SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.22859 · ICML · 2026

From Blind Spots to Gains: Diagnostic-Driven Iterative Training for Large Multimodal Models

Wei Ye, Shikun Zhang, Chaoya Jiang, Yongrui Heng, Hongrui Jia

arXiv · PDF · Open in the Atlas

Code that ran

10 of 10 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; 3 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 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/2602.22859.json · badge: /api/ran/2602.22859.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
hongruijia/DPE canonical (classifier label)
classifier basis: census
10 ran · 0 no recorded run
FunctionStatusWhere it lives
add_default_scores Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/merge_and_convert.py
code served (permissive licence) · get_code("cb8e8922a74cff4b")
How it ran

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

import torch
import numpy as np

torch.manual_seed(0)

# Construct data_list: a list of dictionaries, some with 'score' and some without
data_list = [
    {'value': 1, 'name': 'item1'},  # No 'score' key
    {'value': 2, 'name': 'item2', 'score': 0.5},  # Has 'score' key
    {'value': 3, 'name': 'item3'},  # No 'score' key
]

ARGS = {
    'data_list': data_list,
    'default_score': 1.0,
    'use_existing_scores': True,
}
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
aggregate_by_imagetype Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/regenerate_prompt_with_llm.py
code served (permissive licence) · get_code("c4539c57a9235a68")
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 collections import defaultdict

torch.manual_seed(0)

# Construct a realistic results list to pass to aggregate_by_imagetype
results = [
    {
        'image_type': 'geometric',
        'is_correct': True,
    },
    {
        'image_type': 'geometric',
        'is_correct': False,
        'error_type': 'misidentification',
        'error_description': 'Failed to identify the triangle shape correctly'
    },
    {
        'image_type': 'medical',
        'is_correct': True,
    },
    {
        'image_type': 'medical',
        'is_correct': False,
        'error_type': 'segmentation_error',
        'error_description': 'Incorrect organ boundary detection in CT scan'
    },
    {
        'image_type': 'chart_graph',
        'is_correct': True,
    },
    {
        'image_type': 'text_dense',
        'is_correct': False,
        'error_type': 'ocr_error',
        'error_description': 'Could not properly read dense text in the image due to low resolution and overlapping characters'
    },
    {
        'image_type': 'diagram',
        'is_correct': True,
    },
    {
        'image_type': 'math_formula',
        'is_correct': False,
        'error_type': 'parsing_error',
        'error_description': 'Failed to parse complex mathematical notation'
    },
    {
        'image_type': 'map',
        'is_correct': True,
    },
    {
        'image_type': 'natural_scene',
        'is_correct': True,
    },
]

ARGS = {
    'results': results
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
analyze_weakness_with_llm Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/regenerate_prompt_with_llm.py
code served (permissive licence) · get_code("0267e29344d42128")
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 os

# Mock the environment variable if needed
os.environ['DASHSCOPE_API_KEY'] = 'test-api-key-12345'

# Define the arguments for analyze_weakness_with_llm
ARGS = {
    'img_type': 'indoor',
    'error_descriptions': [
        'Model failed to detect small objects in cluttered scenes',
        'Incorrect color classification in low light conditions',
        'Failed to recognize rotated text in images',
        'Misidentified similar looking objects',
        'Poor performance on partially occluded objects',
        'Confusion between foreground and background elements',
        'Inability to handle extreme aspect ratios'
    ],
    'error_type': 'object_detection'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
b64_to_image Ran
ran, honoured the contract we drafted
hongruijia/DPE/question_evaluate/evaluate.py
code served (permissive licence) · get_code("8db1299d2238c123")
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 Image. 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
from io import BytesIO
from PIL import Image
import numpy as np

# Create a small RGB image
img = Image.new('RGB', (4, 3), color=(255, 0, 0))

# Convert to bytes
img_bytes = BytesIO()
img.save(img_bytes, format='PNG')
img_bytes.seek(0)

# Encode to base64
b64_str = base64.b64encode(img_bytes.getvalue()).decode('utf-8')

ARGS = {
    'b64_str': b64_str
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
b64_to_image Ran
ran, honoured the contract we drafted
hongruijia/DPE/question_evaluate/generate_answers_parallel.py
code served (permissive licence) · get_code("68177b1800fe39d4")
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 Image. 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 argparse
import base64
from io import BytesIO
from PIL import Image
import numpy as np

