SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.03343 · 2026

Neuro-Symbolic Decoding of Neural Activity

Jiajun Wu, Joy Hsu, Ehsan Adeli, Yanchen Wang

arXiv · PDF · Open in the Atlas

Code that ran

14 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 12 failures of our instrument; 1 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/2603.03343.json · badge: /api/ran/2603.03343.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
PPWangyc/neurona canonical (classifier label)
classifier basis: census
14 ran · 1 no recorded run
FunctionStatusWhere it lives
calculate_cost Ran
ran, honoured the contract we drafted
PPWangyc/neurona/src/create_dataset/create_bold5000_scene_graph.py
code served (permissive licence) · get_code("450b9f98a21dfdcd")
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 float, 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 numpy as np

# The function takes prompt_tokens, completion_tokens, and model as parameters
# prompt_tokens and completion_tokens are used in arithmetic operations (division, multiplication)
# model is a string that is compared against specific values

ARGS = {
    'prompt_tokens': 1000,
    'completion_tokens': 500,
    'model': 'gpt-4o-mini'
}
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 float, 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
check_balance Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/create_dataset/create_bold5000_scene_graph.py
code served (permissive licence) · get_code("24ae6638ad6d07bb")
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 datetime

class MockClient:
    def __init__(self):
        self.organization = "org-test123"
        self.api_key = "sk-test1234567890"

ARGS = {
    'client': MockClient()
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
convert_atlas_name Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/models/fmri/simple_cnn.py
code served (permissive licence) · get_code("160985394ca12189")
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 numpy as np

# This function takes a single string parameter and returns a converted atlas name
# based on a simple mapping. No complex objects needed.

ARGS = {
    'atlas_name': 'yeo17'
}
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 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
generate_descriptor_object_exists Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/dataset_utils.py
code served (permissive licence) · get_code("41664c8da8d27551")
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

# Construct valid argument for generate_descriptor_object_exists
subgraph_str = '<subject>red dog</subject><adjective>fluffy</adjective>'

ARGS = {
    'subgraph_str': subgraph_str
}
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
generate_object_exists Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/dataset_utils.py
code served (permissive licence) · get_code("31a1bdd26871de3f")
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 re

def replace_concepts(concept):
    concept = concept.replace(' ', '_')
    concept = concept.replace('-', '_')
    concept = concept.replace('.', '')
    concept = concept.replace("'", '')
    if concept == 'in':
        concept = '_in'
    if concept == 'with':
        concept = '_with'
    if concept == 'and':
        concept = '_and'
    if concept == 'or':
        concept = '_or'
    if concept == 'from':
        concept = '_from'
    if concept == 'is':
        concept = '_is'
    if concept == 'sits':
        concept = 'sit'
    if concept == 'sitting':
        concept = 'sit'
    if concept == 'have':
        concept = 'has'
    if concept == 'contains':
        concept = 'has'
    if concept == 'contain':
        concept = 'has'
    if concept == 'above':
        concept = 'on'
    if concept == 'near':
        concept = 'next_to'
    if concept == 'at':
        concept = '_in'
    if concept == 'carry':
        concept = 'hold'
    if concept == 'wait':
        concept = 'resting'
    if concept == 'perch':
        concept = 'resting'
    if concept == 'around':
        concept = 'surround'
    if concept == 'standing':
        concept = 'stand'
    if concept == 'perform':
        concept = 'display'
    if concept == 'drink':
        concept = 'eat'
    if concept == 'behind':
        concept = 'beside'
    return concept

def generate_object_exists(subgraph_str):
    obj = subgraph_str.split('<object>')[-1].split('</object>')[0]
    obj = replace_concepts(obj)
    obj_exists = 'exists(Object, lambda x: ' + obj + '(x))'
    return obj_exists

# Create test input for generate_object_exists
# The function expects a string with XML-like tags containing an object name
subgraph_str = '<object>dog</object>'

ARGS = {
    'subgraph_str': subgraph_str
}
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
get_concept_grounding_tensor Ran
ran, honoured the contract we drafted
PPWangyc/neurona/src/models/fmri/simple_cnn.py
code served (permissive licence) · get_code("3ae14b50f7b3c995")
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
import torch.nn as nn

torch.manual_seed(0)

