SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.30100 · 2026

CHESS-WORLD-MODEL: A 10M-Game Benchmark for Exact State Tracking from Chess Move Sequences

Terry Lyons, Benjamin Walker

arXiv · PDF · Open in the Atlas

Code that ran

17 of 19 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 2 violated it and 1 has no contract verdict; 11 failures of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2605.30100.json · badge: /api/ran/2605.30100.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
Benjamin-Walker/Chess-World-Model canonical (classifier label)
classifier basis: census
12 ran · 2 no recorded run
niklasf/python-chess canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
datasig-ac-uk/slices community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
SLiCE Ran
ran on a synthesized input · constructed an object; no result was computed
datasig-ac-uk/slices/src/slices/slices.py
code served (permissive licence) · get_code("1581751836cf2f8d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, SLiCE: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'input_dim': 4,
    'hidden_dim': 8,
    'bias': True,
    'block_size': 2,
    'diagonal_dense': False,
    'init_std': 0.01,
    'scale': 1.0,
    'input_dependent_init': False,
    'use_parallel': False,
    'chunk_size': 256,
    'path_mode': 'values',
    'transition_mode': 'euler',
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SLiCE — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SLiCE output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
apply_rope Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/models/transformer.py
code served (permissive licence) · get_code("289ce4c835947e59")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 4, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors for q, k, cos, sin
# q and k are query and key tensors with shape (batch, seq_len, hidden_dim)
# cos and sin are rotation matrices with matching shape
batch_size = 2
seq_len = 3
hidden_dim = 4

q = torch.randn(batch_size, seq_len, hidden_dim)
k = torch.randn(batch_size, seq_len, hidden_dim)
cos = torch.randn(batch_size, seq_len, hidden_dim)
sin = torch.randn(batch_size, seq_len, hidden_dim)

ARGS = {
    "q": q,
    "k": k,
    "cos": cos,
    "sin": sin,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 4, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
binned_exact_sums Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Benjamin-Walker/Chess-World-Model/evaluation/metrics.py
code served (permissive licence) · get_code("2f46735620be1e53")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Build the 'out' dict with model predictions
# The function expects these keys with logit outputs that can be argmax'd
B, T = 2, 4
out = {
    'pieces': torch.randn(B, T, 64, 13),  # 64 piece positions, 13 classes
    'side': torch.randn(B, T, 2),          # 2 classes for side
    'castle': torch.randn(B, T, 4, 16),    # 4 castle positions, 16 classes
    'ep_file': torch.randn(B, T, 8),       # 8 classes for ep file
    'ep_rank': torch.randn(B, T, 8),       # 8 classes for ep rank
    'halfmove': torch.randn(B, T, 2, 151), # 2 halfmove values, 151 classes
    'fullmove': torch.randn(B, T, 2, 16),  # 2 fullmove values, 16 classes
}

# Build states tensor with target values
# Shape: (B, T, 75) where 75 = 64 + 1 + 4 + 1 + 1 + 2 + 2
states = torch.zeros(B, T, 75, dtype=torch.int64)
# Piece targets (0:64)
states[:, :, 0:64] = torch.randint(0, 13, (B, T, 64))
# Side target (64)
states[:, :, 64] = torch.randint(0, 2, (B, T))
# Castle targets (65:69)
states[:, :, 65:69] = torch.randint(0, 16, (B, T, 4))
# EP file target (69)
states[:, :, 69] = torch.randint(0, 8, (B, T))
# EP rank target (70)
states[:, :, 70] = torch.randint(0, 8, (B, T))
# Halfmove targets (71:73)
states[:, :, 71:73] = torch.randint(0, 151, (B, T, 2))
# Fullmove targets (73:75)
states[:, :, 73:75] = torch.randint(0, 16, (B, T, 2))

# Build pad_mask: False means valid, True means pad
# Make most valid with a few padded positions
pad_mask = torch.zeros(B, T, dtype=torch.bool)
pad_mask[0, 3] = True  # One padded position

bin_size = 2

ARGS = {
    'out': out,
    'states': states,
    'pad_mask': pad_mask,
    'bin_size': bin_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
canonicalise_mamba_variant Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/models/factory.py
code served (permissive licence) · get_code("d4304e716ce0c661")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

