SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.03214 · EMNLP · 2025

iVISPAR -An Interactive Visual-Spatial Reasoning Benchmark for VLMs

Elia Bruni, Mohamad Ballout, Julius Mayer, Serwan Jassim, † Farbod, Nosrat Nezami

arXiv · PDF · Open in the Atlas

Code that ran

15 of 16 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; 0 honoured the contract we drafted, 0 violated it and 15 have no contract verdict; 0 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2502.03214.json · badge: /api/ran/2502.03214.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
SharkyBamboozle/iVISPAR canonical (classifier label)
classifier basis not recorded on the link
15 ran · 1 no recorded run
FunctionStatusWhere it lives
a_star Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/find_shortest_move_sequence.py
code served (permissive licence) · get_code("fd2e72f2c704d0e5")
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 3, dtype family int; 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 numpy as np

# Construct valid arguments for the a_star function
# This is a sliding tile puzzle solver using A* search

# For a 3x3 puzzle with a simple configuration
n = 3

# Initial state: 2 tiles at positions [0,0] and [1,1]
initial_state = [
    [0, 0],
    [1, 1]
]

# Goal state: tiles should be at [2,0] and [2,1]
goal_state = [
    [2, 0],
    [2, 1]
]

# Maximum search depth
max_depth = 10

ARGS = {
    "n": n,
    "initial_state": initial_state,
    "goal_state": goal_state,
    "max_depth": max_depth
}
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 3, dtype family int — 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
a_star Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Evaluate/calculate_shortest_path_length.py
code served (permissive licence) · get_code("bc89968ae39cddbb")
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 3, dtype family int; 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 numpy as np

# This is an A* solver for sliding tile puzzles
# The function takes:
# - n: board size (n x n)
# - initial_state: list of [x, y] positions for tiles
# - goal_state: list of [x, y] positions for goal

# For a 3x3 puzzle with 3 tiles, we need valid positions
# Let's create a small 3x3 puzzle with 3 tiles

n = 3

# Initial state: 3 tiles at positions (0,0), (0,1), (1,0)
initial_state = [
    [0, 0],
    [0, 1],
    [1, 0]
]

# Goal state: 3 tiles at positions (1,1), (1,2), (2,0)
goal_state = [
    [1, 1],
    [1, 2],
    [2, 0]
]

ARGS = {
    "n": n,
    "initial_state": initial_state,
    "goal_state": goal_state
}
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 3, dtype family int — 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
calculate_manhattan_heuristic Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/find_shortest_move_sequence.py
code served (permissive licence) · get_code("7f2ed2be9be186fd")
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 int, ndim 0, dtype family int, 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 numpy as np

# Create small sample data for initial and goal states
# Shape: (batch_size=2, dimensions=4)
initial_states = np.array([
    [0.0, 1.0, 2.0, 3.0],
    [1.0, 2.0, 3.0, 4.0]
], dtype=np.float32)

goal_states = np.array([
    [2.0, 3.0, 1.0, 5.0],
    [3.0, 1.0, 5.0, 2.0]
], dtype=np.float32)

ARGS = {
    'initial_states': initial_states,
    'goal_states': goal_states
}
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 int, ndim 0, dtype family int, 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
count_inversions Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/check_is_STP_solvable.py
code served (permissive licence) · get_code("c00a6a002972a2e7")
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 int, ndim 0, dtype family int, 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

torch.manual_seed(0)

# The function count_inversions takes:
# - tiles: a list of positions (tuples with 2 elements each)
# - board_size: an integer representing the board size

# Example: a 3x3 puzzle with tiles at various positions
# We'll create a simple configuration where tiles is a list of (row, col) tuples
tiles = [(0, 0), (1, 1), (2, 0), (0, 2), (1, 0)]
board_size = 3

ARGS = {
    "tiles": tiles,
    "board_size": board_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 int, ndim 0, dtype family int, 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
generate_goal_state Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/generate_STP_configs.py
code served (permissive licence) · get_code("f7f93ad1ff5bc96d")
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 ndarray, ndim 2, dtype family int, shape (15, 2) 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 numpy as np

# Analyze the function:
# - num_elements: int, represents number of tiles to place
# - board_size: int, represents the size of the board (n x n)
# 
# The function creates coordinates for a sliding tile puzzle goal state
# It uses reversed(range(board_size)) and range(board_size) to generate positions
# Returns an np.ndarray of shape (num_elements, 2) with dtype int
#
# Valid test case: 4x4 board with 15 elements (standard sliding puzzle)

ARGS = {
    "num_elements": 15,
    "board_size": 4
}
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 ndarray, ndim 2, dtype family int, shape (15, 2) — shape (15, 2) 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
generate_random_invalid_path Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/find_random_move_sequence.py
code served (permissive licence) · get_code("ee60b7de859d8f6d")
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 3, dtype family int; 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 numpy as np
import random

# Set random seed for reproducibility
random.seed(42)
np.random.seed(42)

# Create a small initial state for a 4x4 sliding tile puzzle
# Each tile is represented as [x, y] coordinates
initial_state = [
    [0, 0],  # Tile 1
    [1, 0],  # Tile 2
    [2, 0],  # Tile 3
    [0, 1],  # Tile 4
]

