SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.04017 · 2026

GeoBrowse: A Geolocation Benchmark for Agentic Tool Use with Expert-Annotated Reasoning Traces

Yi Fung, Rui Min, Tianqing Fang, Yuyang Zhang, Xinyu Geng, Xinyan Liu, Hanwen Wang, Yanjing Xiao

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 6 failures of our instrument; 1 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/2604.04017.json · badge: /api/ran/2604.04017.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
ornamentt/GeoBrowse canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
contains_chinese Ran
ran, violated the contract we drafted
ornamentt/GeoBrowse/Benchmark/codes/data_processing/translate.py
code served (permissive licence) · get_code("3b0c1028ae6d795a")
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 re

_CHINESE_RE = re.compile('[\\u4e00-\\u9fff]')

def contains_chinese(s: str) -> bool:
    return bool(_CHINESE_RE.search(s))

ARGS = {
    "s": "你好世界"
}
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
encode_image_to_base64 Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/Benchmark/codes/infer/baseline.py
code served (permissive licence) · get_code("ae6c2bb194221137")
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 base64
import tempfile
import os

# Create a temporary image file for testing
temp_dir = tempfile.mkdtemp()
image_path = os.path.join(temp_dir, "test_image.png")

# Create a minimal valid PNG file (1x1 pixel)
png_bytes = bytes([
    0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A,  # PNG signature
    0x00, 0x00, 0x00, 0x0D, 0x49, 0x48, 0x44, 0x52,  # IHDR chunk
    0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
    0x08, 0x02, 0x00, 0x00, 0x00, 0x90, 0x77, 0x53,
    0xDE, 0x00, 0x00, 0x00, 0x0C, 0x49, 0x44, 0x41,
    0x54, 0x08, 0x99, 0x63, 0xF8, 0xCF, 0xC0, 0x00,
    0x00, 0x03, 0x01, 0x01, 0x00, 0x18, 0xDD, 0x8D,
    0xB4, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, 0x4E,
    0x44, 0xAE, 0x42, 0x60, 0x82
])

with open(image_path, 'wb') as f:
    f.write(png_bytes)

ARGS = {
    'image_path': image_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 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_response Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/Evaluation/baseline/evaluate.py
pointer only (licence: MIT); the stored copy get_code serves, from leaves-slient/RT-Bench, is pointer only (licence: NONE) · get_code("a4ee90a1bd5be041")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from leaves-slient/RT-Bench, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_until_hit Ran
ran, honoured the contract we drafted
ornamentt/GeoBrowse/Evaluation/baseline/utils.py
code served (permissive licence) · get_code("ce1d671345b6ef99")
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 int, 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
import numpy as np

torch.manual_seed(0)

# Construct arguments for get_until_hit(d, keys, df=None)
# The function iterates through keys and returns the first value found in dict d,
# or returns df if no key is found

# Create a dictionary with some sample data
d = {
    'key1': 'value1',
    'key2': torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]]),
    'key3': 42,
    'key4': [1, 2, 3]
}

# Create a list of keys to search through
keys = ['key_not_found', 'key2', 'key1']

# Default value to return if no key is found
df = 'default_fallback'

ARGS = {
    'd': d,
    'keys': keys,
    'df': df
}
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 Tensor, ndim 2, dtype family int, 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
load_jsonl Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/Evaluation/baseline/evaluate.py
pointer only (licence: MIT); the stored copy get_code serves, from leaves-slient/RT-Bench, is pointer only (licence: NONE) · get_code("e587e0fad56e82f2")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from leaves-slient/RT-Bench, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

normalize_number_str Ran
ran, honoured the contract we drafted
ornamentt/GeoBrowse/System/ckv3/ck_main/gaia_scorer.py
code served (permissive licence) · get_code("dce19b095f363b41")
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

torch.manual_seed(0)

ARGS = {
    "number_str": "$1,234.56"
}
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 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
question_scorer Ran
ran, violated the contract we drafted
ornamentt/GeoBrowse/System/ckv3/ck_main/gaia_scorer.py
code served (permissive licence) · get_code("9b00c498f57a78b5")
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 re
import string
import warnings

# Set up arguments for question_scorer function
ARGS = {
    "model_answer": "42.5",
    "ground_truth": "42.5"
}
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
split_string Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/System/ckv3/ck_main/gaia_scorer.py
code served (permissive licence) · get_code("bd302a2ed08c2f35")
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 re

def split_string(s: str, char_list: list[str]=[',', ';']) -> list[str]:
    pattern = f"[{''.join(char_list)}]"
    return re.split(pattern, s)

ARGS = {
    's': 'apple,banana;orange,grape',
    'char_list': [',', ';']
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tuple_keys_to_str Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/Evaluation/baseline/utils.py
code served (permissive licence) · get_code("67484bbc0c20d4f6")
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