# Setup for canonicalise_mamba_variant function
# The function takes a variant string and returns a canonicalized version
# It looks up the variant in _MAMBA_VARIANT_ALIASES dict and returns the mapped value
# or returns the original variant if not found

ARGS = {
    'variant': 'mamba1'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
clone_model_state_to_cpu Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/training.py
code served (permissive licence) · get_code("da75060faeb09c5e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple model with a few parameters
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

ARGS = {
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_rate Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/training.py
code served (permissive licence) · get_code("01f3007faa62d2bf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function format_rate takes a single float parameter x
# and returns a formatted string representation of it
# We need to provide a float value for x

ARGS = {
    'x': 1234567890.5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
in_queenside Ran
ran, violated the contract we drafted
niklasf/python-chess/chess/gaviota.py
pointer only (licence: GPL-3.0) · get_code("b21a463ebc06fe08")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, niklasf/python-chess, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

map24_b Ran
ran, honoured the contract we drafted
niklasf/python-chess/chess/gaviota.py
pointer only (licence: GPL-3.0) · get_code("9793ec30c4795ad7")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, niklasf/python-chess, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

map88 Ran
ran, honoured the contract we drafted
niklasf/python-chess/chess/gaviota.py
pointer only (licence: GPL-3.0) · get_code("c8f528c58a2be140")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, niklasf/python-chess, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

masked_ce_sum Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/evaluation/metrics.py
code served (permissive licence) · get_code("ace6c857d617c5f5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create small tensors for the function
batch_size = 2
num_classes = 4
spatial_dim = 3

# logits: [..., C] shape - we'll make it [batch, spatial, num_classes]
logits = torch.randn(batch_size, spatial_dim, num_classes)

# targets: [...] shape - same as logits but without last dimension
targets = torch.randint(0, num_classes, (batch_size, spatial_dim))

# valid_mask: same shape as targets (bool)
valid_mask = torch.tensor([
    [True, False, True],
    [True, True, False]
])

ARGS = {
    "logits": logits,
    "targets": targets,
    "valid_mask": valid_mask
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
model_has_nonfinite_params Ran
ran, violated the contract we drafted
Benjamin-Walker/Chess-World-Model/training.py
code served (permissive licence) · get_code("e87873998a700750")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
model = SimpleModel()

ARGS = {
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_datasets Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/evaluation/evaluate_on_test_sets.py
code served (permissive licence) · get_code("8944da0d7f47dbc9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from pathlib import Path

# Since we can't use __file__, we'll create a mock PROJECT_ROOT
PROJECT_ROOT = Path('/mock/project')

# Setup for parse_datasets function
# The function takes dataset_args: list[str] and returns list[tuple[str, str]]
# It either returns DEFAULT_TEST_DATASETS if dataset_args is empty,
# or parses the dataset_args which should be in format "NAME=PATH"

ARGS = {
    'dataset_args': [
        'my_dataset=data/path/to/dataset.jsonl',
        'another_dataset=/absolute/path/to/another.jsonl'
    ]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_checkpoints Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/evaluation/evaluate_on_test_sets.py
code served (permissive licence) · get_code("2dc433c190dca68f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import argparse
import os
import tempfile

# Create temporary directory and checkpoint files for testing
tmpdir = tempfile.mkdtemp()

# Create a checkpoint file
ckpt_file = os.path.join(tmpdir, 'model.pt')
with open(ckpt_file, 'w') as f:
    f.write('dummy checkpoint')

# Create a checkpoint directory with latest.pt
ckpt_dir = os.path.join(tmpdir, 'checkpoints')
os.makedirs(ckpt_dir, exist_ok=True)
latest_file = os.path.join(ckpt_dir, 'latest.pt')
with open(latest_file, 'w') as f:
    f.write('dummy latest checkpoint')

# Create a checkpoints_file listing checkpoint paths
checkpoints_file = os.path.join(tmpdir, 'checkpoints.txt')
with open(checkpoints_file, 'w') as f:
    f.write(f'{ckpt_file}\n')
    f.write('# This is a comment\n')
    f.write(f'{ckpt_dir}\n')

