SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.18779 · 2026

MANGO: Multi-Agent Web Navigation via Global-View Optimization

Tianyi Zhang, Weixi Tong, Yifeng Di

arXiv · PDF · Open in the Atlas

Code that ran

6 of 6 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 0 no recorded run), measured 2026-11-22; 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/2604.18779.json · badge: /api/ran/2604.18779.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
VichyTong/Mango canonical (classifier label)
classifier basis: census
6 ran · 0 no recorded run
FunctionStatusWhere it lives
count_tokens Ran
ran, honoured the contract we drafted
VichyTong/Mango/llm_web_scraper/utils/llm.py
code served (permissive licence) · get_code("04d9feb321c92fde")
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.

class MockUsage:
    def __init__(self, total_tokens):
        self.total_tokens = total_tokens

class MockResponse:
    def __init__(self, total_tokens):
        self.usage = MockUsage(total_tokens)

raw_responses = [
    MockResponse(100),
    MockResponse(250),
    MockResponse(175)
]

ARGS = {
    "raw_responses": raw_responses
}
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 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_final_answer_prompt Ran
ran; our contract draft was wrong, not the code
VichyTong/Mango/llm_web_scraper/prompts/final_answer.py
code served (permissive licence) · get_code("946f958e0aae03a4")
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
from typing import List, Dict

torch.manual_seed(0)

# Construct extracted_contents: List[Dict[str, str]]
extracted_contents = [
    {
        "title": "Document 1",
        "text": "This is the first extracted content piece."
    },
    {
        "title": "Document 2",
        "text": "This is the second extracted content piece with more details."
    },
    {
        "title": "Source",
        "text": "Additional metadata about the source."
    }
]

# Construct user_intent: str
user_intent = "Find information about machine learning frameworks and their applications"

# Define ARGS mapping
ARGS = {
    'extracted_contents': extracted_contents,
    'user_intent': user_intent
}
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
get_navigation_prompt Ran
ran; our contract draft was wrong, not the code
VichyTong/Mango/llm_web_scraper/prompts/simple_navigation.py
code served (permissive licence) · get_code("7110792cb1365676")
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 = {
    "user_intent": "Find information about machine learning",
    "url": "https://example.com"
}
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
load_google_preprocessing_data Ran
ran; our contract draft was wrong, not the code
VichyTong/Mango/evaluation/preprocess/run_preprocessing.py
code served (permissive licence) · get_code("25067f054560a282")
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 os
import tempfile
from pathlib import Path

# Create a temporary directory structure to support the function
temp_dir = tempfile.mkdtemp()
output_dir = temp_dir

# Create the necessary directory structure
google_dir = os.path.join(output_dir, 'webwalker', 'preprocess', 'google', 'my-model')
os.makedirs(google_dir, exist_ok=True)

# Create a sample url_lists.json file with the expected structure
sample_data = [
    {'task_index': 0, 'url': 'https://example.com/0'},
    {'task_index': 1, 'url': 'https://example.com/1'},
    {'task_index': 2, 'url': 'https://example.com/2'}
]

google_file = os.path.join(google_dir, 'url_lists.json')
with open(google_file, 'w', encoding='utf-8') as f:
    json.dump(sample_data, f)

# Define arguments for the function
ARGS = {
    'output_dir': output_dir,
    'model_name': 'tensorblock/my-model'
}
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
load_google_preprocessing_data_webvoyager Ran
ran; our contract draft was wrong, not the code
VichyTong/Mango/evaluation/preprocess/run_preprocessing.py
code served (permissive licence) · get_code("c8727c9ea77eef0f")
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 os
import tempfile
import shutil

# Create a temporary directory structure for the test
temp_base = tempfile.mkdtemp()

# Create the necessary directory structure
webvoyager_dir = os.path.join(temp_base, 'webvoyager', 'preprocess', 'google', 'llama-2-7b')
os.makedirs(webvoyager_dir, exist_ok=True)

# Create sample preprocessing data
sample_data = [
    {'task_id': 'task_1', 'url': 'https://example.com/1'},
    {'task_id': 'task_2', 'url': 'https://example.com/2'},
    {'task_id': 'task_3', 'url': 'https://example.com/3'},
]

url_lists_file = os.path.join(webvoyager_dir, 'url_lists.json')
with open(url_lists_file, 'w', encoding='utf-8') as f:
    json.dump(sample_data, f)

# Set up arguments
ARGS = {
    'output_dir': temp_base,
    'model_name': 'tensorblock/llama-2-7b',
}
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
normalize_model_name_for_path Ran
ran; our contract draft was wrong, not the code
VichyTong/Mango/evaluation/preprocess/run_preprocessing.py
code served (permissive licence) · get_code("011bcb489652a121")
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 normalize_model_name_for_path(model_name: str) -> str:
    if model_name.startswith('tensorblock/'):
        return model_name.replace('tensorblock/', '')
    return model_name

# Test case 1: model name with 'tensorblock/' prefix
model_name_with_prefix = 'tensorblock/my-model'

ARGS = {
    'model_name': model_name_with_prefix
}
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 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 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 6 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

Existing web agents typically initiate exploration from the root URL, which is inefficient for complex websites with deep hierarchical structures. Without a global view of the website's structure, agents frequently fall into navigation traps, explore irrelevant branches, or fail to reach target information within a limited budget. We propose MANGO, a multiagent web navigation method that leverages the website structure to dynamically determine optimal starting points. We formulate URL selection as a multi-armed bandit problem and employ Thompson Sampling to adaptively allocate the navigation budget across candidate URLs. Furthermore, we introduce an episodic memory component to store navigation history, enabling the agent to learn from previous attempts. Experiments on WebVoyager demonstrate that MANGO achieves a success rate of 63.6% when using GPT-5-mini, outperforming the best baseline by 7.3%. Furthermore, on WebWalkerQA, MANGO attains a 52.5% success rate, surpassing the best baseline by 26.8%. We also demonstrate the generalizability of MANGO using both open-source and closed-source models as backbones. Our data and code are open-source and available at https://github.com/VichyTong/Mango.

For agents

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

get_harvested_code_for_paper("2604.18779")
get_code_for_paper("2604.18779")
have("2604.18779")

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

curl https://syntology.ai/api/ran/2604.18779.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.18779.svg)](https://syntology.ai/paper/2604.18779)

Connect an agent — have() is free.