# The function accesses outputs['execution_traces'][0][1][1].tensor
# So we need to construct a nested structure that matches this indexing

# Create a small tensor to be the grounding tensor
grounding_tensor = torch.randn(2, 4)

# Build the nested structure: execution_traces[0][1][1].tensor
# We need an object that has a .tensor attribute
class TensorHolder:
    def __init__(self, tensor):
        self.tensor = tensor

tensor_holder = TensorHolder(grounding_tensor)

# execution_traces[0][1][1] should be the tensor_holder
# execution_traces[0][1] should be a list or tuple containing it
inner_list = [None, tensor_holder]

# execution_traces[0] should be a list or tuple containing inner_list
outer_list = [None, inner_list]

# execution_traces should be a list containing outer_list
execution_traces = [outer_list]

# parsings is also required (though not used in function body)
parsings = []

outputs = {
    'execution_traces': execution_traces,
    'parsings': parsings
}

ARGS = {
    'outputs': outputs
}
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
get_npy_paths Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/loader/fqa.py
code served (permissive licence) · get_code("54578ef789c2744f")
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 os
import tempfile
import shutil

# Create a temporary directory structure with .npy files
temp_dir = tempfile.mkdtemp()
subject = "subject_01"
split = "train"

# Create the directory structure
split_dir = os.path.join(temp_dir, subject, split)
os.makedirs(split_dir, exist_ok=True)

# Create some dummy .npy files
import numpy as np
np.save(os.path.join(split_dir, "data_001.npy"), np.array([1, 2, 3]))
np.save(os.path.join(split_dir, "data_002.npy"), np.array([4, 5, 6]))
np.save(os.path.join(split_dir, "info.txt"), "not an npy file")

ARGS = {
    "data_dir": temp_dir,
    "subject": subject,
    "split": split
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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
load_atlas Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/models/fmri/simple_cnn.py
code served (permissive licence) · get_code("791729dfc9ce0813")
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 pandas as pd
import io

# Create a minimal CSV content that mimics an atlas file
csv_content = """Component,Yeo_networks17
1,1
2,1
3,2
4,2
5,3
6,3
"""

# Create a temporary file-like object with the CSV data
atlas_df = pd.read_csv(io.StringIO(csv_content))

# Write to a temporary path for the function to read
import tempfile
import os

temp_dir = tempfile.mkdtemp()
atlas_path = os.path.join(temp_dir, "atlas.csv")
atlas_df.to_csv(atlas_path, index=False)

ARGS = {
    'atlas_path': atlas_path,
    'atlas_name': 'Yeo_networks17'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — 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
load_best_model Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/utils.py
code served (permissive licence) · get_code("67124850e8a71839")
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 SimpleModel. 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 torch
import torch.nn as nn
import tempfile
import shutil

torch.manual_seed(0)

# Create a minimal model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

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

# Create a dummy model and save its state dict
dummy_model = SimpleModel()
train_hop_value = "1"
atlas_value = "test_atlas"
model_path = os.path.join(temp_dir, f'best_model_{train_hop_value}_{atlas_value}.pth')
torch.save(dummy_model.state_dict(), model_path)

# Create the model instance to load weights into
model = SimpleModel()

ARGS = {
    "model": model,
    "output_dir": temp_dir,
    "train_hop": train_hop_value,
    "atlas": atlas_value,
}
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 SimpleModel — a SimpleModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModel 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
move_batch_to_device Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/utils.py
code served (permissive licence) · get_code("0678ca15b4c19cb8")
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

torch.manual_seed(0)

# Create a batch dict with mixed content
batch = {
    "input_ids": torch.randn(2, 4),
    "attention_mask": torch.randn(2, 4),
    "labels": torch.randn(2, 3),
    "metadata": "some string that should not be moved",
    "count": 42,
}

# Set device to CPU
device = "cpu"

ARGS = {
    "batch": batch,
    "device": device,
}
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
plot_bar_chart Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
PPWangyc/neurona/src/utils/plot_utils.py
code served (permissive licence) · get_code("21e32e2becf034a7")
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 matplotlib.pyplot as plt
import seaborn as sns

