SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.05191 · ACL · 2023

COLA: Contextualized Commonsense Causal Reasoning from the Causal Inference Perspective

Yangqiu Song, Jiayao Zhang, Hongming Zhang, Ginny Wong, Simon See, Tianqing Fang, Weiqi Wang, Zhaowei Wang, Quyet Do

arXiv · PDF · Open in the Atlas

Code that ran

17 of 18 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 14 have no contract verdict; 3 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/2305.05191.json · badge: /api/ran/2305.05191.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
hkust-knowcomp/cola canonical (classifier label)
classifier basis: census
17 ran · 1 no recorded run
FunctionStatusWhere it lives
chunk_list Ran
ran on a synthesized input
hkust-knowcomp/cola/SRL_parsing.py
code served (permissive licence) · get_code("c0ae1f3a83a9162e")
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. 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

torch.manual_seed(0)

# chunk_list takes a sequence (list-like) and chunks it by batch_size
# It uses: len(seq_list), range, min, and list slicing
# seq_list should be any sequence that supports len() and slicing

# Create a simple list of items to chunk
seq_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
batch_size = 3

ARGS = {
    "seq_list": seq_list,
    "batch_size": batch_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 list — a list 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
decode_generation Ran
ran on a synthesized input
hkust-knowcomp/cola/covariate_util.py
code served (permissive licence) · get_code("88b6d26ae9b9bbcd")
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 numpy as np

class MockTokenizer:
    def __init__(self, pad_token_id=0):
        self.pad_token_id = pad_token_id
    
    def batch_decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False):
        # Mock decoder: convert token ids to simple text
        result = []
        for ids in token_ids:
            text = " ".join([f"tok_{int(id_)}" for id_ in ids if id_ != -100])
            result.append(text)
        return result

# Create mock tokenizer
tokenizer = MockTokenizer(pad_token_id=0)

# Create pred_ids: list of 2 arrays, each with shape (3, 5) - batch of 3, seq length 5
pred_ids = [
    np.array([[1, 2, 3, 4, 5], [6, 7, 8, 9, 10], [11, 12, 13, 14, 15]], dtype=np.int64),
    np.array([[16, 17, 18, 19, 20], [21, 22, 23, 24, 25], [26, 27, 28, 29, 30]], dtype=np.int64)
]

# Create input_ids: list of 2 arrays with same structure
input_ids = [
    np.array([[100, 101, 102, 103, 104], [105, 106, 107, 108, 109], [110, 111, 112, 113, 114]], dtype=np.int64),
    np.array([[115, 116, 117, 118, 119], [120, 121, 122, 123, 124], [125, 126, 127, 128, 129]], dtype=np.int64)
]

# Test with a non-gpt model
model_name_or_path = "bert-base-uncased"

ARGS = {
    "pred_ids": pred_ids,
    "input_ids": input_ids,
    "tokenizer": tokenizer,
    "model_name_or_path": model_name_or_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 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
delta_bar Ran
ran on a synthesized input
hkust-knowcomp/cola/src/metrics.py
code served (permissive licence) · get_code("1540e18377812a02")
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 float64, ndim 0, 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 numpy as np

# Create sample data for p_xd
# Shape: (Z, num_attributes, 2) where last dim is [count_0, count_1]
# Z = number of samples, num_attributes = 3
p_xd = np.array([
    [[10, 5], [8, 3], [6, 2]],
    [[12, 4], [9, 2], [7, 1]],
    [[11, 6], [7, 4], [5, 3]],
    [[9, 7], [10, 1], [8, 2]]
], dtype=float)

# Create sample data for p_dy
# Shape: (num_classes, 2) where last dim is [count_0, count_1]
p_dy = np.array([
    [15, 8],
    [12, 5],
    [10, 3],
    [8, 6]
], dtype=float)

ARGS = {
    'p_xd': p_xd,
    'p_dy': p_dy,
    'eps': 0.1,
    'ord': 1,
    'normalization': True,
    'direct_match': False,
    'use_cooccur': False,
    'res_norm': False,
    'pxa_all_norm': False
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
delta_pxd Ran
ran on a synthesized input
hkust-knowcomp/cola/src/metrics.py
code served (permissive licence) · get_code("25b707022ce8b4fd")
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 float64, ndim 0, 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 numpy as np

# Analyze the function: delta_pxd(p_xd, p_dy, p_xy, **kwargs)
# The function accesses:
# - p_dy[0][0]: indexing into first element twice, so 2D array-like
# - p_xy[:, 0]: slicing first column, so 2D array-like
# - np.nanmean(): operates on arrays

