SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.24150 · 2026

A visual observation on the geometry of UMAP projections of the difference vectors of antonym and synonym word pair embeddings

Rami Luisto

arXiv · PDF · Open in the Atlas

Code that ran

19 of 25 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; 1 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 17 failures of our instrument; 6 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/2603.24150.json · badge: /api/ran/2603.24150.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
ramiluisto/CuriousSwirl canonical (classifier label)
classifier basis: census
19 ran · 6 no recorded run
FunctionStatusWhere it lives
balance_classes Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/classify.py
code served (permissive licence) · get_code("4e66903f7714773f")
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 numpy as np
import logging

logging.basicConfig(level=logging.INFO)

# Create sample data for different classes
# Class 0: 5 samples of 4 features each
X_class_0 = np.random.randn(5, 4).astype(np.float32)

# Class 1: 3 samples of 4 features each (smallest class)
X_class_1 = np.random.randn(3, 4).astype(np.float32)

# Class 2: 7 samples of 4 features each
X_class_2 = np.random.randn(7, 4).astype(np.float32)

ARGS = {
    "X_list": [X_class_0, X_class_1, X_class_2],
    "class_names": ["class_A", "class_B", "class_C"],
    "random_state": 42,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
build_features Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ramiluisto/CuriousSwirl/semsim/classify.py
code served (permissive licence) · get_code("65507ab8bdad9e9f")
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 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 numpy as np

# Create small embedding pairs
emb1_1 = np.random.randn(4)  # embedding of dim 4
emb1_2 = np.random.randn(4)

emb2_1 = np.random.randn(4)
emb2_2 = np.random.randn(4)

emb_pairs = [
    (emb1_1, emb1_2),
    (emb2_1, emb2_2),
]

ARGS = {
    'emb_pairs': emb_pairs,
    'input_type': 'difference',
}
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 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
build_projection_params Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/projection_cache.py
code served (permissive licence) · get_code("aea8a7d36b40146f")
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 numpy as np
from typing import List

torch.manual_seed(0)

ARGS = {
    'pair_types': ['type_a', 'type_b', 'type_c'],
    'symmetrize': True,
    'standardize': False,
    'metric': 'euclidean',
}
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 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
compute_pca_projection Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ramiluisto/CuriousSwirl/semsim/projections.py
code served (permissive licence) · get_code("51e4bf7eff968db9")
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 numpy as np
import logging

logging.basicConfig(level=logging.INFO)

# Create a small feature matrix with distinct dimensions
# n_samples=7, n_features=5 (more features than components we'll request)
X = np.random.randn(7, 5).astype(np.float32)

ARGS = {
    "X": X,
    "max_components": 4,
    "n_components": 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — 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
cosine_similarity Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/stats.py
code served (permissive licence) · get_code("3500710b647d99ab")
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 numpy as np

# Create two sample numpy arrays for the cosine similarity function
v1 = np.array([1.0, 2.0, 3.0, 4.0])
v2 = np.array([2.0, 3.0, 4.0, 5.0])

ARGS = {
    "v1": v1,
    "v2": v2,
}
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 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
dot_product Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/stats.py
code served (permissive licence) · get_code("9a9338b265d08fe3")
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 numpy as np

# Create two small numpy vectors for the dot product
v1 = np.array([1.0, 2.0, 3.0, 4.0])
v2 = np.array([5.0, 6.0, 7.0, 8.0])

ARGS = {
    "v1": v1,
    "v2": v2,
}
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 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
euclidean_distance Ran
ran, honoured the contract we drafted
ramiluisto/CuriousSwirl/semsim/stats.py
code served (permissive licence) · get_code("45e2b883e4e52baa")
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 numpy as np

# Create two small numpy arrays of different sizes to test the euclidean distance function
v1 = np.array([1.0, 2.0, 3.0, 4.0])
v2 = np.array([5.0, 6.0, 7.0, 8.0])

ARGS = {
    "v1": v1,
    "v2": v2,
}
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 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
extract_transformer_embeddings Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/embeddings.py
code served (permissive licence) · get_code("db18cbb63ad64fdb")
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
import numpy as np
from typing import Set, Dict
import logging

torch.manual_seed(0)
logging.basicConfig(level=logging.INFO)

