SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2307.02477 · 2023

Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks

arXiv · PDF · Open in the Atlas

Code that ran

13 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 12 failures of our instrument; 2 no recorded run), measured 2026-11-21; 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/2307.02477.json · badge: /api/ran/2307.02477.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
zhaofengwu/counterfactual-evaluation canonical (classifier label)
classifier basis: census
13 ran · 2 no recorded run
FunctionStatusWhere it lives
encode_images Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/drawing/eval_utils.py
code served (permissive licence) · get_code("dd09dc3fabcd7c46")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 512) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal processor stand-in
class MockProcessor:
    def __call__(self, images, return_tensors=None):
        # Simulate processor output: returns a dict with pixel_values tensor
        batch_size = len(images) if isinstance(images, list) else 1
        return {
            'pixel_values': torch.randn(batch_size, 3, 224, 224)
        }
    
    def to(self, device):
        # Return a dict-like object that can be unpacked with **
        class ProcessorOutput(dict):
            def to(self, device):
                # Convert tensors to device and return self
                for key in self:
                    if isinstance(self[key], torch.Tensor):
                        self[key] = self[key].to(device)
                return self
        
        output = ProcessorOutput(pixel_values=torch.randn(2, 3, 224, 224))
        return output.to(device)

# Create a minimal model stand-in
class MockVisionModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.device_attr = 'cpu'
        self.linear = nn.Linear(224 * 224 * 3, 512)
    
    @property
    def device(self):
        return self.device_attr
    
    def get_image_features(self, pixel_values):
        # Flatten and project to embedding dimension (512)
        batch_size = pixel_values.shape[0]
        flattened = pixel_values.reshape(batch_size, -1)
        embeddings = self.linear(flattened)
        return embeddings

# Create minimal processor that works with the function
class FunctionalProcessor:
    def __call__(self, images, return_tensors=None):
        # Return an object that can be converted to dict and unpacked
        batch_size = len(images) if isinstance(images, list) else 2
        return ProcessorOutputDict(
            pixel_values=torch.randn(batch_size, 3, 224, 224)
        )

class ProcessorOutputDict(dict):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
    
    def to(self, device):
        for key in self:
            if isinstance(self[key], torch.Tensor):
                self[key] = self[key].to(device)
        return self

# Create test images (list of dummy tensors)
images = [
    torch.randn(3, 224, 224),
    torch.randn(3, 224, 224)
]

# Create processor
processor = FunctionalProcessor()

# Create model
model = MockVisionModel()

ARGS = {
    'images': images,
    'processor': processor,
    'model': model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 512) — shape (2, 512) 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
escape Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/query_ccc.py
code served (permissive licence) · get_code("8e2c1bf8f5114da3")
How it ran

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

import torch

def escape(str):
    return str.replace('\n', '\\n')

# The function takes a single parameter: str (a string)
# It replaces newline characters with the escaped sequence '\n'
# We need to provide a string with newlines to make this meaningful

ARGS = {
    "str": "hello\nworld\ntest"
}
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
estimate_pass_at_k Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/programming/generation/human_eval/evaluation.py
pointer only (licence: MIT); the stored copy get_code serves, from rucaibox/bamboo, is pointer only (licence: NONE) · get_code("e080db2eaa81fdb3")
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 rucaibox/bamboo, 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.

get_captions Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/drawing/eval_utils.py
code served (permissive licence) · get_code("4fd941db8b2c2e88")
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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'obj': 'elephant'
}
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_label Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/sample.py
code served (permissive licence) · get_code("6cd76a2436afdd9f")
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

# Analyzing the function:
# - expr: a string containing two numbers separated by '+'
# - base: an integer representing the base for conversion
# The function splits the expression, converts from the given base to base 10,
# adds them, and converts back to the original base.

