfortable using as the new development baseline.

Performance result:

In a matched fixed-depth benchmark, the current Ascendant engine completed the same search work in approximately 5.72% less total time than the baseline.

That does NOT automatically mean 5.72% stronger chess.

The important part is that the candidate also passed the bounded correctness tests used during this development cycle.

Engine-v-engine testing:

I also built an automated match system that:

• uses the same Scratch runtime
  
• matches actual wall-clock thinking time
  
• uses paired openings
  
• reverses colours
  
• uses the same opening-book policy
  
• checks moves with an independent chess arbiter
  
• records games and results
  

The long match became much more expensive to run than expected because of Scratch/headless-runtime behaviour, so I eventually stopped it rather than spending several more days testing.

The usable completed sample currently stands at:

GoK Ascendant: 7 wins
Baseline GoK: 4 wins

There was also one unfinished very long game and one baseline timeout/fault.

Important:

I am NOT claiming a +Elo value or statistically proven superiority from 11 completed games.

The result is encouraging, but it is still a relatively small and incomplete sample.

What I can currently claim is:

✓ the first Ascendant optimisation cycle passed its correctness checks
✓ the current engine is measurably faster in the controlled benchmark
✓ limited fair match evidence is currently favourable
~ statistically significant playing-strength superiority is still unproven

Because of that, I have promoted this engine to become the new GoK Ascendant development baseline.

The next stage is NOT release yet.

I'm starting a second improvement cycle to see whether the engine can be pushed further before doing the final strength test.

The challenge continues.

From the time of your original posting I am estimating your baseline GoK NNUE might be around version 7.3383.2318 (you can verify by pressing ‘v’). I played a short 20-games series against that baseline on TurboWarp (1sec), and current GoK NNUE won 12.5 - 7.5. Doesn't matter so much which version it was, what I am saying is the project underwent improvements since, although I will slow it down now and focus on GoK Classic.

HOLY VERSION NUMBER

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

u underestimate my abilities

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

Why not? Single improvements are always possible. The engine was created by hundreds of single improvements.

ArnoHu wrote:

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

Why not? Single improvements are always possible. The engine was created by hundreds of single improvements.

Yeah, but you have more experience, you will probably make more improvements than him in a given time. Of course "probably

austin_intel001jrahs wrote:

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

u underestimate my abilities

Well, I didn’t say you've no abilities, maybe what I said was confusing. Anyway, good luck, and as arnohu says, single improvements are always possible.

TheRandomFrenchGuy wrote:

ArnoHu wrote:

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

Why not? Single improvements are always possible. The engine was created by hundreds of single improvements.

Yeah, but you have more experience, you will probably make more improvements than him in a given time. Of course "probably

austin_intel001jrahs wrote:

TheRandomFrenchGuy wrote:

Good luck with the challenge, I personnally abandonnes it months ago after 2 years, hope for you you will beat years of optimization, but don’t worry if you can’t make any improvement, don’t forget that gok is known as the best chess engine in scratch. Good luck.

u underestimate my abilities

Well, I didn’t say you've no abilities, maybe what I said was confusing. Anyway, good luck, and as arnohu says, single improvements are always possible.

Thanks! I will try my best and make sure that will not happen XD

ArnoHu wrote:

austin_intel001jrahs wrote:

UPDATE #2 — THE FIRST ASCENDANT BASELINE IS COMPLETE

A lot has happened since the challenge started.

The first full improvement cycle is now complete.

After mapping Game of Kings, establishing a controlled baseline, testing individual optimisation ideas and validating the resulting engine, I now have the first version of GoK Ascendant that I am comfortable using as the new development baseline.

Performance result:

In a matched fixed-depth benchmark, the current Ascendant engine completed the same search work in approximately 5.72% less total time than the baseline.

That does NOT automatically mean 5.72% stronger chess.

The important part is that the candidate also passed the bounded correctness tests used during this development cycle.

Engine-v-engine testing:

I also built an automated match system that:

• uses the same Scratch runtime
  
• matches actual wall-clock thinking time
  
• uses paired openings
  
• reverses colours
  
• uses the same opening-book policy
  
• checks moves with an independent chess arbiter
  
• records games and results
  

The long match became much more expensive to run than expected because of Scratch/headless-runtime behaviour, so I eventually stopped it rather than spending several more days testing.

The usable completed sample currently stands at:

GoK Ascendant: 7 wins
Baseline GoK: 4 wins

There was also one unfinished very long game and one baseline timeout/fault.

Important:

I am NOT claiming a +Elo value or statistically proven superiority from 11 completed games.

The result is encouraging, but it is still a relatively small and incomplete sample.

What I can currently claim is:

✓ the first Ascendant optimisation cycle passed its correctness checks
✓ the current engine is measurably faster in the controlled benchmark
✓ limited fair match evidence is currently favourable
~ statistically significant playing-strength superiority is still unproven

Because of that, I have promoted this engine to become the new GoK Ascendant development baseline.

The next stage is NOT release yet.

I'm starting a second improvement cycle to see whether the engine can be pushed further before doing the final strength test.

The challenge continues.

From the time of your original posting I am estimating your baseline GoK NNUE might be around version 7.3383.2318 (you can verify by pressing ‘v’). I played a short 20-games series against that baseline on TurboWarp (1sec), and current GoK NNUE won 12.5 - 7.5. Doesn't matter so much which version it was, what I am saying is the project underwent improvements since, although I will slow it down now and focus on GoK Classic.

Final result 40 - 26. Here is your target.