SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.05930 · 2024

WebLINX: Real-World Website Navigation with Multi-Turn Dialogue

arXiv · PDF · Open in the Atlas

Code that ran

21 of 27 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; 4 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 15 failures of our instrument; 6 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/2402.05930.json · badge: /api/ran/2402.05930.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
McGill-NLP/weblinx canonical (classifier label)
classifier basis: census
19 ran · 4 no recorded run
McGill-NLP/webllama canonical (classifier label)
classifier basis: classifier_2025
2 ran · 2 no recorded run
FunctionStatusWhere it lives
cast_to_float Ran
ran, honoured the contract we drafted
McGill-NLP/weblinx/weblinx/processing/outputs.py
code served (permissive licence) · get_code("fa80bfc9b9704f8c")
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 float, 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

def cast_to_float(value):
    """
    Checks if a value is an int or a float.
    """
    if value is None:
        return None
    if isinstance(value, str):
        value = value.strip()
    try:
        return float(value)
    except ValueError:
        return None

ARGS = {
    "value": 42.5
}
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 float, 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
compose_formatters Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/format.py
code served (permissive licence) · get_code("1ec7b369dd522796")
How it ran

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

import torch
import numpy as np

# Define a minimal Turn class that the formatters will receive
class Turn:
    def __init__(self):
        self.text = "Hello world"
        self.id = 1
        self.metadata = {"key": "value"}

# Define some simple formatter functions that take a Turn and return a dict
def formatter_1(turn: 'Turn') -> dict:
    return {"text": turn.text, "processed": True}

def formatter_2(turn: 'Turn') -> dict:
    return {"id": turn.id, "length": len(turn.text)}

def formatter_3(turn: 'Turn') -> dict:
    return {"metadata": turn.metadata, "timestamp": 12345}

# Create the arguments
ARGS = {
    "formatters": [formatter_1, formatter_2, formatter_3]
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
filter_bboxes Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/html.py
code served (permissive licence) · get_code("4fcb1951b117a6e6")
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.

# Setup code to construct valid arguments for filter_bboxes

ARGS = {
    "bboxes": {
        "widget_1": {
            "x": 10,
            "y": 20,
            "width": 50,
            "height": 60
        },
        "widget_2": {
            "x": 100,
            "y": 150,
            "width": 5,
            "height": 5
        },
        "widget_3": {
            "x": 200,
            "y": 300,
            "width": 30,
            "height": 8
        },
        "widget_4": {
            "x": 0,
            "y": 0,
            "width": -10,
            "height": 20
        }
    },
    "min_height": 10,
    "min_width": 10,
    "viewport_height": 500,
    "viewport_width": 400
}
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
format_action_arg_value Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/processing/outputs.py
code served (permissive licence) · get_code("bc51f017163fe65f")
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

# The function takes a single parameter: arg_value
# It expects a string and performs string operations on it
# Let's test with various string inputs that exercise different code paths

# Test case 1: quoted string
arg_value_1 = '"hello"'

# Test case 2: float string
arg_value_2 = '3.14'

# Test case 3: positive integer string
arg_value_3 = '42'

# Test case 4: negative integer string
arg_value_4 = '-99'

# Test case 5: plain string with whitespace
arg_value_5 = '  plain_text  '

ARGS = {
    'arg_value': '"hello"'
}
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
format_arg_item Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/format.py
code served (permissive licence) · get_code("4d03c832e78532b4")
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

class Turn:
    def __init__(self, args):
        self.args = args

# Create a minimal Turn object with args dictionary
turn = Turn({
    'value': '  some example text that might be long  ',
    'text': 'Hello world',
    'other_arg': 'test'
})

# Define ARGS dictionary with all parameters for format_arg_item
ARGS = {
    'turn': turn,
    'name': 'value',
    'strip': True,
    'max_length': 40,
    'return_as': 'dict'
}
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
format_prev_turns Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/processing/prompt.py
code served (permissive licence) · get_code("a899f2c938e9474d")
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.