# Analyze the function: tuple_keys_to_str recursively processes dicts, lists, and other values
# It converts dict keys to strings (str(k)) and recurses into lists and dict values
# The function only needs a single parameter: a dict (or nested structure)

# Test case 1: Simple dict with tuple keys
test_dict_1 = {
    (1, 2): "value1",
    (3, 4): "value2"
}

# Test case 2: Nested dict with tuple keys and lists
test_dict_2 = {
    (5, 6): [
        {(7, 8): "nested_value"},
        "list_item"
    ],
    "normal_string_key": {
        (9, 10): torch.tensor([1, 2, 3, 4])
    }
}

# Test case 3: Complex nested structure
test_dict_3 = {
    (11, 12): [
        [
            {(13, 14): np.array([1, 2, 3])},
            "deep_item"
        ]
    ],
    ("a", "b", "c"): {
        (15,): [1, 2, 3],
        ("x", "y"): "final_value"
    }
}

# The function signature has one parameter 'd'
ARGS = {
    "d": test_dict_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 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
wrapped_trying Ran
ran; our contract draft was wrong, not the code
ornamentt/GeoBrowse/Evaluation/baseline/utils.py
code served (permissive licence) · get_code("160fcf9f63160edb")
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 random
import time

def simple_func():
    return "success"

def failing_func():
    raise ValueError("Test error")

def timeout_func():
    time.sleep(1)
    return "done"

ARGS = {
    "func": simple_func,
    "default_return": None,
    "max_times": 3,
    "wait_error_names": ("TimeoutError", "ConnectionError"),
    "reraise": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
rich, 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_template Not yet run
unverified (harvested, no recorded run)
ornamentt/GeoBrowse/System/ckv3/agents/agent.py
code served (permissive licence) · get_code("400fc092b5b46134")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 5 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 3 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.

FunctionAlso present inEarliest-datedOthers named
encode_image_to_base64 3 other papers
2 dated earlier, 1 the same month
2505.14362 (2025-05)
DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning
2509.02807 (2025-09)
PixFoundation 2.0: Do Video Multi-Modal LLMs Use Motion in Visual Grounding?
2604.05818 (2026-04)
WikiSeeker: Rethinking the Role of Vision-Language Models in Knowledge-Based Visual Question Answering
extract_response 2 other papers
1 dated earlier, 1 the same month
2507.02592 (2025-07)
WebSailor: Navigating Super-human Reasoning for Web Agent
2604.16349 (2026-04)
Benchmarking Real-Time Question Answering via Executable Code Workflows
load_jsonl 2 other papers
1 dated earlier, 1 the same month
2507.02592 (2025-07)
WebSailor: Navigating Super-human Reasoning for Web Agent
2604.16349 (2026-04)
Benchmarking Real-Time Question Answering via Executable Code Workflows
normalize_number_str 1 other paper
1 dated earlier
2407.10718 (2024-07)
Sibyl: Simple yet Effective Agent Framework for Complex Real-world Reasoning
split_string 1 other paper
1 dated earlier
2407.10718 (2024-07)
Sibyl: Simple yet Effective Agent Framework for Complex Real-world Reasoning

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

Deep research agents integrate fragmented evidence through multi-step tool use. BrowseComp offers a text-only testbed for such agents, but existing multimodal benchmarks rarely require both weak visual cues composition and BrowseComp-style multi-hop verification. Geolocation is a natural testbed because answers depend on combining multiple ambiguous visual cues and validating them with open-web evidence. Thus, we introduce GeoBrowse, a geolocation benchmark that combines visual reasoning with knowledge-intensive multi-hop queries. Level 1 tests extracting and composing fragmented visual cues, and Level 2 increases query difficulty by injecting long-tail knowledge and obfuscating key entities. To support evaluation, we provide an agentic workflow GATE with five think-with-image tools and four knowledge-intensive tools, and release expert-annotated stepwise traces grounded in verifiable evidence for trajectory-level analysis. Experiments show that GATE outperforms direct inference and open-source agents, indicating that no-tool, searchonly or image-only setups are insufficient. Gains come from coherent, level-specific tool-use plans rather than more tool calls, as they more reliably reach annotated key evidence steps and make fewer errors when integrating into the final decision. The GeoBrowse bernchmark and codes are provided in https://github.com/ornamentt/GeoBrowse

For agents

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

get_harvested_code_for_paper("2604.04017")
get_code_for_paper("2604.04017")
have("2604.04017")

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

curl https://syntology.ai/api/ran/2604.04017.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.04017.svg)](https://syntology.ai/paper/2604.04017)

Connect an agent — have() is free.