# Set up the argument parser with a gpu attribute
parser = argparse.ArgumentParser(description='使用vLLM生成模型答案')
args = parser.parse_args([])
args.gpu = 0

# Create a valid base64-encoded image
def create_test_image_b64(width=300, height=300):
    """Create a small test image and encode it as base64"""
    img = Image.new('RGB', (width, height), color=(73, 109, 137))
    img_bytes = BytesIO()
    img.save(img_bytes, format='PNG')
    img_bytes.seek(0)
    b64_str = base64.b64encode(img_bytes.read()).decode('utf-8')
    return b64_str

# Create test cases: normal size and small size
normal_size_b64 = create_test_image_b64(300, 300)
small_size_b64 = create_test_image_b64(100, 100)

ARGS = {
    'b64_str': normal_size_b64,
    'min_size': 224
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
convert_to_parquet Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/merge_and_convert.py
code served (permissive licence) · get_code("a358bb64d8f0c7a4")
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 json
import pandas as pd
import os
import tempfile

# Create a temporary directory for output
temp_dir = tempfile.mkdtemp()

# Create a temporary JSON file with valid test data
test_data = [
    {
        "question": "What is 2+2?",
        "answer": "4",
        "score": 0.9,
        "image": "image1.jpg",
        "question_type": "math",
        "image_type": "diagram"
    },
    {
        "question": "What is the capital of France?",
        "answer": "Paris",
        "score": 0.95,
        "image": "image2.jpg",
        "question_type": "geography",
        "image_type": "map"
    },
    {
        "question": "Describe photosynthesis",
        "answer": "Process by which plants convert light to chemical energy",
        "score": 0.5,
        "image": "image3.jpg",
        "question_type": "biology",
        "image_type": "photo"
    },
    {
        "question": "What is DNA?",
        "answer": "Deoxyribonucleic acid",
        "score": 0.3,
        "image": "image4.jpg",
        "question_type": "biology",
        "image_type": "diagram"
    },
    {
        "question": "Calculate the derivative",
        "answer": "2x",
        "score": 0.85,
        "image": "image5.jpg",
        "question_type": "math",
        "image_type": "formula"
    },
    {
        "question": "What is Newton's first law?",
        "answer": "An object in motion stays in motion",
        "score": 0.2,
        "image": "image6.jpg",
        "question_type": "physics",
        "image_type": "diagram"
    }
]

json_file = os.path.join(temp_dir, "test_data.json")
with open(json_file, 'w') as f:
    json.dump(test_data, f)

ARGS = {
    "json_file": json_file,
    "output_dir": temp_dir,
    "save_name": "test_conversion",
    "min_score": 0.5,
    "max_score": 1.0,
    "use_existing_scores": True,
    "filter_by_score": True,
    "target_samples": 3,
    "enable_补充": True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
decode_base64_image Ran
ran, honoured the contract we drafted
hongruijia/DPE/question_evaluate/score_questions.py
code served (permissive licence) · get_code("6c3a03cdc78f51d2")
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 Image. 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 argparse
import base64
from io import BytesIO
from PIL import Image
import numpy as np

parser = argparse.ArgumentParser(description='为生成的问题评分(自采样一致性)')
args = parser.parse_args()

def decode_base64_image(b64_str):
    try:
        if not b64_str:
            return None
        img_bytes = base64.b64decode(b64_str)
        return Image.open(BytesIO(img_bytes)).convert('RGB')
    except Exception as e:
        print(f'[{args.suffix}] 图片解码失败: {e}')
        return None

# Create a small valid base64-encoded image
img = Image.new('RGB', (4, 3), color='red')
buffer = BytesIO()
img.save(buffer, format='PNG')
buffer.seek(0)
b64_image = base64.b64encode(buffer.getvalue()).decode('utf-8')

ARGS = {
    'b64_str': b64_image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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_weakness_prompt Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/regenerate_prompt_with_llm.py
code served (permissive licence) · get_code("99097aa76bc784c2")
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 json