# Create sample data
p_xd = np.array([[1.0, 2.0, 3.0, 4.0],
                  [5.0, 6.0, 7.0, 8.0]])  # 2x4 array

p_dy = np.array([[0.5, 1.5],
                 [2.5, 3.5]])  # 2x2 array, accessed as p_dy[0][0]

p_xy = np.array([[10.0, 20.0, 30.0],
                 [40.0, 50.0, 60.0]])  # 2x3 array, sliced as p_xy[:, 0]

ARGS = {
    'p_xd': p_xd,
    'p_dy': p_dy,
    'p_xy': p_xy
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
duplicate_list Ran
ran on a synthesized input
hkust-knowcomp/cola/sample_intervention.py
code served (permissive licence) · get_code("b7cf9f9b8c2c0cf5")
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 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

torch.manual_seed(0)

ARGS = {
    "my_list": [1, 2, 3],
    "length": 7
}
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 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
format_args Ran
ran on a synthesized input
hkust-knowcomp/cola/covariate_util.py
code served (permissive licence) · get_code("07d0e75610183e3c")
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.

from dataclasses import dataclass, field
import torch

# Examine the function: it takes an 'args' parameter
# The function calls:
# - asdict(args)  -> args must be a dataclass instance
# - args.__class__.__name__  -> get class name
# - iterates over k, v pairs checking if k.endswith('_token')
# - builds formatted string output

# Create a simple dataclass with various fields including some *_token fields
@dataclass
class SampleArgs:
    model_name: str = "bert-base"
    learning_rate: float = 1e-4
    batch_size: int = 32
    max_length: int = 128
    api_token: str = "secret123"
    access_token: str = "access456"
    num_epochs: int = 10
    dropout: float = 0.1

# Create an instance
args_instance = SampleArgs()

ARGS = {
    "args": args_instance
}
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_config_dict Ran
ran on a synthesized input
hkust-knowcomp/cola/evaluate.py
code served (permissive licence) · get_code("398d9dc4825e6ceb")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function simply constructs and returns a dict with the given parameters
# No special attributes or methods are called on any of them
# So we can use simple values for all parameters

ARGS = {
    'dm': 64,           # D - likely a dimension parameter (int)
    'nm': 128,          # S - likely a size/model dimension parameter (int)
    'pn': 256,          # Q - likely a projection/query dimension (int)
    'co': 32,           # C - likely a count/channels parameter (int)
    'rn': 8,            # E - likely an embedding/rank parameter (int)
    'eps': 1e-5         # eps - epsilon for numerical stability (float)
}
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
get_f1 Ran
ran on a synthesized input
hkust-knowcomp/cola/evaluate.py
code served (permissive licence) · get_code("f71f4792bb23e87b")
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 numpy as np

# Analyze the function:
# - causal_label_list: used in element-wise multiplication with ground_truth_list, then np.sum()
# - ground_truth_list: used in element-wise multiplication with causal_label_list, then np.sum()
# Both need to be numpy arrays that can be multiplied element-wise and summed

# Create small numpy arrays with binary labels (0s and 1s for F1 score context)
causal_label_list = np.array([1, 0, 1, 1, 0])
ground_truth_list = np.array([1, 1, 1, 0, 0])

ARGS = {
    "causal_label_list": causal_label_list,
    "ground_truth_list": ground_truth_list,
}
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 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
get_input_pair Ran
ran on a synthesized input
hkust-knowcomp/cola/temporal_score_util.py
code served (permissive licence) · get_code("a138ca7ac0f58678")
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

torch.manual_seed(0)

# The function takes two lists of strings
# text_event_list: list of text strings (punctuation will be removed from the end)
# outcome_event_list: list of outcome strings (first letter will be lowercased)

text_event_list = [
    "This is a sample text.",
    "Another event happened!"
]

outcome_event_list = [
    "Outcome One",
    "Outcome Two"
]

ARGS = {
    "text_event_list": text_event_list,
    "outcome_event_list": outcome_event_list
}
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 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
get_macro_f1 Ran
ran on a synthesized input
hkust-knowcomp/cola/evaluate.py
code served (permissive licence) · get_code("e725c859a48e41ec")
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 numpy as np

# Analyze the function:
# get_macro_f1 takes two parameters:
# 1. causal_label_list - used with arithmetic operations (*, 1 -)
# 2. ground_truth_list - used with arithmetic operations (*, 1 -)
# Both are treated as numpy arrays/lists that support element-wise operations

# Create small numpy arrays with binary values (0 or 1)
causal_label_list = np.array([1, 0, 1, 1, 0, 1])
ground_truth_list = np.array([1, 0, 1, 0, 0, 1])

