SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2212.10097 · 2022

Optimization Techniques for Unsupervised Complex Table Reasoning via Self-Training Framework

arXiv · PDF · Open in the Atlas

Code that ran

17 of 20 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 17 have no contract verdict; 0 failures of our instrument; 3 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/2212.10097.json · badge: /api/ran/2212.10097.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
leezythu/uctr canonical (classifier label)
classifier basis not recorded on the link
13 ran · 3 no recorded run
leezythu/uctr-st canonical (classifier label)
classifier basis not recorded on the link
4 ran · 0 no recorded run
FunctionStatusWhere it lives
ds_arg Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/process_program.py
code served (permissive licence) · get_code("b75be4391719a2b0")
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)

# Analyzing the ds_arg function:
# - logic: a dict with 'func' (str) and 'args' (list) keys
# - all: a list that gets appended to
# - pos_d: a dict mapping strings to values

# The function processes a nested logic structure, where args can be dicts or strings

# Create minimal test data
logic = {
    'func': 'eq',
    'args': ['arg1', 'arg2']
}

all_list = []

pos_d = {
    'arg1': 'value1',
    'arg2': 'value2'
}

ARGS = {
    'logic': logic,
    'all': all_list,
    'pos_d': pos_d
}
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
find_index Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/fill_in_template.py
code served (permissive licence) · get_code("534cdb9a35bb5f05")
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 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

torch.manual_seed(0)

# Construct a simple 2D table (list of lists) with some values
table = [
    [10, 20, 30],
    [40, 50, 60],
    [70, 80, 90]
]

# The value to search for
val = 50

ARGS = {
    "val": val,
    "table": table
}
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 tuple, ndim 1, dtype family int — 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
find_num Ran
ran on a synthesized input
leezythu/uctr/TAT-QA/finqa/collect_data.py
code served (permissive licence) · get_code("18e12d28c390eca0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Analyze what find_num expects:
# - program: a string containing operations, parentheses, commas, numbers, and possibly comments/const
# The function processes a program string by:
# 1. Removing all operation names
# 2. Removing parentheses
# 3. Splitting by comma and collecting items that don't start with '#' or 'const'

# Create test programs with various formats
program1 = "add(5, 3), subtract(10, 2), multiply(4, 6)"
program2 = "table_sum(#1, #2), divide(100, 5), 42, const_value"
program3 = "greater(x, y), exp(z), table_max(a, b, c)"
program4 = "add(1, 2), #comment, const_x, 99, subtract(10, 5)"

# Test with a simple program
test_program = "add(#1, table_max(2, 3)), subtract(const_x, 15), multiply(x, y), 42"

ARGS = {
    "program": test_program
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
find_pos Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/process_program.py
code served (permissive licence) · get_code("4f0aabb25ef5e927")
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

torch.manual_seed(0)

# Analyze the function:
# - find_pos takes 'rows' parameter
# - rows is indexed as rows[0], rows[1], etc. (list-like)
# - rows[j] is indexed by i (list-like)
# - rows[j][i] are individual elements that can be dict keys (hashable)
# - The function builds a dict mapping elements to strings

# Construct valid rows argument
rows = [
    ['a', 'b', 'c'],           # rows[0] - header row
    ['x', 'y', 'z'],           # rows[1] - first data row
    ['a', 'new_val', 'c'],     # rows[2] - second data row with one new element
]

ARGS = {
    'rows': rows
}
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
find_pos Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/syn.py
code served (permissive licence) · get_code("a8deec34bf20a0a4")
How it ran

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

import torch
import numpy as np

# Analyzing the function find_pos:
# - It takes a parameter 'table'
# - It iterates over keys in 'table' (so table must be iterable like a dict)
# - It accesses table[key] to get values
# - The function builds a new dict 'd' with keys like 'c0', 'val_0', etc.

# The simplest valid argument is a dictionary with some key-value pairs
table = {
    'key_a': 'value_1',
    'key_b': 42,
    'key_c': [1, 2, 3],
    'key_d': {'nested': True}
}

ARGS = {
    'table': table
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
fuzzy_compare_filter Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/execution/APIs.py
code served (permissive licence) · get_code("9b7f0f8a8a9d658e")
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 sample DataFrame with mixed content (dates and numbers)
df = pd.DataFrame({
    'dates': ['January 15, 2020', 'February 3, 2021', 'March 22, 2019', 'April 10, 2020'],
    'values': ['Price: 100.5', 'Cost = 250', 'Amount: 75.25', 'Total 500']
})

# Define ARGS dictionary with all required parameters
ARGS = {
    't': df,
    'col': 'dates',
    'val': 'January 2020',
    'type': 'eq'
}
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
fuzzy_match_filter Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/execution/APIs.py
code served (permissive licence) · get_code("bd791c82a8efe829")
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