# Create a minimal mock tokenizer
class MockTokenizer:
    def __init__(self):
        self.vocab_size = 1000
    
    def encode(self, text, add_special_tokens=False):
        # Simple mock: return consistent token ids based on text hash
        # Ensure single-token variants map to single ids
        hash_val = hash(text) % (self.vocab_size - 1)
        return [hash_val]

# Create a minimal transformer model with embeddings
class MockTransformer(nn.Module):
    def __init__(self):
        super().__init__()
        self.embeddings = nn.Module()
        # Add word_embeddings as a simple embedding layer
        self.embeddings.word_embeddings = nn.Embedding(1000, 16)
    
    def eval(self):
        return self
    
    def to(self, device):
        return super().to(device)

# Mock AutoModel and AutoTokenizer
class MockAutoModel:
    @staticmethod
    def from_pretrained(model_name):
        return MockTransformer()

class MockAutoTokenizer:
    @staticmethod
    def from_pretrained(model_name):
        return MockTokenizer()

# Patch transformers module before import in function
import sys
class MockTransformersModule:
    AutoModel = MockAutoModel
    AutoTokenizer = MockAutoTokenizer

sys.modules['transformers'] = MockTransformersModule()

# Create test vocabulary
vocabulary = {'hello', 'world', 'test', 'python', 'embedding'}

ARGS = {
    'model_name': 'bert-base-uncased',
    'vocabulary': vocabulary,
    'device': 'cpu',
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, transformers — 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
generate_antonym_pairs Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/pairs.py
code served (permissive licence) · get_code("733c302dd47abcb2")
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 logging
from typing import Dict, List, Set, Tuple
import numpy as np

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Create sample data
stuttgart_pairs = [
    ("big", "small"),
    ("hot", "cold"),
    ("fast", "slow"),
]

curated_pairs = [
    ("happy", "sad"),
    ("light", "dark"),
    ("big", "tiny"),  # Will be deduplicated with stuttgart pair
]

# Create embeddings dictionary with numpy arrays for words that appear in pairs
embeddings = {
    "big": np.random.randn(4),
    "small": np.random.randn(4),
    "hot": np.random.randn(4),
    "cold": np.random.randn(4),
    "fast": np.random.randn(4),
    "slow": np.random.randn(4),
    "happy": np.random.randn(4),
    "sad": np.random.randn(4),
    "light": np.random.randn(4),
    "dark": np.random.randn(4),
    "tiny": np.random.randn(4),
}

ARGS = {
    "stuttgart_pairs": stuttgart_pairs,
    "curated_pairs": curated_pairs,
    "embeddings": embeddings,
}
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 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
generate_shuffled_pairs Ran
ran on a synthesized input
ramiluisto/CuriousSwirl/semsim/pairs.py
code served (permissive licence) · get_code("f5c3c779c57bee35")
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 random
from typing import Dict, List, Set, Tuple

# Construct valid arguments for generate_shuffled_pairs

# original_pairs: A list of (word1, word2) tuples representing semantic pairs
original_pairs = [
    ("cat", "dog"),
    ("car", "bicycle"),
    ("happy", "sad"),
    ("big", "small"),
]

# semantic_exclusion_set: Set of canonical (sorted) pair tuples to exclude
# These are pairs we don't want to appear in the shuffled output
semantic_exclusion_set = {
    ("cat", "dog"),
    ("big", "small"),
}

# random_state: Random seed for reproducibility
random_state = 42

ARGS = {
    "original_pairs": original_pairs,
    "semantic_exclusion_set": semantic_exclusion_set,
    "random_state": random_state,
}
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 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
generate_synonym_pairs Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/pairs.py
code served (permissive licence) · get_code("524abf9bd354578b")
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 logging
import numpy as np
from typing import Dict, List, Tuple

logging.basicConfig(level=logging.INFO)

# Define the PairList type
PairList = List[Tuple[str, str]]

# Create sample Stuttgart pairs - some with embeddings, some without
stuttgart_pairs = [
    ("cat", "feline"),
    ("dog", "canine"),
    ("happy", "joyful"),
    ("big", "large"),
    ("unknown_word_1", "unknown_word_2"),
    ("small", "tiny"),
]