ARGS = {
    "causal_label_list": causal_label_list,
    "ground_truth_list": ground_truth_list,
}
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 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
get_preprocess_function Ran
ran on a synthesized input
hkust-knowcomp/cola/temporal_score_util.py
code served (permissive licence) · get_code("0d2028a1f903d498")
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 function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a mock tokenizer that has a __call__ method
class MockTokenizer:
    def __call__(self, texts, max_length=None, padding=False, truncation=False):
        # Return a dict with 'input_ids' and 'attention_mask' keys
        # Simulate tokenization by creating dummy token ids
        batch_size = len(texts)
        token_ids = [[1, 2, 3, 4] for _ in range(batch_size)]
        attention_masks = [[1, 1, 1, 1] for _ in range(batch_size)]
        return {
            'input_ids': token_ids,
            'attention_mask': attention_masks
        }

# Create mock examples dict with text and outcome columns
examples = {
    'text': ['hello world', 'foo bar'],
    'outcome': ['positive', 'negative']
}

tokenizer = MockTokenizer()
mask_token = '[MASK]'
max_length = 512
text_column = 'text'
outcome_column = 'outcome'

ARGS = {
    'text_column': text_column,
    'outcome_column': outcome_column,
    'tokenizer': tokenizer,
    'mask_token': mask_token,
    'max_length': max_length
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_preprocess_function Ran
ran on a synthesized input
hkust-knowcomp/cola/covariate_util.py
code served (permissive licence) · get_code("b72601e68fddf6cc")
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 function. 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
from transformers import PreTrainedTokenizer

torch.manual_seed(0)

class MockTokenizer:
    """Minimal tokenizer stand-in that mimics HuggingFace tokenizer behavior."""
    def __call__(self, text, padding=False, truncation=True, return_tensors=None):
        # Simulate tokenization: split text into tokens and create token IDs
        tokens = text.split()
        token_ids = [101] + [i % 1000 for i in range(len(tokens))] + [102]  # CLS + tokens + SEP
        
        result = {
            'input_ids': token_ids,
            'attention_mask': [1] * len(token_ids),
        }
        
        if return_tensors == 'pt':
            result['input_ids'] = torch.tensor([result['input_ids']], dtype=torch.long)
            result['attention_mask'] = torch.tensor([result['attention_mask']], dtype=torch.long)
        
        return result

class MockDataArgs:
    """Minimal data arguments class."""
    def __init__(self):
        self.text_column = 'text'
        self.pad_to_max_length = False

# Create example batch with text column
examples = {
    'text': [
        'The cat sat',
        'on the mat',
        'and looked around',
        'very carefully today',
        'The dog ran',
        'through the park',
        'very quickly now',
        'chasing a squirrel',
    ]
}

ARGS = {
    'data_args': MockDataArgs(),
    'tokenizer': MockTokenizer(),
    'model_name_or_path': 't5-base',
    'joint': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_preprocess_function Ran
ran on a synthesized input
hkust-knowcomp/cola/mlm_util.py
code served (permissive licence) · get_code("3304c2a3b4189c4f")
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 function. 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 logging
from dataclasses import dataclass

torch.manual_seed(0)

# Create a minimal logger
logger = logging.getLogger("test_logger")
logger.setLevel(logging.WARNING)

# Create a minimal tokenizer stand-in class
class MinimalTokenizer:
    def __init__(self):
        self.model_max_length = 512
        self.mask_token = "[MASK]"
    
    def __call__(self, texts, padding=False, truncation=True, max_length=None):
        # Return a dict with 'input_ids' and optionally other token type fields
        batch_size = len(texts)
        return {
            'input_ids': [[101, 2054, 2003, 102] for _ in range(batch_size)],
            'attention_mask': [[1, 1, 1, 1] for _ in range(batch_size)],
            'token_type_ids': [[0, 0, 0, 0] for _ in range(batch_size)]
        }

# Create a minimal data_args stand-in class
@dataclass
class DataArgs:
    max_seq_length: int = 256
    pad_to_max_length: bool = False