ARGS = {
    't': pd.DataFrame({
        'col_A': ['hello world', 'foo bar', 'test data'],
        'col_B': ['some value', 'another entry', 'more text']
    }),
    'col': 'col_A',
    'val': 'hello',
    'negate': False
}
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
get_description Ran
ran on a synthesized input
leezythu/uctr/TAT-QA/finqa/collect_data.py
code served (permissive licence) · get_code("e625b89f3bb8669a")
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

# Construct a valid table structure
# The function searches for 'num' in table[i][j] for i,j >= 1
# table[0][0] is unused, table[i][0] are row names, table[0][j] are column names
# table[i][j] for i,j >= 1 contains the data to search

table = [
    [None, "Color", "Size", "Material"],  # row 0: headers (table[0][0] unused, then col names)
    ["Fruits", ["apple", "banana"], ["orange"], ["grape", "kiwi"]],  # row 1
    ["Vegetables", ["carrot"], ["lettuce", "spinach"], ["broccoli"]],  # row 2
    ["Metals", ["iron", "gold"], ["silver"], ["copper", "bronze"]],  # row 3
]

# The function looks for 'num' (a string or item) in the nested lists
# Let's search for something that exists
num = "apple"

ARGS = {
    "num": num,
    "table": table
}
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
get_evidence Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/fill_in_template.py
code served (permissive licence) · get_code("b3ccf463e4486dce")
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.

# Setup code for get_evidence function

# Construct feed_dict with 'c' prefixed keys for 'count' type
feed_dict = {
    'c0': 'value1',
    'c1': 'value2',
    'c2': 'value3'
}

# Construct must_have list with values that exist in the table
must_have = ['apple', 'banana', 'cherry']

# Construct type parameter
type_param = 'row2text'

# Construct table_index
table_index = 0

# Construct doc string
doc = 'document_1'

# Construct table as 2D list (4 rows x 3 columns with distinct sizes)
table = [
    ['header1', 'header2', 'header3'],
    ['apple', 'red', '5'],
    ['banana', 'yellow', '3'],
    ['cherry', 'red', '7']
]

# Construct is_header as 2D boolean list (4 rows x 3 columns)
# Row 0 is header, other rows are data
is_header = [
    [True, True, True],
    [False, False, False],
    [False, False, False],
    [False, False, False]
]

# Define ARGS dict mapping all parameters
ARGS = {
    'feed_dict': feed_dict,
    'must_have': must_have,
    'type': type_param,
    'table_index': table_index,
    'doc': doc,
    'table': table,
    'is_header': is_header
}
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
get_mapping Ran
ran on a synthesized input
leezythu/uctr-st/TAT-QA/generate_pseudo.py
code served (permissive licence) · get_code("defe43aaba4c6447")
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 math

# Construct table argument
# The function iterates through table[i][j] and checks if fact is in the string
# or if the numeric value matches
table = [
    ['10', '20.5', 'text with 30 in it'],
    ['$1,000', '(50%)', 'description'],
    ['data', '100', 'another entry']
]

# Construct para argument
# The function accesses p['text'] and p['order'] for each paragraph
para = [
    {'text': 'This paragraph contains the fact 10 and some more text', 'order': 0},
    {'text': 'Another paragraph with value 20.5 mentioned here', 'order': 1},
    {'text': 'Final paragraph with reference to 100 in the content', 'order': 2}
]

# Construct facts argument
# The function checks if facts match table entries or paragraph text
# It handles floats, integers, and strings
facts = [10, 20.5, '100', 'text with 30 in it']

ARGS = {
    'table': table,
    'para': para,
    'facts': facts
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_para_mapping Ran
ran on a synthesized input
leezythu/uctr-st/TAT-QA/generate_pseudo.py
code served (permissive licence) · get_code("a4ec9baf2fc7a4f5")
How it ran

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

import torch

def get_para_mapping(para, fact):
    for p in para:
        text = p['text'].lower()
        rel_index = p['order']
        try:
            start_idx = text.index(fact.lower())
        except:
            continue
        end_idx = start_idx + len(fact)
        return (rel_index, [start_idx, end_idx])
    return (None, None)


# Construct valid arguments
para = [
    {
        'text': 'The quick brown fox jumps over the lazy dog',
        'order': 0
    },
    {
        'text': 'Another paragraph with some text',
        'order': 1
    },
    {
        'text': 'Third paragraph containing a fact here',
        'order': 2
    }
]

fact = 'brown fox'

ARGS = {
    'para': para,
    'fact': fact
}
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 tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_table_mapping Ran
ran on a synthesized input
leezythu/uctr-st/TAT-QA/generate_pseudo.py
code served (permissive licence) · get_code("ce5896bb3ae606e2")
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

# The function iterates through a 2D table (list of lists)
# and searches for a string representation of 'fact' or numeric equivalence
# It returns [i, j] indices or None

# Create a sample table with mixed content
table = [
    ["Name", "Value", "Status"],
    ["Item1", "42", "Active"],
    ["Item2", "($100.50)", "Inactive"],
    ["Item3", "95%", "Pending"]
]

# Create a fact to search for - can be a string or number
fact = 42

ARGS = {
    "table": table,
    "fact": fact
}
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
locate_num Ran
ran on a synthesized input
leezythu/uctr/TAT-QA/finqa/collect_data.py
code served (permissive licence) · get_code("a79f0ace7e53331c")
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 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

# The function locate_num searches through a 2D table structure
# looking for a number within nested containers at each position.
# It returns the (i, j) coordinates where num is found, or (None, None).