# Build argparse.Namespace with required attributes
args_namespace = argparse.Namespace(
    checkpoint=[ckpt_file],
    checkpoints_file=checkpoints_file
)

ARGS = {
    'args': args_namespace
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
resolve_checkpoint_path Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/evaluation/evaluate_on_test_sets.py
code served (permissive licence) · get_code("1ec2c07817c05ff1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()

# Create a checkpoint file named 'latest.pt' in the directory
latest_checkpoint_path = os.path.join(temp_dir, 'latest.pt')
with open(latest_checkpoint_path, 'w') as f:
    f.write('')

# Test case 1: path is a directory containing latest.pt
ARGS = {
    'path': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
resolve_gated_deltanet_model_kwargs Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/models/factory.py
code served (permissive licence) · get_code("e91f6d66beeca4bb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import argparse

# Create a minimal argparse.Namespace with all expected attributes
args = argparse.Namespace(
    ff_mult=4,
    gated_deltanet_head_dim=48,
    gated_deltanet_expand_v=2.0,
    gated_deltanet_use_gate=True,
    gated_deltanet_use_short_conv=True,
    gated_deltanet_allow_neg_eigval=True,
    gated_deltanet_conv_size=4,
    gated_deltanet_attn_mode='chunk',
    gated_deltanet_norm_eps=1e-06
)

ARGS = {'args': args}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
rotate_half Ran
ran; our contract draft was wrong, not the code
Benjamin-Walker/Chess-World-Model/models/transformer.py
code served (permissive licence) · get_code("0d5b4082f0e044b9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function takes a tensor and performs a rotation operation on it.
# It expects a tensor with shape (..., feature_dim) where feature_dim is even.
# The function slices the last dimension into even and odd indices,
# stacks them, and flattens back.

# Create a simple tensor with shape (2, 4) - batch of 2, feature dim of 4
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
run_in_background Ran
ran, honoured the contract we drafted
niklasf/python-chess/chess/engine.py
pointer only (licence: GPL-3.0) · get_code("e1b2a8eaa981880d")
How it ran

The body of this sample is withheld: GPL-3.0, as GitHub detects it on this paper's repository, niklasf/python-chess, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

compute_state_loss Not yet run
unverified (harvested, no recorded run)
Benjamin-Walker/Chess-World-Model/evaluation/metrics.py
code served (permissive licence) · get_code("b62c82168a1e521d")
How it ran

no invocation recorded.

resolve_mamba_model_kwargs Not yet run
unverified (harvested, no recorded run)
Benjamin-Walker/Chess-World-Model/models/factory.py
code served (permissive licence) · get_code("2591ab122614aa59")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 19 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

World models require state tracking, which is the ability to maintain a correct latent state across action sequences. Existing benchmarks are often synthetic or languagebased, limiting their value as tests of structured state updates in realistic domains. We introduce CHESS-WORLD-MODEL, a large-scale state-tracking benchmark built from 10 million real chess games, where models predict the exact board state reached after a sequence of legal moves. Alongside a held-out real-game split, we include an out-of-distribution split from uniformly random legal play, which tests whether models learn the transition rules rather than shortcuts from common human positions. Prior theoretical and empirical work has shown that Transformers struggle to state-track, while input-dependent linear RNNs require expressive state-transition matrices to do so. We therefore benchmark a causal Transformer, block-diagonal SLiCE, Mamba-3, and Gated DeltaNet with negative eigenvalues under a matched interface and training protocol. The recurrent models strongly outperform the Transformer at 3 and 8 million parameters. Real-game performance saturates above 18 million parameters, but the random-uniform split remains discriminative up to 40 million, exposing failures otherwise hidden by scale. Additionally, ablations show that less expressive state-transition mechanisms reduce performance on the out-of-distribution split for all three recurrent models. Together, these results establish CHESS-WORLD-MODEL as a practical large-scale benchmark for state tracking that exposes failures model scale would otherwise conceal.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2605.30100")
get_code_for_paper("2605.30100")
have("2605.30100")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2605.30100.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2605.30100.svg)](https://syntology.ai/paper/2605.30100)

Connect an agent — have() is free.