# Create token2id_mapping
token2id_mapping = {
    "relation_0": 0,
    "relation_1": 1,
    "relation_2": 2
}

ARGS = {
    "data_args": DataArgs(max_seq_length=256, pad_to_max_length=False),
    "tokenizer": MinimalTokenizer(),
    "logger": logger,
    "token2id_mapping": token2id_mapping
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_probs Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/cola/src/metric_utils.py
pointer only (licence: MIT); the stored copy get_code serves, from insundaycathy/event-causality-extraction, is pointer only (licence: NONE) · get_code("50bc28c5417a08cb")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from insundaycathy/event-causality-extraction, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

proc_data Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/cola/src/metric_utils.py
pointer only (licence: MIT); the stored copy get_code serves, from insundaycathy/event-causality-extraction, is pointer only (licence: NONE) · get_code("3c1f2a1e109ff596")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from insundaycathy/event-causality-extraction, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

return_match_idx Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hkust-knowcomp/cola/src/metrics.py
code served (permissive licence) · get_code("d5011fe46a81a130")
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 1, dtype family int, shape (3,) 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 test matrix
# Shape: (Z, features) where Z is number of samples
# The function computes norm of columns 1: relative to column 0
# Making it (5, 4) - 5 samples, 4 features
mat = np.array([
    [1.0, 1.05, 1.02, 0.98],
    [2.0, 2.08, 2.01, 1.99],
    [3.0, 3.06, 3.03, 2.97],
    [4.0, 4.09, 4.02, 3.98],
    [5.0, 5.07, 5.01, 4.99]
], dtype=np.float64)

ARGS = {
    "mat": mat,
    "ord": None,
    "eps": 0.1,
    "return_arr": False
}
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 1, dtype family int, shape (3,) — shape (3,) 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
unpacked_data Ran
ran on a synthesized input
hkust-knowcomp/cola/temporal_score_util.py
code served (permissive licence) · get_code("e8757970675c9907")
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

# Construct valid arguments for unpacked_data function

# The function expects:
# - example_list: list of dicts with text_column and outcome_column keys
# - text_column: string key to access text events
# - outcome_column: string key to access outcome events

# Looking at the function body:
# - It iterates over example_list
# - For each example, it accesses cur_example[text_column] and cur_example[outcome_column]
# - Both should be lists (since get_input_pair iterates over them with for loops)
# - text_event_list and outcome_event_list are passed to get_input_pair
# - get_input_pair expects iterables of strings (it applies _remove_punct and _sent_lowercase)

text_column = 'texts'
outcome_column = 'outcomes'

example_list = [
    {
        'texts': ['First event.', 'Second event.'],
        'outcomes': ['Outcome A', 'Outcome B']
    },
    {
        'texts': ['Third event.', 'Fourth event.'],
        'outcomes': ['Outcome C']
    }
]

ARGS = {
    'example_list': example_list,
    'text_column': text_column,
    'outcome_column': outcome_column
}
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 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
get_token2id_mapping Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/cola/mlm_util.py
code served (permissive licence) · get_code("bfb9f39475bae9a7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 18 distinct functions with an identity, 3 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 2 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.

FunctionAlso present inEarliest-datedOthers named
get_probs 1 other paper
1 later
2311.09648 (2023-11)
Event Causality Is Key to Computational Story Understanding
proc_data 1 other paper
1 later
2311.09648 (2023-11)
Event Causality Is Key to Computational Story Understanding
return_match_idx 1 other paper
1 dated earlier
2202.00436 (2022-02)
ROCK : Causal Inference Principles for Reasoning about Commonsense Causality

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

Detecting commonsense causal relations (causation) between events has long been an essential yet challenging task. Given that events are complicated, an event may have different causes under various contexts. Thus, exploiting context plays an essential role in detecting causal relations. Meanwhile, previous works about commonsense causation only consider two events and ignore their context, simplifying the task formulation. This paper proposes a new task to detect commonsense causation between two events in an event sequence (i.e., context), called contextualized commonsense causal reasoning. We also design a zero-shot framework: COLA (Contextualized Commonsense Causality Reasoner) to solve the task from the causal inference perspective. This framework obtains rich incidental supervision from temporality and balances covariates from multiple timestamps to remove confounding effects. Our extensive experiments show that COLA 1 can detect commonsense causality more accurately than baselines. Cause 1. Emma felt hungry. Cause 2. Emma was doing her job. Emma made a steak in the kitchen. 🥩 Context-Free 🤔 ✅ Which cause is more plausible? Cause 1. Emma felt hungry. Cause 2. Emma was doing her job. Emma made a steak in the kitchen. 🥩 Context 1: Emma exercised for a while. ✅ 🏃 Cause 1. Emma felt hungry. Cause 2. Emma was doing her job.

For agents

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

get_harvested_code_for_paper("2305.05191")
get_code_for_paper("2305.05191")
have("2305.05191")

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

curl https://syntology.ai/api/ran/2305.05191.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.05191.svg)](https://syntology.ai/paper/2305.05191)

Connect an agent — have() is free.