ARGS = {
    "expr": "12+34",  # A valid expression with two numbers in some base
    "base": 16        # Base 16 (hexadecimal) - a common base for testing
}
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
load_data Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/chess/query.py
code served (permissive licence) · get_code("819f8be42e847f42")
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 tempfile
import os

# Create a temporary file with test data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("hello world *\n")
temp_file.write("foo bar *\n")
temp_file.write("test data *\n")
temp_file.close()

ARGS = {
    'data_file': temp_file.name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_output Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/eval_ccc.py
code served (permissive licence) · get_code("5352f13bd6bc44b3")
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 re

def parse_output(output):
    original_output = output
    if output.endswith(', with each digit incremented as needed.'):
        output = output[:-len(', with each digit incremented as needed.')]
    if output.endswith('End of response.'):
        output = output[:-len('End of response.')]
    output = output.rstrip()
    if (match := re.search('^[0-9A-Z]+$', output)) is not None:
        return output
    if (match := re.search('\\n([0-9A-Z]+ \\+ [0-9A-Z]+ = )?([0-9A-Z]+)( \\(?(in )?base-[0-9]+\\)?)?\\.?$', output)) is not None:
        return match.groups()[-3]
    if (match := re.search('^([0-9A-Z]+ \\+ [0-9A-Z]+ = )?([0-9A-Z]+)( \\(?(in )?base-[0-9]+\\)?)?\\.?$', output)) is not None:
        return match.groups()[-3]
    if (match := re.search('(would be|is|as)( therefore)?:?\\s*([0-9]+\\. )?([0-9A-Z]+ \\+ [0-9A-Z]+ = )?([0-9A-Z]+)\\.?$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('in base-[0-9]+:?\\s+([0-9]+\\. )?([0-9A-Z]+)\\.?$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('(written as|is|would be|the number) ([0-9A-Z]+)( \\(?(in )?base-[0-9]+\\)?)?(, which is (equal|equivalent) to [0-9A-Z +*=^]+( in (decimal|base-[0-9]+))?)?\\.$', output)) is not None:
        return match.groups()[-7]
    if (match := re.search('\\n[0-9]+\\. ([0-9A-Z]+)\\.?$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('(([0-9A-Z]+\\+1=)?[0-9A-Z]+(, |\\n))+([0-9A-Z]+\\+1=)?([0-9A-Z]+)\\n*\\.*$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('(answer|result): ([0-9A-Z]+)\\.?$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('\\.\\.\\. ([0-9A-Z]+)\\.?$', output)) is not None:
        return match.groups()[-1]
    if (match := re.search('(T|t)he next number after [0-9A-Z]+( in base-[0-9]+)? is:?\\s*([0-9A-Z]+)(,| \\()', output)) is not None:
        return match.groups()[-2]
    print('Failed to parse output:', original_output)
    assert False

# Test cases for the parse_output function
test_output_1 = "ABC123"
test_output_2 = "A + B = C (in base-16)."
test_output_3 = "The result is: XYZ."
test_output_4 = "answer: 42."
test_output_5 = "... ABC123."

ARGS = {
    "output": "ABC123"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
query_openai Ran
ran on a synthesized input
zhaofengwu/counterfactual-evaluation/query_utils.py
code served (permissive licence) · get_code("36f593974514e07c")
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 coroutine. 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 asyncio
import json
from math import ceil, log2
from random import random
import sys
from unittest.mock import AsyncMock, MagicMock

# Mock openai module since we can't import the real one
class MockOpenAIError(Exception):
    pass

class MockChatCompletion:
    @staticmethod
    async def acreate(**kwargs):
        return {
            'choices': [
                {
                    'finish_reason': 'stop',
                    'message': {'content': 'This is a mock response'}
                }
            ]
        }

mock_openai = MagicMock()
mock_openai.ChatCompletion = MockChatCompletion()
mock_openai.error.APIError = MockOpenAIError
mock_openai.error.TryAgain = MockOpenAIError
mock_openai.error.Timeout = MockOpenAIError
mock_openai.error.APIConnectionError = MockOpenAIError
mock_openai.error.ServiceUnavailableError = MockOpenAIError
mock_openai.error.RateLimitError = MockOpenAIError