ARGS = {
    "n": 4,
    "initial_state": initial_state,
    "max_steps": 10,
}
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 3, dtype family int — 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
generate_random_valid_path Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/find_random_move_sequence.py
code served (permissive licence) · get_code("3da90419aa44ae40")
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 3, dtype family int; 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 numpy as np
import random

# Set seed for reproducibility
random.seed(0)
np.random.seed(0)

# Create a valid initial state for a sliding tile puzzle
# For a 4x4 board, we need 15 tiles (leaving one empty space)
# Each tile is represented as [x, y] coordinates
n = 4
initial_state = [
    [0, 0], [1, 0], [2, 0], [3, 0],
    [0, 1], [1, 1], [2, 1], [3, 1],
    [0, 2], [1, 2], [2, 2], [3, 2],
    [0, 3], [1, 3], [2, 3]
    # Note: [3, 3] is empty (not included in state)
]

ARGS = {
    "n": n,
    "initial_state": initial_state,
    "max_steps": 10
}
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 3, dtype family int — 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
get_blank_position Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/check_is_STP_solvable.py
code served (permissive licence) · get_code("0ecaec1e92634ba7")
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 int, ndim 0, dtype family int, 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

torch.manual_seed(0)

# The function expects:
# - state: something with a .tolist() method (like a numpy array or tensor)
# - board_size: an integer

# Create a state representing a sliding puzzle board
# For a 3x3 puzzle with blank at position (1, 2), we'd have 8 tiles and 1 blank
# The state should be a 2D array/tensor where all positions except one are filled

board_size = 3
# Create a state: positions (0,0) to (2,2) with one position missing
# Let's say blank is at position (1, 2) - so we exclude that
state_list = []
for row in range(board_size):
    for col in range(board_size):
        if not (row == 1 and col == 2):  # Skip the blank position
            state_list.append([row, col])

state = np.array(state_list)

ARGS = {
    'state': state,
    'board_size': board_size
}
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 int, ndim 0, dtype family int, 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
get_neighbors Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/find_random_move_sequence.py
code served (permissive licence) · get_code("6cf8ccfcf1db2c95")
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 3, dtype family int; 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

# Analyze the function:
# - state: appears to be a list of lists, where each inner list is [x, y] coordinates
# - n: an integer representing the grid size (n x n)
# The function generates neighboring states by sliding tiles into empty positions

# For a simple 2x2 puzzle, we need:
# - A state represented as a list of [x, y] positions
# - n representing the grid dimension

# Create a small valid state for a 2x2 grid
# In a sliding puzzle, we have n*n - 1 tiles (one empty position)
# For a 2x2 grid, we have 3 tiles
# Let's represent positions as [x, y] where x, y are in range [0, n)
state = [
    [0, 0],  # tile at position (0, 0)
    [0, 1],  # tile at position (0, 1)
    [1, 0],  # tile at position (1, 0)
    # (1, 1) is the empty position - not in state
]

n = 2

ARGS = {
    "state": state,
    "n": n
}
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 3, dtype family int — 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
get_tile_color Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/encode_config_to_json.py
code served (permissive licence) · get_code("eb9ce1cba6692683")
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 numpy as np

ARGS = {
    'tile_index': 5,
    'board_size': 5
}
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 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
has_unfilled_c1_bins_np Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/configuration_utilities.py
code served (permissive licence) · get_code("5294dea7fd5c07db")
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 numpy as np

# Construct arguments for has_unfilled_c1_bins_np
# The function expects:
# - complexity_bins: a 3D numpy array, indexed by [:, c1_indices, :]
# - found_complexity: a scalar value to compare against c1_range
# - complexity_bin_size: a scalar for comparison
# - c1_range: an iterable of values to filter by

# Create a small 3D complexity_bins array
# Shape: (batch=2, c1_dim=4, feature_dim=3)
complexity_bins = np.array([
    [[1, 2, 3], [4, 5, 6], [7, 8, 9], [10, 11, 12]],
    [[2, 3, 4], [5, 6, 7], [8, 9, 10], [11, 12, 13]]
], dtype=np.float32)

# found_complexity: a scalar that will be compared against c1_range values
found_complexity = 2.5

# complexity_bin_size: threshold for comparison
complexity_bin_size = 20.0

# c1_range: values to filter indices by
c1_range = np.array([1.0, 2.0, 3.0, 4.0])

ARGS = {
    "complexity_bins": complexity_bins,
    "found_complexity": found_complexity,
    "complexity_bin_size": complexity_bin_size,
    "c1_range": c1_range,
}
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 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
is_solvable Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/check_is_STP_solvable.py
code served (permissive licence) · get_code("e0d7d25dfd66dd61")
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 numpy as np

