SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.16702 · 2023

Rethinking Tabular Data Understanding with Large Language Models

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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 0 have no contract verdict; 9 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/2312.16702.json · badge: /api/ran/2312.16702.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
Leolty/tablellm canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
best_match_based_on_content Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/table.py
code served (permissive licence) · get_code("b8b021484ec64f77")
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 pandas as pd

# Create a simple DataFrame with some columns
# The function searches through columns to find one where all head() values contain unmatched_name
df = pd.DataFrame({
    'user_id': ['alice', 'bob', 'charlie'],
    'email': ['alice@example.com', 'bob@example.com', 'charlie@example.com'],
    'name_field': ['alice_name', 'alice_name', 'alice_name'],
    'other': ['x', 'y', 'z']
})

unmatched_name = 'alice_name'

ARGS = {
    'df': df,
    'unmatched_name': unmatched_name
}
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 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
construct_markdown_table Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/data.py
code served (permissive licence) · get_code("488db90e1ef07f73")
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

ARGS = {
    'header': ['Name', 'Age', 'City'],
    'rows': [['Alice', '30', 'NYC'], ['Bob', '25', 'LA']],
    'header_aware': True
}
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
flatten Ran
ran, honoured the contract we drafted
Leolty/tablellm/evaluate.py
code served (permissive licence) · get_code("f4e20da00a5d2f24")
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

def flatten(lst):
    flat_list = []
    for i in lst:
        if isinstance(i, list):
            flat_list.extend(flatten(i))
        else:
            flat_list.append(i)
    return flat_list

# The function takes a single parameter 'lst' which should be a (possibly nested) list
# It recursively flattens nested lists into a single flat list

ARGS = {
    "lst": [1, [2, 3, [4, 5]], 6, [7, [8, 9]]]
}
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 int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_cot_prompt Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Leolty/tablellm/run_helper.py
code served (permissive licence) · get_code("0cf45bc29c09c932")
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 sys
from unittest.mock import MagicMock, patch

torch.manual_seed(0)

# Mock the prompt modules to avoid file I/O
sys.modules['prompt'] = MagicMock()
sys.modules['prompt.wtq'] = MagicMock()
sys.modules['prompt.wtq.cot'] = MagicMock()
sys.modules['prompt.tabfact'] = MagicMock()
sys.modules['prompt.tabfact.cot'] = MagicMock()

# Set up mock cot_prompt values
sys.modules['prompt.wtq.cot'].cot_prompt = "This is a WTQ COT prompt"
sys.modules['prompt.tabfact.cot'].cot_prompt = "This is a TabFact COT prompt"

ARGS = {
    'dataset_name': 'wtq'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
prompt, 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
load_dataset Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/run_helper.py
code served (permissive licence) · get_code("a6475c2b07a6ec20")
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 json
import tempfile
import os

# Create temporary JSON files for testing
temp_dir = tempfile.mkdtemp()

# Create a mock wtq.json file
wtq_data = [
    {"id": 1, "question": "What is the capital?", "table": {"rows": [["Country", "Capital"]]}}
]
wtq_file = os.path.join(temp_dir, "wtq.json")
with open(wtq_file, 'w') as f:
    json.dump(wtq_data, f)

# Create a mock tabfact.json file
tabfact_data = [
    {"id": 1, "statement": "The fact is true", "table": {"rows": []}}
]
tabfact_file = os.path.join(temp_dir, "tabfact.json")
with open(tabfact_file, 'w') as f:
    json.dump(tabfact_data, f)

# Create a mock custom dataset file
custom_data = [
    {"id": 1, "data": "custom"}
]
custom_file = os.path.join(temp_dir, "custom.json")
with open(custom_file, 'w') as f:
    json.dump(custom_data, f)

# Patch the open function to intercept file operations for the standard paths
import builtins
original_open = builtins.open

def patched_open(file, *args, **kwargs):
    if file == 'data/wtq.json':
        return original_open(wtq_file, *args, **kwargs)
    elif file == 'data/tabfact.json':
        return original_open(tabfact_file, *args, **kwargs)
    else:
        return original_open(file, *args, **kwargs)

builtins.open = patched_open

# Define ARGS with all parameters
ARGS = {
    'dataset_name': 'wtq',
    'dataset_file': None
}
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
markdown_to_df Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/execute.py
code served (permissive licence) · get_code("1925013cb0ab7c9b")
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 pandas as pd
import csv
from io import StringIO
import unicodedata

markdown_string = """| Name | Age | City |
|------|-----|------|
| Alice | 30 | NYC |
| Bob | 25 | LA |"""

