SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.08700 · 2024

DyKnow: Dynamically Verifying Time-Sensitive Factual Knowledge in LLMs

arXiv · PDF · Open in the Atlas

Code that ran

12 of 12 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, 2 violated it and 0 have no contract verdict; 10 failures of our instrument; 0 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/2404.08700.json · badge: /api/ran/2404.08700.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
sislab-unitn/dyknow canonical (classifier label)
classifier basis: census
12 ran · 0 no recorded run
FunctionStatusWhere it lives
encode_inputs Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/RAG/generate_rag_answers.py
code served (permissive licence) · get_code("1bddd74f6ee2e001")
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 BatchEncoding. 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
from transformers import BatchEncoding

torch.manual_seed(0)

class MockTokenizer:
    def __init__(self, model_name):
        self.model_name = model_name
    
    def apply_chat_template(self, messages, return_tensors=None, padding=False):
        # Return a simple tensor for mistral
        return torch.tensor([[101, 2054, 2003, 102]])
    
    def __call__(self, text, return_tensors=None, padding=False):
        # Return BatchEncoding with input_ids and attention_mask for llama2 and others
        input_ids = torch.tensor([[101, 2054, 2003, 102], [101, 2054, 2003, 102]])
        attention_mask = torch.ones_like(input_ids)
        return BatchEncoding({
            'input_ids': input_ids,
            'attention_mask': attention_mask
        })

# Test with mistral model
mistral_tokenizer = MockTokenizer('mistral-7b')
llama_tokenizer = MockTokenizer('llama2-7b')
other_tokenizer = MockTokenizer('gpt2')

ARGS = {
    'q': 'What is the capital of France?',
    'context': 'France is a country in Europe. Paris is the capital of France.',
    'tokenizer': mistral_tokenizer,
    'model_name': 'mistral-7b'
}
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 BatchEncoding — a BatchEncoding output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a BatchEncoding output the recorded ndim 1 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
extract_answer Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/analyze_replies.py
code served (permissive licence) · get_code("17600052c8ca574e")
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 re
from typing import Dict, List, Optional, Set
from datetime import datetime

# Construct valid arguments for extract_answer function

answers = [
    "Alice|S:+2020-01-15T10:30:00Z",
    "Bob|S:+2021-05-20T14:45:00Z|E:+2021-12-31T23:59:59Z",
    "Charlie|E:+2022-06-10T08:00:00Z",
]

exceptions = {
    "person": {
        "employee": {
            "status": ["Diana"]
        }
    }
}

category = "person"
element = "employee"
attribute = "status"

ARGS = {
    "answers": answers,
    "exceptions": exceptions,
    "category": category,
    "element": element,
    "attribute": attribute,
}
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
extract_category Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/analyze_replies.py
code served (permissive licence) · get_code("1f0f74255167ba0c")
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 extract_category(file_to_analyze: str):
    return '_'.join(file_to_analyze.split('_')[:-1])

ARGS = {
    'file_to_analyze': 'model_checkpoint_v1_final'
}
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
extract_target Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_editing/utils.py
code served (permissive licence) · get_code("2ffb6d860fccc774")
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

# Set seed for reproducibility
torch.manual_seed(0)

# The function extract_target expects:
# 1. answers: List[str] - list of answer strings with format "name|span1|span2"
# 2. exceptions: dict - nested dict structure for filtering answers
# 3. category: str - category name
# 4. element: str - element name
# 5. attribute: Optional[str] - optional attribute name

# Build valid test data
answers = [
    "target_name|S:0",
    "other_name|S:1|E:5",
    "excluded_name|S:2",
]

exceptions = {
    "test_category": {
        "test_element": {
            "test_attribute": ["excluded_name"]
        }
    }
}

category = "test_category"
element = "test_element"
attribute = "test_attribute"

ARGS = {
    "answers": answers,
    "exceptions": exceptions,
    "category": category,
    "element": element,
    "attribute": attribute,
}
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
generate_answers Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/RAG/generate_rag_answers.py
code served (permissive licence) · get_code("c31983832684378f")
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
from argparse import Namespace
from transformers import BatchEncoding

torch.manual_seed(0)

# Create a minimal mock tokenizer
class MockTokenizer:
    def __init__(self):
        self.eos_token_id = 2
    
    def apply_chat_template(self, messages, return_tensors=None, padding=False):
        # Return a simple tensor for mistral
        return torch.tensor([[1, 2, 3, 4]])
    
    def __call__(self, text, return_tensors=None, padding=False):
        # Return BatchEncoding for llama2 and default case
        input_ids = torch.tensor([[1, 2, 3, 4]])
        attention_mask = torch.tensor([[1, 1, 1, 1]])
        return BatchEncoding({
            'input_ids': input_ids,
            'attention_mask': attention_mask
        })
    
    def batch_decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False):
        # Return list of strings
        return ["This is a decoded text response"]