# Build a realistic stats dict based on how generate_weakness_prompt uses it
stats = {
    "chart_images": {
        "accuracy": 0.65,
        "total_evaluated": 50,
        "error_descriptions_by_type": {
            "misread_axis": [
                "Model confused x-axis label with y-axis",
                "Failed to identify logarithmic scale",
                "Misinterpreted axis range"
            ],
            "wrong_value": [
                "Off by factor of 2 in reading bar height",
                "Estimated 150 instead of 120"
            ]
        }
    },
    "diagram_images": {
        "accuracy": 0.78,
        "total_evaluated": 40,
        "error_descriptions_by_type": {
            "spatial_relation": [
                "Confused left/right position",
                "Missed hierarchical nesting"
            ]
        }
    },
    "photo_images": {
        "accuracy": 0.92,
        "total_evaluated": 100,
        "error_descriptions_by_type": {}
    }
}

ARGS = {
    "stats": stats
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
requests — 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
merge_by_placement Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/scripts_Qwen-VL-7B/model_merger.py
code served (permissive licence) · get_code("32cbd7aa87e2f64b")
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 (4, 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
from torch.distributed._tensor import Placement, Shard, Replicate

torch.manual_seed(0)

# Create small test tensors
# For Shard placement, we need multiple tensors that can be concatenated
tensor1 = torch.randn(2, 4)
tensor2 = torch.randn(2, 4)
tensors = [tensor1, tensor2]

# Create a Shard placement object with dim=0
placement = Shard(dim=0)

ARGS = {
    "tensors": tensors,
    "placement": placement,
}
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 (4, 4) — shape (4, 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
merge_json_files Ran
ran; our contract draft was wrong, not the code
hongruijia/DPE/question_evaluate/merge_and_convert.py
code served (permissive licence) · get_code("532a063c4d397d63")
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 json
import os
import tempfile

# Create temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create test JSON files
test_file_1 = os.path.join(temp_dir, 'data1.json')
test_file_2 = os.path.join(temp_dir, 'data2.json')
output_file = os.path.join(temp_dir, 'merged_output.json')

# Write test data to input files
data1 = [
    {"id": 1, "name": "item1"},
    {"id": 2, "name": "item2"}
]
with open(test_file_1, 'w') as f:
    json.dump(data1, f)

data2 = [
    {"id": 3, "name": "item3"},
    {"id": 4, "name": "item4"}
]
with open(test_file_2, 'w') as f:
    json.dump(data2, f)

# Define arguments
ARGS = {
    'input_files': [test_file_1, test_file_2],
    'output_file': output_file
}
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

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 1 later) and 9 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
merge_by_placement 6 other papers
4 dated earlier, 1 the same month, 1 later
2410.21236 (2024-10)
Flaming-hot Initiation with Regular Execution Sampling for Large Language Models
2505.12081 (2025-05)
VisionReasoner: Unified Reasoning-Integrated Visual Perception via Reinforcement Learning
2505.13417 (2025-05)
AdaptThink: Reasoning Models Can Learn When to Think
and 3 more not named here

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

As Large Multimodal Models (LMMs) scale up and reinforcement learning (RL) methods mature, LMMs have made notable progress in complex reasoning and decision making. Yet training still relies on static data and fixed recipes, making it difficult to diagnose capability blind spots or provide dynamic, targeted reinforcement. Motivated by findings that test driven error exposure and feedback based correction outperform repetitive practice, we propose Diagnostic-driven Progressive Evolution (DPE), a spiral loop where diagnosis steers data generation and reinforcement, and each iteration re-diagnoses the updated model to drive the next round of targeted improvement. DPE has two key components. First, multiple agents annotate and quality control massive unlabeled multimodal data, using tools such as web search and image editing to produce diverse, realistic samples. Second, DPE attributes failures to specific weaknesses, dynamically adjusts the data mixture, and guides agents to generate weakness focused data for targeted reinforcement. Experiments on Qwen3-VL-8B-Instruct and Qwen2.5-VL-7B-Instruct show stable, continual gains across eleven benchmarks, indicating DPE as a scalable paradigm for continual LMM training under open task distributions. Our code, models, and data are publicly available at https://github.com/hongruijia/DPE.

For agents

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

get_harvested_code_for_paper("2602.22859")
get_code_for_paper("2602.22859")
have("2602.22859")

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

curl https://syntology.ai/api/ran/2602.22859.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.22859.svg)](https://syntology.ai/paper/2602.22859)

Connect an agent — have() is free.