class Turn:
    def __init__(self, index, intent):
        self.index = index
        self.intent = intent

class Replay:
    def __init__(self, turns):
        self.turns = turns
    
    def __getitem__(self, key):
        if isinstance(key, slice):
            return [self.turns[i] for i in range(*key.indices(len(self.turns)))]
        return self.turns[key]

def format_intent_func(turn, return_as=str):
    if return_as == str:
        return f"Turn {turn.index}: {turn.intent}"
    else:
        return {"turn_index": turn.index, "intent": turn.intent}

# Create sample turns
turns = [
    Turn(0, "action_0"),
    Turn(1, "action_1"),
    Turn(2, "action_2"),
    Turn(3, "action_3"),
    Turn(4, "action_4"),
    Turn(5, "action_5"),
    Turn(6, "action_6"),
]

# Create replay object
replay = Replay(turns)

# Current turn to get previous turns for
current_turn = Turn(4, "current_action")

ARGS = {
    "replay": replay,
    "turn": current_turn,
    "format_intent": format_intent_func,
    "turn_sep": " ; ",
    "num_prev_turns": 5,
    "return_as": str,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_bracket_length Ran
ran, honoured the contract we drafted
McGill-NLP/weblinx/weblinx/processing/truncation.py
code served (permissive licence) · get_code("c6518874884b35e2")
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 int, ndim 0, dtype family int, 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
from transformers import AutoTokenizer

# Create a minimal stand-in tokenizer class that implements the required interface
class StandInTokenizer:
    def __init__(self):
        pass
    
    def tokenize(self, text, add_special_tokens=False):
        # Simple tokenization: split by spaces and special characters
        # Return a list of tokens with length proportional to input
        if not text:
            return []
        # Simple mock: return list of characters as tokens
        return list(text)

# Use a real tokenizer from transformers for more realistic behavior
try:
    tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
except Exception:
    # Fallback to stand-in if download fails
    tokenizer = StandInTokenizer()

ARGS = {
    "elem_open_bracket": "<",
    "elem_close_bracket": ">",
    "tokenizer": tokenizer,
}
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 int, ndim 0, dtype family int, 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
get_descendants Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/processing/dom.py
code served (permissive licence) · get_code("44d3924c80fe3091")
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

# Create a simple tree-like structure using a stand-in class
class TreeNode:
    def __init__(self, value, children=None):
        self.value = value
        self.children = children if children is not None else []
    
    def __iter__(self):
        # This makes the node iterable, returning its children
        return iter(self.children)

# Build a small tree structure
leaf1 = TreeNode("leaf1", [])
leaf2 = TreeNode("leaf2", [])
leaf3 = TreeNode("leaf3", [])
internal_node1 = TreeNode("internal1", [leaf1, leaf2])
internal_node2 = TreeNode("internal2", [leaf3])
root = TreeNode("root", [internal_node1, internal_node2])

ARGS = {
    "node": root,
    "max_depth": 2,
    "current_depth": 0
}
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
get_env_dict Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/envs.py
code served (permissive licence) · get_code("abbcbd290c538f35")
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 tempfile
from pathlib import Path

# Create a temporary .env file with some environment variables
with tempfile.NamedTemporaryFile(mode='w', suffix='.env', delete=False) as f:
    f.write("API_KEY=secret123\n")
    f.write("DATABASE_URL=postgres://localhost\n")
    f.write("DEBUG=true\n")
    temp_env_path = f.name

ARGS = {
    'filepath': temp_env_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
get_list_from_records_by_key Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/recs.py
code served (permissive licence) · get_code("b070440172fb7a66")
How it ran

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

import torch

torch.manual_seed(0)

# The function expects:
# - records: dict (note: docstring says "list of dictionaries" but signature says dict)
#   Looking at the function body: [r[key] for r in records]
#   This iterates over records, so records must be iterable
#   Each r must support indexing with r[key]
#   So records should be a list of dicts, not a dict itself
# - key: str to extract from each record