# Create a minimal mock model
class MockModel(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = torch.nn.Linear(4, 4)
    
    def generate(self, input_ids, attention_mask=None, max_new_tokens=10, pad_token_id=None):
        # Return generated token ids
        batch_size = input_ids.shape[0]
        return torch.cat([input_ids, torch.full((batch_size, max_new_tokens), 1)], dim=1)
    
    def to(self, device):
        return super().to(device)

# Create instances
tokenizer = MockTokenizer()
model = MockModel()

# Create args namespace
args = Namespace(
    model_name='mistral-7b',
    device='cpu',
    max_length=20
)

# Create test questions dictionary
questions = {
    'q1': 'What is the capital of France?',
    'q2': 'Who wrote Romeo and Juliet?'
}

# Create context string
context = 'Paris is the capital of France. William Shakespeare wrote Romeo and Juliet.'

ARGS = {
    'questions': questions,
    'model': model,
    'tokenizer': tokenizer,
    'args': args,
    'context': context
}
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 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_questions Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/utils.py
code served (permissive licence) · get_code("f48962cb47272ac5")
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 json
import tempfile
import os

# Create a temporary GRC JSON file
grc_data = {
    "category1": {
        "element1": {
            "attribute1": {
                "questions": {
                    "type1": "What is the answer?",
                    "type2": "How does it work?"
                }
            },
            "attribute2": {
                "questions": {
                    "type1": "When did it happen?",
                    "type2": "Where is it?"
                }
            }
        },
        "element2": {
            "attribute1": {
                "questions": {
                    "type1": "Why is this?",
                    "type2": "Who knows?"
                }
            }
        }
    },
    "category2": {
        "element3": {
            "attribute1": {
                "questions": {
                    "type1": "Can you explain?",
                    "type2": "Is it true?"
                }
            }
        }
    },
    "athletes_byPayment": {
        "athlete1": {
            "questions": {
                "type1": "How much does he earn?",
                "type2": "What is his rank?"
            }
        },
        "athlete2": {
            "questions": {
                "type1": "What team is he on?",
                "type2": "How many goals?"
            }
        }
    },
    "companies_byRevenue": {
        "company1": {
            "questions": {
                "type1": "What is the revenue?",
                "type2": "How many employees?"
            }
        }
    }
}

# Write to temporary file
temp_fd, temp_path = tempfile.mkstemp(suffix='.json')
try:
    with os.fdopen(temp_fd, 'w') as f:
        json.dump(grc_data, f)
    
    ARGS = {
        'grc_path': temp_path,
        'prompt': 'additional context'
    }
finally:
    pass
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
is_exception Ran
ran, violated the contract we drafted
sislab-unitn/dyknow/models_editing/utils.py
code served (permissive licence) · get_code("04d2ecf0f938a61e")
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 torch
import numpy as np
from typing import Dict, List, Optional

torch.manual_seed(0)

# Build the exceptions dictionary with nested structure
# The function checks: category -> element -> [attribute or answer_name]
exceptions = {
    "math": {
        "addition": ["answer_1", "answer_2"],
        "subtraction": {
            "simple": ["answer_3", "answer_4"],
            "complex": ["answer_5"]
        }
    },
    "science": {
        "physics": ["answer_6"],
        "chemistry": {
            "organic": ["answer_7", "answer_8"],
            "inorganic": ["answer_9"]
        }
    }
}

ARGS = {
    "answer_name": "answer_1",
    "category": "math",
    "element": "addition",
    "attribute": None,
    "exceptions": exceptions
}
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 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
is_exception Ran
ran, violated the contract we drafted
sislab-unitn/dyknow/models_output/analyze_replies.py
code served (permissive licence) · get_code("821b26ef3a3a1e9f")
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 torch
import numpy as np
from typing import Dict, List, Optional, Set

torch.manual_seed(0)

# Construct arguments for is_exception function
answer_name = "answer_001"
category = "type_A"
element = "elem_1"
attribute = "attr_x"

# Build the exceptions dict with nested structure
exceptions = {
    "type_A": {
        "elem_1": {
            "attr_x": ["answer_001", "answer_002"],
            "attr_y": ["answer_003"]
        },
        "elem_2": ["answer_004"]
    },
    "type_B": {
        "elem_3": {
            "attr_z": ["answer_005"]
        }
    }
}

ARGS = {
    "answer_name": answer_name,
    "category": category,
    "element": element,
    "attribute": attribute,
    "exceptions": exceptions
}
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 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
load_json Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/utils.py
code served (permissive licence) · get_code("c5182d6eab527d0c")
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 json
import tempfile
import os

# Create a temporary JSON file for the function to load
temp_dir = tempfile.mkdtemp()
temp_json_path = os.path.join(temp_dir, "test.json")

# Write a simple JSON structure to the temporary file
test_data = {
    "key1": "value1",
    "key2": [1, 2, 3],
    "key3": {"nested": "object"}
}

with open(temp_json_path, 'w') as f:
    json.dump(test_data, f)

# Define the arguments for load_json
ARGS = {
    "path": temp_json_path
}
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
prepare_targets Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_editing/utils.py
code served (permissive licence) · get_code("12ea596f4cbe7637")
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.