ARGS = {
    "markdown_string": markdown_string
}
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
print_partial_markdown Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/data.py
code served (permissive licence) · get_code("c905653bda7ccc66")
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 pandas as pd
import numpy as np

# Create a sample DataFrame with various data types
df = pd.DataFrame({
    'Name': ['Alice', 'Bob', 'Charlie', 'David', 'Eve', 'Frank', 'Grace', 'Henry'],
    'Age': [25, 30, 35, 40, 28, 33, 29, 38],
    'Score': [85.5, 92.3, 78.9, 88.1, 91.2, 86.7, 89.4, 87.2],
    'City': ['NYC', 'LA', 'Chicago', 'Houston', 'Phoenix', 'Philadelphia', 'San Antonio', 'San Diego']
})

ARGS = {
    'df': df,
    'keep': 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
print_partial_markdown Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/execute.py
code served (permissive licence) · get_code("050b33c679b5e164")
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 pandas as pd
import numpy as np

# Create a sample dataframe with multiple rows and columns
data = {
    'A': np.arange(10),
    'B': np.arange(10, 20),
    'C': np.arange(20, 30),
}
df = pd.DataFrame(data)

ARGS = {
    'df': df,
    'keep': 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
remove_merged_suffixes Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/execute.py
code served (permissive licence) · get_code("f8e49cb5d1919217")
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 pandas as pd
import re

def remove_merged_suffixes(df):
    pattern = re.compile('^(.*) \\.\\d+$')
    for col in df.columns:
        for idx, value in df[col].items():
            match = pattern.match(str(value))
            if match:
                new_value = match.group(1).strip()
                if new_value in df[col].drop(idx).values or new_value == col:
                    df.at[idx, col] = new_value
    return df

# Create a sample DataFrame with values that match the pattern
df = pd.DataFrame({
    'col1': ['value1 .1', 'value2 .2', 'value1', 'other'],
    'col2': ['item1 .5', 'item2 .3', 'item1', 'different'],
    'col3': ['test .99', 'sample .7', 'test', 'unique']
})

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 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
transpose Ran
ran; our contract draft was wrong, not the code
Leolty/tablellm/utils/table.py
code served (permissive licence) · get_code("91e307626f8b6a11")
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 pandas as pd
import numpy as np

# Create a simple DataFrame with distinct dimensions (4 rows, 3 columns)
df = pd.DataFrame(
    np.arange(12).reshape(4, 3),
    columns=['A', 'B', 'C']
)

ARGS = {
    'df': df,
    'heading_deduplication': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
load_results Not yet run
unverified (harvested, no recorded run)
Leolty/tablellm/evaluate.py
code served (permissive licence) · get_code("c0044887a847001b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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

Large Language Models (LLMs) have shown to be capable of various tasks, yet their capability in interpreting and reasoning over tabular data remains an underexplored area. In this context, this study investigates from three core perspectives: the robustness of LLMs to structural perturbations in tables, the comparative analysis of textual and symbolic reasoning on tables, and the potential of boosting model performance through the aggregation of multiple reasoning pathways. We discover that structural variance of tables presenting the same content reveals a notable performance decline, particularly in symbolic reasoning tasks. This prompts the proposal of a method for table structure normalization. Moreover, textual reasoning slightly edges out symbolic reasoning, and a detailed error analysis reveals that each exhibits different strengths depending on the specific tasks. Notably, the aggregation of textual and symbolic reasoning pathways, bolstered by a mix self-consistency mechanism, resulted in achieving SOTA performance, with an accuracy of 73.6% on WIKITABLEQUESTIONS, representing a substantial advancement over previous existing table processing paradigms of LLMs.

For agents

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

get_harvested_code_for_paper("2312.16702")
get_code_for_paper("2312.16702")
have("2312.16702")

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

curl https://syntology.ai/api/ran/2312.16702.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.16702.svg)](https://syntology.ai/paper/2312.16702)

Connect an agent — have() is free.