# Create a simple 2D table where each cell contains a collection
# The function checks "if num in table[i][j]", so each cell needs to be
# a container that supports the 'in' operator (list, set, tuple, etc.)

table = [
    [set(), {1, 2, 3}, {4, 5, 6}],
    [{7, 8}, {9, 10, 11}, {12, 13}],
    [{14}, {15, 16, 17}, {18, 19, 20}]
]

num = 11

ARGS = {
    "num": num,
    "table": table
}
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 tuple, ndim 1, dtype family int — 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
normalize Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/syn.py
code served (permissive licence) · get_code("c4ae14071abdbcdf")
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 random
import json
import re

class MockRow:
    def __init__(self, cell_content):
        self.cell_content = cell_content

# Create a row object with cell_content attribute
# The function processes items in cell_content looking for [[...]] patterns
# and replaces them with the part after the pipe
row_obj = MockRow([
    "some text [[link|display text]] more text",
    "normal item without pattern",
    "another [[internal|external]] reference",
    "no brackets here"
])

ARGS = {
    "row": row_obj
}
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
normalize Ran
ran on a synthesized input
leezythu/uctr-st/SEM-TAB-FACTS/syn.py
code served (permissive licence) · get_code("cfb7267da4dc4f92")
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 torch

# Construct a simple 2D table with string entries containing newlines and extra spaces
table = [
    ["hello\nworld", "foo  bar", "test\n\nmultiple"],
    ["line1\nline2", "a    b    c", "x\ny\nz"]
]

ARGS = {
    "table": table
}
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 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
obj_compare Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/execution/APIs.py
code served (permissive licence) · get_code("f36c3eba21146e55")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import math

torch_seed = __import__('torch').manual_seed(0)

ARGS = {
    'num1': 3.14159,
    'num2': 3.14160,
    'round': False,
    'type': 'eq'
}
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 bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
paraphrase Ran
ran on a synthesized input
leezythu/uctr/FEVEROUS/process_program.py
code served (permissive licence) · get_code("3941e0c5c9a3493d")
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)

ARGS = {
    'func': 'eq'
}
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
execute_all Not yet run
unverified (harvested, no recorded run)
leezythu/uctr/FEVEROUS/execution/execute.py
code served (permissive licence) · get_code("bb7c3358b37dc877")
How it ran

no invocation recorded.

gen_feed Not yet run
unverified (harvested, no recorded run)
leezythu/uctr/FEVEROUS/syn.py
code served (permissive licence) · get_code("59e65dbe081b83ae")
How it ran

no invocation recorded.

gen_feed Not yet run
unverified (harvested, no recorded run)
leezythu/uctr/SEM-TAB-FACTS/syn.py
code served (permissive licence) · get_code("0f626cad0a001441")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 19 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, 0 later) and 18 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
execute_all 1 other paper
1 dated earlier
2004.14579 (2020-04)
Logic2Text: High-Fidelity Natural Language Generation from Logical Forms

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

Structured tabular data is a fundamental data type in numerous fields, and the capacity to reason over tables is crucial for answering questions and validating hypotheses. However, constructing labeled data for complex reasoning tasks is labor intensive, and the quantity of annotated data remains insufficient to support the intricate demands of real-world applications. To address the insufficient annotation challenge, we present a self-training framework for unsupervised complex tabular reasoning (UCTR-ST) by generating diverse synthetic data with complex logic. Specifically, UCTR-ST incorporates several essential techniques: we aggregate diverse programs and execute them on tables based on a "Program-Management" component, and we bridge the gap between programs and text with a powerful "Program-Transformation" module that generates natural language sentences with complex logic. Furthermore, we optimize the procedure using a "Table-Text Manipulator" to handle joint table-text reasoning scenarios. The entire framework utilizes self-training techniques to leverage the unlabeled training data, which results in significant performance improvements when tested on real-world data. Experimental results demonstrate that UCTRST achieves above 90% of the supervised model performance on different tasks and domains, reducing the dependence on manual annotation. Additionally, our approach can serve as a data augmentation technique, significantly boosting the performance of supervised models in low-resourced domains.

For agents

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

get_harvested_code_for_paper("2212.10097")
get_code_for_paper("2212.10097")
have("2212.10097")

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

curl https://syntology.ai/api/ran/2212.10097.json

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

[![Syntology run record](https://syntology.ai/api/ran/2212.10097.svg)](https://syntology.ai/paper/2212.10097)

Connect an agent — have() is free.