# Create a valid state for a 3x3 sliding puzzle (8-puzzle)
# State should be a 2D numpy array of shape (n, 2) where n = board_size^2 - 1
# Each row represents (row, col) position of a tile (excluding the blank)

board_size = 3
# For a 3x3 puzzle, we have 8 tiles (9 positions total, 1 blank)
# Create a solvable configuration
state = np.array([
    [0, 0],  # tile 1 at position (0, 0)
    [0, 1],  # tile 2 at position (0, 1)
    [0, 2],  # tile 3 at position (0, 2)
    [1, 0],  # tile 4 at position (1, 0)
    [1, 1],  # tile 5 at position (1, 1)
    [1, 2],  # tile 6 at position (1, 2)
    [2, 0],  # tile 7 at position (2, 0)
    [2, 1],  # tile 8 at position (2, 1)
    # blank at position (2, 2)
], dtype=np.int32)

ARGS = {
    "state": state,
    "goal_state": None
}
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 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
load_params_from_json Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/configuration_utilities.py
code served (permissive licence) · get_code("4c1d2ded97e0ed89")
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 json
import os
import tempfile

# Create a temporary directory structure and JSON file for testing
temp_dir = tempfile.mkdtemp()
data_dir = os.path.join(temp_dir, 'Data', 'Params')
os.makedirs(data_dir, exist_ok=True)

# Create a sample JSON parameters file
sample_params = {
    'learning_rate': 0.001,
    'batch_size': 32,
    'epochs': 10,
    'model_name': 'test_model'
}

json_file_path = os.path.join(data_dir, 'test_params.json')
with open(json_file_path, 'w') as f:
    json.dump(sample_params, f)

# Mock the __file__ variable to point to our temp structure
import sys
from unittest.mock import patch

# Create a minimal test by passing just the filename
ARGS = {
    'file_name': 'test_params.json'
}
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
sample_board_states Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/configuration_utilities.py
code served (permissive licence) · get_code("f5e28286301de4d3")
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 ndarray, ndim 2, dtype family int, shape (3, 2) 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 numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# The function takes two parameters:
# - num_geoms: number of geometries to sample (int)
# - board_size: size of the board (int)
# 
# The function samples num_geoms random positions on a board_size x board_size board
# without replacement, so num_geoms must be <= board_size ** 2

num_geoms = 3
board_size = 4

ARGS = {
    "num_geoms": num_geoms,
    "board_size": board_size,
}
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 ndarray, ndim 2, dtype family int, shape (3, 2) — shape (3, 2) 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
translate_moves_to_commands Ran
ran on a synthesized input
SharkyBamboozle/iVISPAR/Source/Configure/encode_config_to_json.py
code served (permissive licence) · get_code("3521c890a4ffb8bb")
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 torch
import numpy as np

# Set seed for reproducibility
torch.manual_seed(0)

# Define the arguments for translate_moves_to_commands

# shortest_move_sequence: A 3D list representing board states for all tiles/objects
# Each element is a state, each state contains positions for multiple objects
# Using simple (x, y) coordinate tuples for positions
shortest_move_sequence = [
    # State 0: initial positions
    [(0, 0), (1, 1), (2, 2)],
    # State 1: first object moves right
    [(1, 0), (1, 1), (2, 2)],
    # State 2: second object moves up
    [(1, 0), (1, 2), (2, 2)],
    # State 3: third object moves left
    [(1, 0), (1, 2), (1, 2)],
]

# landmarks: A list of dictionaries with geom_nr, body, and color
landmarks = [
    {'geom_nr': 1, 'body': 'tile', 'color': 'red'},
    {'geom_nr': 2, 'body': 'tile', 'color': 'blue'},
    {'geom_nr': 3, 'body': 'tile', 'color': 'green'},
]

ARGS = {
    'shortest_move_sequence': shortest_move_sequence,
    'landmarks': landmarks,
}
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 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
expand_config_file Not yet run
unverified (harvested, no recorded run)
SharkyBamboozle/iVISPAR/Source/Configure/extend_config_parameter.py
code served (permissive licence) · get_code("77a3b9080a5db16c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 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 16 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

Vision-Language Models (VLMs) are known to struggle with spatial reasoning and visual alignment. To help overcome these limitations, we introduce iVISPAR, an interactive multimodal benchmark designed to evaluate the spatial reasoning capabilities of VLMs acting as agents. iVISPAR is based on a variant of the sliding tile puzzle-a classic problem that demands logical planning, spatial awareness, and multi-step reasoning. The benchmark supports visual 3D, 2D, and text-based input modalities, enabling comprehensive assessments of VLMs' planning and reasoning skills. We evaluate a broad suite of state-of-theart open-source and closed-source VLMs, comparing their performance while also providing optimal path solutions and a human baseline to assess the task's complexity and feasibility for humans. Results indicate that while VLMs perform better on 2D tasks compared to 3D or text-based settings, they struggle with complex spatial configurations and consistently fall short of human performance, illustrating the persistent challenge of visual alignment. This underscores critical gaps in current VLM capabilities, highlighting their limitations in achieving human-level cognition. Project website: https://microcosm.ai/ivispar.

For agents

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

get_harvested_code_for_paper("2502.03214")
get_code_for_paper("2502.03214")
have("2502.03214")

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

curl https://syntology.ai/api/ran/2502.03214.json

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

[![Syntology run record](https://syntology.ai/api/ran/2502.03214.svg)](https://syntology.ai/paper/2502.03214)

Connect an agent — have() is free.