# Create sample records as a list of dictionaries
records = [
    {"name": "Alice", "age": 30, "score": 95},
    {"name": "Bob", "age": 25, "score": 87},
    {"name": "Charlie", "age": 35, "score": 92},
]

key = "name"

ARGS = {
    "records": records,
    "key": key,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_speaker Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/processing/prompt.py
code served (permissive licence) · get_code("576db6c2c80cba48")
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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'utterance': 'say something speaker="instructor" to the user',
    'instructor_name': 'User',
    'navigator_name': 'Assistant',
    'default_name': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_truncation_offsets Ran
ran, honoured the contract we drafted
McGill-NLP/weblinx/weblinx/processing/truncation.py
code served (permissive licence) · get_code("e8d383f64048fb06")
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

torch.manual_seed(0)

# Construct a tokens dict with "input_ids" and "offset_mapping"
# offset_mapping is a list/array where each element is a tuple (start, end) of character offsets
tokens = {
    "input_ids": torch.tensor([[101, 2054, 2003, 102]]),  # dummy token IDs
    "offset_mapping": [
        (0, 0),      # [CLS] token
        (1, 5),      # word 1
        (6, 10),     # word 2
        (11, 15),    # word 3
        (15, 16),    # [SEP] token
        (16, 20),    # word 4
        (21, 25),    # word 5
        (26, 30),    # word 6
    ]
}

length = 8
num_tokens_to_remove = 2

ARGS = {
    "tokens": tokens,
    "length": length,
    "num_tokens_to_remove": num_tokens_to_remove,
}
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
group_record_to_dict Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/recs.py
code served (permissive licence) · get_code("176890ae96f63695")
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.

from copy import deepcopy
from collections import defaultdict
from typing import Any, List, Dict

# Construct records - a list of dictionaries with common keys
records = [
    {
        "category": "A",
        "type": 1,
        "value": 100,
        "name": "record1"
    },
    {
        "category": "A",
        "type": 1,
        "value": 200,
        "name": "record2"
    },
    {
        "category": "B",
        "type": 2,
        "value": 300,
        "name": "record3"
    },
    {
        "category": "B",
        "type": 2,
        "value": 400,
        "name": "record4"
    },
]

# Keys to group by
keys = ["category", "type"]

# Boolean flags
remove_keys = False
copy_flag = True

ARGS = {
    "records": records,
    "keys": keys,
    "remove_keys": remove_keys,
    "copy": copy_flag
}
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
has_elem_in_viewport Ran
ran, violated the contract we drafted
McGill-NLP/weblinx/weblinx/utils/html.py
code served (permissive licence) · get_code("1fec2e813060a926")
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.

# Stand-in class for webtasks.Turn
class Turn:
    def __init__(self, index, bboxes, viewport_width, viewport_height):
        self.index = index
        self.bboxes = bboxes
        self.viewport_width = viewport_width
        self.viewport_height = viewport_height

# Create a turn with some bboxes
turn = Turn(
    index=0,
    bboxes={
        "1234": {"x": 10, "y": 20, "width": 100, "height": 50},
        "5678": {"x": 200, "y": 300, "width": 5, "height": 10},
    },
    viewport_width=800,
    viewport_height=600
)

# Create element attributes dictionary
attrib_dict = {
    "data-webtasks-id": "1234",
    "class": "foo bar",
    "id": "my-element"
}

ARGS = {
    "turn": turn,
    "attrib_dict": attrib_dict,
    "key": "data-webtasks-id",
    "min_height": 2,
    "min_width": 2,
    "verbose": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
identity Ran
ran, honoured the contract we drafted
McGill-NLP/weblinx/weblinx/processing/prompt.py
code served (permissive licence) · get_code("6b063e5a47833dd4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) 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

torch.manual_seed(0)