# Analyze the function to understand what it needs:
# 1. category: str - a category name like 'athletes_byPayment', 'countries_byGDP', or 'organizations'
# 2. original: dict - a nested dict with structure depending on category
# 3. exceptions: dict - defaults to EXCEPTIONS, but can be overridden

# For 'athletes_byPayment' category (simplest case):
# original[category][element]['answers'] = List[str]
# Each answer string is formatted as "name|span1|span2" or "name|span1"
# We need to find entries with "S:" prefix in single-span answers

# Build test data for 'athletes_byPayment' category
original_athletes = {
    'athletes_byPayment': {
        'Athlete1': {
            'answers': [
                'Answer1|S:something',
                'Answer2|E:something',
            ]
        },
        'Athlete2': {
            'answers': [
                'Answer3|S:something',
                'Answer4|span1|span2',
            ]
        }
    }
}

# Build test data for 'countries_byGDP' category (has attributes)
original_countries = {
    'countries_byGDP': {
        'USA': {
            'gdp': {
                'answers': [
                    'Value1|S:data',
                    'Value2|E:data',
                ]
            },
            'population': {
                'answers': [
                    'Value3|S:data',
                    'Value4|E:data',
                ]
            }
        },
        'China': {
            'gdp': {
                'answers': [
                    'Value5|S:data',
                    'Value6|E:data',
                ]
            }
        }
    }
}

# Build test data for 'organizations' category (has attributes)
original_organizations = {
    'organizations': {
        'Company1': {
            'size': {
                'answers': [
                    'Small|S:attr',
                    'Large|E:attr',
                ]
            }
        }
    }
}

ARGS = {
    'category': 'athletes_byPayment',
    'original': original_athletes,
    'exceptions': {
        'athletes_byPayment': {
            'Lionel Messi': 'Argentina national association football team',
            'Neymar Jr.': 'Brazil national football team',
        }
    }
}
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
questionAnswering Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/models/api.py
code served (permissive licence) · get_code("8734c422f0011297")
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
from unittest.mock import Mock, MagicMock

torch.manual_seed(0)

# Create a mock OpenAI client
client = Mock()

# Mock the completions.create method for davinci-002
completion_response = Mock()
completion_response.choices = [Mock()]
completion_response.choices[0].text = " This is a test answer"
client.completions.create = Mock(return_value=completion_response)

# Mock the chat.completions.create method for other models
chat_response = Mock()
chat_response.choices = [Mock()]
chat_response.choices[0].message = Mock()
chat_response.choices[0].message.content = "This is a chat answer"
client.chat.completions.create = Mock(return_value=chat_response)

# Create args object with necessary attributes
args = Mock()
args.prompt = "You are a helpful assistant."
args.use_prompt = True
args.max_length = 100

# Define the ARGS dictionary
ARGS = {
    'question': "What is the capital of France?",
    'model': 'gpt-4',
    'client': client,
    'args': args
}
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
write_roman Ran
ran; our contract draft was wrong, not the code
sislab-unitn/dyknow/models_output/utils.py
code served (permissive licence) · get_code("32e0a42879a44343")
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

# Analyze the function: write_roman takes a single parameter 'num' of type int
# The function converts an integer to its Roman numeral representation
# It iterates through a ROMAN dictionary and builds the string representation

ARGS = {
    "num": 1994
}
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

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 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

LLMs acquire knowledge from massive data snapshots collected at different timestamps. Their knowledge is then commonly evaluated using static benchmarks. However, factual knowledge is generally subject to time-sensitive changes, and static benchmarks cannot address those cases. We present an approach to dynamically evaluate the knowledge in LLMs and their time-sensitiveness against Wikidata, a publicly available up-to-date knowledge graph. We evaluate the time-sensitive knowledge in twenty-four private and open-source LLMs, as well as the effectiveness of four editing methods in updating the outdated facts. Our results show that 1) outdatedness is a critical problem across state-of-the-art LLMs; 2) LLMs output inconsistent answers when prompted with slight variations of the question prompt; and 3) the performance of the state-of-the-art knowledge editing algorithms is very limited, as they can not reduce the cases of outdatedness and output inconsistency.

For agents

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

get_harvested_code_for_paper("2404.08700")
get_code_for_paper("2404.08700")
have("2404.08700")

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

curl https://syntology.ai/api/ran/2404.08700.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.08700.svg)](https://syntology.ai/paper/2404.08700)

Connect an agent — have() is free.