# Create sample data dictionary with numeric values
data_dict = {
    'metric_a': 500,
    'metric_b': 1200,
    'metric_c': 800,
    'metric_d': 1500,
}

# Define the keys to plot
keys = ['metric_a', 'metric_b', 'metric_c', 'metric_d']

# Define display names for the bars
names = ['Metric A', 'Metric B', 'Metric C', 'Metric D']

# Create color palette
colors = sns.color_palette('flare', len(keys))

ARGS = {
    'data_dict': data_dict,
    'keys': keys,
    'names': names,
    'xlabel': 'Metrics',
    'ylabel': 'Count',
    'title': 'Sample Bar Chart',
    'colors': colors,
    'fontsize': 12,
    'linewidth': 1.5,
    'bar_width': 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, seaborn — 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
plot_pie_chart Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/plot_utils.py
code served (permissive licence) · get_code("6397dce231531d80")
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 matplotlib.pyplot as plt
import seaborn as sns

# Create a sample data dictionary
data_dict = {
    'category_a': 25,
    'category_b': 35,
    'category_c': 20,
    'category_d': 20
}

# Define keys to extract from data_dict
keys = ['category_a', 'category_b', 'category_c', 'category_d']

# Define display names for the pie slices
names = ['Category A', 'Category B', 'Category C', 'Category D']

# Define title for the chart
title = 'Sample Distribution'

# Define custom colors (optional, but providing them to test that parameter)
colors = ['#FF6B6B', '#4ECDC4', '#45B7D1', '#FFA07A']

# Define fontsize
fontsize = 14

ARGS = {
    'data_dict': data_dict,
    'keys': keys,
    'names': names,
    'title': title,
    'colors': colors,
    'fontsize': fontsize
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, seaborn — 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
process_args Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/utils.py
code served (permissive licence) · get_code("7d8ec730be793b26")
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 os
import tempfile

class Args:
    def __init__(self):
        self.dataset = "test_dataset"
        self.subject = "subject_01"
        self.output_dir = tempfile.gettempdir()
        self.ground_attr_region = True
        self.ground_attr_relation = True
        self.ground_rel_region = True
        self.ground_rel_relation = True
        self.ground_rel_guided_relation = "guided_test"
        self.atlas = "MNI152"
        self.train_hop = 2
        self.selected_regions = ["region_a", "region_b"]
        self.removed_regions = None

args_obj = Args()

ARGS = {
    "args": args_obj,
    "make_dir": True
}
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
replace_concepts Ran
ran; our contract draft was wrong, not the code
PPWangyc/neurona/src/utils/dataset_utils.py
code served (permissive licence) · get_code("33ec8a7891fbd7a5")
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)

ARGS = {
    'concept': 'hello world'
}
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
process_image Not yet run
unverified (harvested, no recorded run)
PPWangyc/neurona/src/create_dataset/create_bold5000_scene_graph.py
code served (permissive licence) · get_code("91cb92ed1cdd73a8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 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 15 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

We propose NEURONA, a neuro-symbolic framework for fMRI decoding and concept grounding in neural activity. Leveraging image-and video-based fMRI question-answering datasets, NEURONA learns to decode interacting concepts from visual stimuli based on patterns of fMRI responses, integrating symbolic reasoning and compositional execution with fMRI grounding across brain regions. We demonstrate that incorporating structural priors (e.g., compositional predicateargument dependencies between concepts) into the decoding process significantly improves both decoding accuracy over precise queries, and notably, generalization to unseen queries at test time. With NEURONA, we highlight neuro-symbolic frameworks as promising tools for understanding neural activity. Our code is available at https://github.com/PPWangyc/neurona. * Equal contribution. † Equal mentorship.

For agents

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

get_harvested_code_for_paper("2603.03343")
get_code_for_paper("2603.03343")
have("2603.03343")

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

curl https://syntology.ai/api/ran/2603.03343.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.03343.svg)](https://syntology.ai/paper/2603.03343)

Connect an agent — have() is free.