ARGS = {
    "x": torch.randn(2, 4)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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_list_monotonically_increasing Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
McGill-NLP/weblinx/weblinx/utils/recs.py
code served (permissive licence) · get_code("3ae8877f0adf8db3")
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

torch.manual_seed(0)

# Test case 1: monotonically increasing list
lst_increasing = [1, 2, 3, 4, 5]

ARGS = {
    "lst": lst_increasing
}
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 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
list_all_non_alphanum_chars Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/processing/outputs.py
code served (permissive licence) · get_code("dcda9d27514120e0")
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

# Construct arguments for list_all_non_alphanum_chars
ARGS = {
    "other_chars_allowed": ['!', '@', '#', '$']
}
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
open_html_with_encodings Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/html.py
code served (permissive licence) · get_code("63c7e3d03045c0db")
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 tempfile
import os

# Create a temporary HTML file with valid UTF-8 encoding
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.html")

html_content = """<!DOCTYPE html>
<html>
<head>
    <title>Test Page</title>
</head>
<body>
    <h1>Hello World</h1>
    <p>This is a test HTML file.</p>
</body>
</html>"""

with open(temp_file, 'w', encoding='utf-8') as f:
    f.write(html_content)

ARGS = {
    'path': temp_file,
    'encodings': ['utf-8', 'latin-1'],
    'raise_error': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
remap_bboxes Ran
ran; our contract draft was wrong, not the code
McGill-NLP/webllama/examples/browsergym/agent.py
code served (permissive licence) · get_code("84bd76bf303ce78b")
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) 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

torch.manual_seed(0)

# The function remap_bboxes takes:
# 1. bboxes: a dictionary where keys are old ids and values are bbox data
# 2. attrs_map: a dictionary mapping old ids to new ids

# Create a simple bboxes dictionary with old ids as keys
bboxes = {
    0: torch.tensor([10.0, 20.0, 30.0, 40.0]),
    1: torch.tensor([50.0, 60.0, 70.0, 80.0]),
    2: torch.tensor([15.0, 25.0, 35.0, 45.0])
}

# Create attrs_map that maps old ids to new ids
attrs_map = {
    0: 'cat',
    1: 'dog',
    2: 'bird'
}

ARGS = {
    'bboxes': bboxes,
    'attrs_map': attrs_map
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
remove_latex Ran
ran, violated the contract we drafted
McGill-NLP/webllama/app/Results.py
code served (permissive licence) · get_code("d31d4a5938635d95")
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) 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

torch.manual_seed(0)

# Analyze the function:
# - Takes parameter 'x'
# - Checks if x is a tuple, str, or other type
# - If tuple, recursively applies remove_latex to each element
# - If str, performs string operations (startswith, partition, rpartition)
# - Otherwise returns x unchanged

# Test cases for the function:
test_cases = [
    # Simple string with LaTeX
    "prefix{content}suffix",
    # String without braces
    "simple_string",
    # String starting with 'Unnamed'
    "Unnamed: something",
    # Tuple of strings
    ("string1{latex}end", "Unnamed: test", "no_braces"),
    # Non-string type
    42,
    # Nested tuple
    (("inner{content}outer", "Unnamed"), "simple"),
    # Empty string
    "",
    # String with multiple braces
    "start{first}middle{second}end",
]

# The function parameter is 'x', which can be:
# - a string (with or without LaTeX)
# - a tuple (containing strings or other tuples)
# - any other type

ARGS = {
    "x": "example{LaTeX}text"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
shorten Ran
ran; our contract draft was wrong, not the code
McGill-NLP/weblinx/weblinx/utils/format.py
code served (permissive licence) · get_code("d20962095e5b96d0")
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 = {
    "text": "The quick brown fox jumps over the lazy dog. This is a longer text that will be truncated.",
    "max_length": 20
}
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
build_cond_df Not yet run
unverified (harvested, no recorded run)
McGill-NLP/webllama/app/Results.py
code served (permissive licence) · get_code("168f0e5f647cc7c9")
How it ran

no invocation recorded.