# Create embeddings dict with numpy arrays
# Only include embeddings for some of the words
embeddings = {
    "cat": np.random.randn(4),
    "feline": np.random.randn(4),
    "dog": np.random.randn(4),
    "canine": np.random.randn(4),
    "happy": np.random.randn(4),
    "joyful": np.random.randn(4),
    "big": np.random.randn(4),
    "large": np.random.randn(4),
    "small": np.random.randn(4),
    "tiny": np.random.randn(4),
    # Note: "unknown_word_1" and "unknown_word_2" are NOT in embeddings
}

ARGS = {
    "stuttgart_pairs": stuttgart_pairs,
    "embeddings": embeddings,
}
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 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
get_nouns Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/data.py
code served (permissive licence) · get_code("b90f750b418d59aa")
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 set. 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

# Create a sample DataFrame with the structure expected by get_nouns
# The function looks for rows where 'pos' == 'nouns' and extracts 'word1' and 'word2'
df = pd.DataFrame({
    'pos': ['nouns', 'nouns', 'verbs', 'nouns', 'adjectives'],
    'word1': ['cat', 'dog', 'run', 'bird', 'happy'],
    'word2': ['feline', 'canine', 'sprint', 'avian', 'joyful']
})

ARGS = {
    'df': df
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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_vocabulary Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/data.py
code served (permissive licence) · get_code("f60cc27077f1980b")
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 set. 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 torch

torch.manual_seed(0)

# Create a simple DataFrame with 'word1' and 'word2' columns
df = pd.DataFrame({
    'word1': ['hello', 'world', 'foo'],
    'word2': ['test', 'example', 'bar']
})

ARGS = {
    'df': df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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_embeddings Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/embeddings.py
code served (permissive licence) · get_code("e298f4b865e41113")
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 tempfile
import pickle
from pathlib import Path
import numpy as np

# Create a temporary directory for test files
tmpdir = Path(tempfile.gettempdir())

# Create a pickle file with embeddings
pickle_path = tmpdir / "test_embeddings.pkl"
embeddings_dict = {
    "hello": np.array([1.0, 2.0, 3.0, 4.0]),
    "world": np.array([5.0, 6.0, 7.0, 8.0]),
    "test": np.array([9.0, 10.0, 11.0, 12.0])
}
with open(pickle_path, 'wb') as f:
    pickle.dump(embeddings_dict, f)

ARGS = {
    "path": pickle_path
}
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 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_stuttgart Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/data.py
code served (permissive licence) · get_code("6f9a74ca00952c93")
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 DataFrame. 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 tempfile
from pathlib import Path
import pandas as pd

# Create a temporary directory with mock Stuttgart dataset files
tmpdir = tempfile.TemporaryDirectory()
tmppath = Path(tmpdir.name)

# Create mock Stuttgart dataset files
STUTTGART_FILES = ['adjectives.train', 'adjectives.test', 'adjectives.val', 'nouns.train', 'nouns.test', 'nouns.val', 'verbs.train', 'verbs.test', 'verbs.val']

for filename in STUTTGART_FILES:
    filepath = tmppath / filename
    with filepath.open('w', encoding='utf-8') as f:
        f.write("happy\tsad\t1\n")
        f.write("big\tsmall\t1\n")
        f.write("fast\tquick\t0\n")

ARGS = {
    'stuttgart_dir': tmppath
}
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 DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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_confusion_matrix Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ramiluisto/CuriousSwirl/semsim/plots.py
code served (permissive licence) · get_code("2523a48b4c15f71f")
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 Figure. 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 logging
from pathlib import Path
from typing import Dict, List, Optional, Union
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Create a small confusion matrix (4x4 for 4 classes)
cm = np.array([
    [45, 5, 2, 1],
    [3, 48, 4, 2],
    [2, 6, 42, 3],
    [1, 1, 2, 51]
], dtype=np.int32)

# Define class names
class_names = ['Class_A', 'Class_B', 'Class_C', 'Class_D']

# Define title
title = 'Test Confusion Matrix'

# Define save path (using a temporary path, but we won't actually save)
save_path = Path('/tmp/confusion_matrix.png')

# Define figsize
figsize = (10, 8)

# Define dpi
dpi = 100

ARGS = {
    'cm': cm,
    'class_names': class_names,
    'title': title,
    'save_path': save_path,
    'figsize': figsize,
    'dpi': dpi
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy, seaborn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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_metric_distributions Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ramiluisto/CuriousSwirl/semsim/plots.py
code served (permissive licence) · get_code("28c4a8b9ae88c062")
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 Figure. 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 logging
from pathlib import Path
import numpy as np
import matplotlib
matplotlib.use('Agg')  # Non-interactive backend
import matplotlib.pyplot as plt

logging.basicConfig(level=logging.INFO)