sys.modules['openai'] = mock_openai

# Create temporary history file path (won't actually write in this setup)
ARGS = {
    'prompt': 'What is the capital of France?',
    'model_name': 'gpt-3.5-turbo',
    'system_msg': 'You are a helpful assistant.',
    'history': [
        {'role': 'system', 'content': 'Previous system message'},
        {'role': 'user', 'content': 'Previous user message'},
        {'role': 'assistant', 'content': 'Previous assistant message'}
    ],
    'max_tokens': 100,
    'temperature': 0.7,
    'retry': 2,
    'n': 1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
openai — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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_number Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/sample.py
code served (permissive licence) · get_code("5aa746044ba7f7d5")
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 random

def sample_number(n_digits, base):
    digits = '0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ'[:base]
    number = ''.join((str(random.choice(digits[1:] if i == 0 else digits)) for i in range(n_digits)))
    assert number[0] != '0' and len(number) == n_digits
    return number

ARGS = {
    'n_digits': 5,
    'base': 16
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sample_single Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/sample.py
code served (permissive licence) · get_code("dc7985be0af1dc39")
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 random

ARGS = {
    'n_digits': 3,
    'base': 16,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
templatize Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/query_ccc.py
code served (permissive licence) · get_code("2c5dcccc856725c9")
How it ran

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

import torch

torch.manual_seed(0)

expr = "9"
base = 10

ARGS = {
    "expr": expr,
    "base": base
}
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
templatize Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/chess/query.py
code served (permissive licence) · get_code("9639157e1c524625")
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.

# Setup code for templatize function

ARGS = {
    'mode': 'real_world',
    'pgn_string': 'pawn',
    'cot': True,
    'is_control': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
unescape Ran
ran; our contract draft was wrong, not the code
zhaofengwu/counterfactual-evaluation/arithmetic/eval_ccc.py
code served (permissive licence) · get_code("274b5d31316b3739")
How it ran

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

import torch
import numpy as np

def unescape(str):
    return str.replace('\\n', '\n')

ARGS = {
    'str': 'Hello\\nWorld\\nTest'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
encode_classes Not yet run
unverified (harvested, no recorded run)
zhaofengwu/counterfactual-evaluation/drawing/eval_utils.py
code served (permissive licence) · get_code("dd4aafd6d5e91a3c")
How it ran

no invocation recorded.

load_data Not yet run
unverified (harvested, no recorded run)
zhaofengwu/counterfactual-evaluation/arithmetic/query.py
code served (permissive licence) · get_code("19996a3c6b913d67")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 14 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
estimate_pass_at_k 20 other papers
4 dated earlier, 16 later
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
2207.14255 (2022-07)
Efficient Training of Language Models to Fill in the Middle
2210.14868 (2022-10)
Multi-lingual Evaluation of Code Generation Models
and 17 more not named here

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 2026-11-21.

Repositories linked to this paper

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

Abstract

The impressive performance of recent language models across a wide range of tasks suggests that they possess a degree of abstract reasoning skills. Are these skills general and transferable, or specialized to specific tasks seen during pretraining? To disentangle these effects, we propose an evaluation framework based on "counterfactual" task variants that deviate from the default assumptions underlying standard tasks. Across a suite of 11 tasks, we observe nontrivial performance on the counterfactual variants, but nevertheless find that performance substantially and consistently degrades compared to the default conditions. This suggests that while current LMs may possess abstract task-solving skills to an extent, they often also rely on narrow, non-transferable procedures for task-solving. These results motivate a more careful interpretation of language model performance that teases apart these aspects of behavior.

For agents

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

get_harvested_code_for_paper("2307.02477")
get_code_for_paper("2307.02477")
have("2307.02477")

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

curl https://syntology.ai/api/ran/2307.02477.json

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

[![Syntology run record](https://syntology.ai/api/ran/2307.02477.svg)](https://syntology.ai/paper/2307.02477)

Connect an agent — have() is free.