get_tree_repr_simple Not yet run
unverified (harvested, no recorded run)
McGill-NLP/weblinx/weblinx/processing/dom.py
code served (permissive licence) · get_code("af79280883191a10")
How it ran

no invocation recorded.

load_and_clean_df Not yet run
unverified (harvested, no recorded run)
McGill-NLP/webllama/app/Results.py
code served (permissive licence) · get_code("78b8c5d1e64ad4a3")
How it ran

no invocation recorded.

prune_tree Not yet run
unverified (harvested, no recorded run)
McGill-NLP/weblinx/weblinx/processing/dom.py
code served (permissive licence) · get_code("7c1d79a2be5ce989")
How it ran

no invocation recorded.

reduce_list_of_lengths Not yet run
unverified (harvested, no recorded run)
McGill-NLP/weblinx/weblinx/processing/truncation.py
code served (permissive licence) · get_code("0ac8898c0f3f43ec")
How it ran

no invocation recorded.

resolve_cache_path Not yet run
unverified (harvested, no recorded run)
McGill-NLP/weblinx/weblinx/utils/hydra.py
code served (permissive licence) · get_code("736f8a0adf254065")
How it ran

no invocation recorded.

Also present in other papers' code

Of 27 samples lifted from this paper's linked repositories, 27 carry a function identity and 0 do not. Of the 27 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 2 the same month, 2 later) and 25 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
identity 69 other papers
46 dated earlier, 1 the same month, 22 later · 5 only as an assembly brick
1706.07296 (2017-06)
A Minimal Developmental Model Can Increase Evolvability in Soft Robots
1802.09419 (2018-02)
Stochastic Hyperparameter Optimization through Hypernetworks
1804.06055 (2018-04)
Co-occurrence Feature Learning from Skeleton Data for Action Recognition and Detection with Hierarchical Aggregation
and 66 more not named here
get_descendants 3 other papers
1 dated earlier, 1 the same month, 1 later
2401.01614 (2024-01)
GPT-4V(ision) is a Generalist Web Agent, if Grounded
2402.15057 (2024-02)
On the Multi-turn Instruction Following for Conversational Web Agents
2409.11295 (2024-09)
EIA: Environmental Injection Attack on Generalist Web Agents for Privacy Leakage

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

We propose the problem of conversational web navigation, where a digital agent controls a web browser and follows user instructions to solve real-world tasks in a multi-turn dialogue fashion. To support this problem, we introduce WEBLINX - a large-scale benchmark of 100K interactions across 2300 expert demonstrations of conversational web navigation. Our benchmark covers a broad range of patterns on over 150 real-world websites and can be used to train and evaluate agents in diverse scenarios. Due to the magnitude of information present, Large Language Models (LLMs) cannot process entire web pages in real-time. To solve this bottleneck, we design a retrieval-inspired model that efficiently prunes HTML pages by ranking relevant elements. We use the selected elements, along with screenshots and action history, to assess a variety of models for their ability to replicate human behavior when navigating the web. Our experiments span from small text-only to proprietary multimodal LLMs. We find that smaller finetuned decoders surpass the best zero-shot LLMs (including GPT-4V), but also larger finetuned multimodal models which were explicitly pretrained on screenshots. However, all finetuned models struggle to generalize to unseen websites. Our findings highlight the need for large multimodal models that can generalize to novel settings. Our code, data and models are available for research: https://mcgill-nlp.github.io/weblinx

For agents

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

get_harvested_code_for_paper("2402.05930")
get_code_for_paper("2402.05930")
have("2402.05930")

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

curl https://syntology.ai/api/ran/2402.05930.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.05930.svg)](https://syntology.ai/paper/2402.05930)

Connect an agent — have() is free.