# Create sample metric data for different pair types
metrics_by_type = {
    'type_A': np.random.randn(15),
    'type_B': np.random.randn(20),
    'type_C': np.random.randn(18),
}

metric_name = 'correlation_score'
title = 'Metric Distribution Across Pair Types'
save_path = None  # No file I/O in sandbox
figsize = (10, 6)
dpi = 300

ARGS = {
    'metrics_by_type': metrics_by_type,
    'metric_name': metric_name,
    'title': title,
    'save_path': save_path,
    'figsize': figsize,
    'dpi': dpi,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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_pvalue_heatmap Ran
ran; our contract draft was wrong, not the code
ramiluisto/CuriousSwirl/semsim/plots.py
code served (permissive licence) · get_code("720477544f913473")
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 Figure. 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 logging
from pathlib import Path
from typing import Dict, List, Optional, Union
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

# Set up logging
logging.basicConfig(level=logging.INFO)

# Create valid arguments for plot_pvalue_heatmap
pvalues = {
    "comparison_A": {
        "metric_1": 0.01,
        "metric_2": 0.03,
        "metric_3": 0.001,
    },
    "comparison_B": {
        "metric_1": 0.02,
        "metric_2": 0.015,
        "metric_3": 0.04,
    },
    "comparison_C": {
        "metric_1": 0.005,
        "metric_2": 0.025,
        "metric_3": 0.012,
    },
}

title = "p-value Heatmap"
save_path = None  # No file I/O
figsize = (10, 8)
dpi = 300

ARGS = {
    "pvalues": pvalues,
    "title": title,
    "save_path": save_path,
    "figsize": figsize,
    "dpi": dpi,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy, seaborn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Figure — a Figure output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Figure 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
symmetrize_features Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ramiluisto/CuriousSwirl/semsim/classify.py
code served (permissive licence) · get_code("b06f15870f2100a1")
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 float, shape (4, 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 numpy as np

# Create a small feature matrix for testing
# Using shape (2, 4) - 2 samples, 4 features
X = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
], dtype=np.float32)

ARGS = {
    'X': X,
    'base_input_type': 'difference'
}
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 float, shape (4, 4) — shape (4, 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
compute_tsne_projection Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/semsim/projections.py
code served (permissive licence) · get_code("8c2201a9c54f1361")
How it ran

no invocation recorded.

compute_umap_projection Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/semsim/projections.py
code served (permissive licence) · get_code("4776fba74086339d")
How it ran

no invocation recorded.

extract_word2vec_embeddings Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/semsim/embeddings.py
code served (permissive licence) · get_code("0d46a4f5b229b35f")
How it ran

no invocation recorded.

get_embedding_path Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/config.py
code served (permissive licence) · get_code("f087285272edbb72")
How it ran

no invocation recorded.

get_model_slug Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/config.py
code served (permissive licence) · get_code("d0ad9ca728f04022")
How it ran

no invocation recorded.

get_pair_path Not yet run
unverified (harvested, no recorded run)
ramiluisto/CuriousSwirl/config.py
code served (permissive licence) · get_code("39d62c2346fbb8ed")
How it ran

no invocation recorded.

Also present in other papers' code

Of 25 samples lifted from this paper's linked repositories, 25 carry a function identity and 0 do not. Of the 25 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 25 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

Antonyms, or opposites, are sometimes defined as word pairs that have all of the same contextually relevant properties but one. Seeing how transformer models seem to encode concepts as directions, this begs the question if one can detect "antonymity" in the geometry of the embedding vectors of word pairs, especially based on their difference vectors. Such geometrical studies are then naturally contrasted by comparing antonymic pairs to their opposites; synonyms. This paper started as an exploratory project on the complexity of the systems needed to detect the geometry of the embedding vectors of antonymic word pairs. What we now report is a curious "swirl" that appears across embedding models in a somewhat specific projection configuration.

For agents

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

get_harvested_code_for_paper("2603.24150")
get_code_for_paper("2603.24150")
have("2603.24150")

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

curl https://syntology.ai/api/ran/2603.24150.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.24150.svg)](https://syntology.ai/paper/2603.24150)

Connect an agent — have() is free.