SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.03548 · 2026

SEAD: Self-Evolving Agent for Multi-Turn Service Dialogue

Wei Zhang, Jie Wang, Yuqin Dai, Zichen Luo, Ning Gao

arXiv · PDF · Open in the Atlas

Code that ran

19 of 20 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, 1 violated it and 0 have no contract verdict; 14 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/2602.03548.json · badge: /api/ran/2602.03548.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
Da1yuqin/SEAD canonical (classifier label)
classifier basis: census
19 ran · 1 no recorded run
FunctionStatusWhere it lives
build_chat_messages Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/SEAD/vllm_service/vllm_server.py
code served (permissive licence) · get_code("96f494ea053bac2d")
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 logging
from typing import List, Dict, Any

logging.basicConfig(level=logging.INFO)

# Set up arguments for build_chat_messages function
base_system_prompt_template = """You are a helpful assistant.
Initial cooperation: ${cooperation}
Initial emotion: ${emotion}
Initial trust: ${trust}
Specific behaviors: ${behaviors}
"""

dialogue_history = [
    {
        "role": "user",
        "content": "<user_reply>Hello, I need help with my account.</user_reply>"
    },
    {
        "role": "assistant",
        "content": "I'd be happy to help you with your account. What seems to be the issue?"
    },
    {
        "role": "user",
        "content": "<user_reply>I can't reset my password.</user_reply>"
    }
]

user_params = {
    "cooperation": "high",
    "emotion": "neutral",
    "trust": "medium",
    "behaviors": ["be helpful", "respond quickly", "stay professional"]
}

mode = "call_chatbot"

ARGS = {
    "base_system_prompt_template": base_system_prompt_template,
    "dialogue_history": dialogue_history,
    "user_params": user_params,
    "mode": mode
}
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
calculate_completion_rate Ran
ran, honoured the contract we drafted
Da1yuqin/SEAD/utils/analyze_chatbot_mistakes.py
code served (permissive licence) · get_code("3c337ab1bef2b5b1")
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
from typing import List, Dict

torch.manual_seed(0)

# Construct test dialogues
# The function checks if d.get('batch_statuses', []) exists and if d['batch_statuses'][-1] == 1
dialogues = [
    {
        'batch_statuses': [0, 0, 1]  # success_count += 1 (last status is 1)
    },
    {
        'batch_statuses': [1, 0, 0]  # success_count += 0 (last status is 0)
    },
    {
        'batch_statuses': [1]  # success_count += 1 (last status is 1)
    },
    {
        'id': 'dialogue_4'  # success_count += 0 (no batch_statuses key)
    },
    {
        'batch_statuses': []  # success_count += 0 (empty list)
    }
]

ARGS = {
    'dialogues': dialogues
}
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
check_vllm_service_health Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/for_evaluation/Baseline_test_local_models_vllm.py
code served (permissive licence) · get_code("5b7295d80bd7a043")
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. 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, Tuple

torch.manual_seed(0)

# The function check_vllm_service_health takes a list of server URLs and a timeout
# It makes HTTP requests to check health status
# We need to provide valid URLs (they don't need to actually work since exceptions are caught)

ARGS = {
    "server_urls": [
        "http://localhost:8000",
        "http://localhost:8001",
        "http://127.0.0.1:8002"
    ],
    "timeout": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, requests, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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_initial_state Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/analyze_chatbot_mistakes.py
code served (permissive licence) · get_code("b364d075c39bce10")
How it ran

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

import re
from typing import List, Dict, Tuple

# Create a valid dialogue dict with ground_truth_history
dialogue_with_states = {
    'ground_truth_history': [
        {
            'role': 'system',
            'content': 'System message'
        },
        {
            'role': 'user',
            'content': 'Hello, I am interested. <user_cooperation>3</user_cooperation> <user_emotion>2</user_emotion> <user_trust>1</user_trust>'
        }
    ]
}

dialogue_empty = {
    'ground_truth_history': []
}

dialogue_short = {
    'ground_truth_history': [
        {
            'role': 'system',
            'content': 'Only one message'
        }
    ]
}

dialogue_no_tags = {
    'ground_truth_history': [
        {
            'role': 'system',
            'content': 'System message'
        },
        {
            'role': 'user',
            'content': 'Hello without tags'
        }
    ]
}

ARGS = {
    'dialogue': dialogue_with_states
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_score Ran
ran, honoured the contract we drafted
Da1yuqin/SEAD/utils/analyze_chatbot_mistakes.py
code served (permissive licence) · get_code("fb8290dd4e4e685b")
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 re
import torch

def extract_score(content: str, tag: str) -> int:
    """Extract score from XML tag"""
    try:
        match = re.search(f'<{tag}[^>]*>(\\d+)</{tag}>', content, re.IGNORECASE)
        return int(match.group(1)) if match else None
    except:
        return None

# Setup code to construct valid arguments
content = "<score>42</score>"
tag = "score"

ARGS = {
    "content": content,
    "tag": tag
}
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 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
extract_stages_from_dialogue Ran
ran, honoured the contract we drafted
Da1yuqin/SEAD/for_evaluation/get_metrics_from_jsonl.py
code served (permissive licence) · get_code("949edfca00534d9f")
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 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 re

def extract_tag_value(text: str, tag_name: str, cast_type=None):
    """Extract value from XML-style tags"""
    pattern = f'<{tag_name}[^>]*>(.*?)</{tag_name}>'
    match = re.search(pattern, text, re.DOTALL | re.IGNORECASE)
    if match:
        value_str = match.group(1).strip()
        if cast_type:
            try:
                return cast_type(value_str)
            except (ValueError, TypeError):
                return None
        return value_str
    return None

def extract_stages_from_dialogue(dialogue_item):
    """
    Extract all stages that appear in the dialogue
    Returns: List[int]
    """
    ground_truth_history = dialogue_item.get('ground_truth_history', [])
    stages = []
    for msg in ground_truth_history:
        if msg['role'] == 'assistant':
            content = msg['content']
            stage_val = extract_tag_value(content, 'stage', int)
            if stage_val is not None:
                stages.append(stage_val)
    return stages


# Construct valid arguments
dialogue_item = {
    'ground_truth_history': [
        {
            'role': 'user',
            'content': 'Hello, what is the first stage?'
        },
        {
            'role': 'assistant',
            'content': 'The first stage is <stage>1</stage>.'
        },
        {
            'role': 'user',
            'content': 'What about the second stage?'
        },
        {
            'role': 'assistant',
            'content': 'The second stage is <stage>2</stage> of the process.'
        },
        {
            'role': 'assistant',
            'content': 'Then we move to <stage>3</stage>.'
        }
    ]
}

ARGS = {
    'dialogue_item': dialogue_item
}
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 int — 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
extract_tag_value Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/for_evaluation/get_metrics_from_jsonl.py
code served (permissive licence) · get_code("0b7e36b128f91713")
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 re

def extract_tag_value(text: str, tag_name: str, cast_type=None):
    """Extract value from XML-style tags"""
    pattern = f'<{tag_name}[^>]*>(.*?)</{tag_name}>'
    match = re.search(pattern, text, re.DOTALL | re.IGNORECASE)
    if match:
        value_str = match.group(1).strip()
        if cast_type:
            try:
                return cast_type(value_str)
            except (ValueError, TypeError):
                return None
        return value_str
    return None


ARGS = {
    "text": "<greeting>Hello World</greeting>",
    "tag_name": "greeting",
    "cast_type": None
}
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_user_final_status Ran
ran, honoured the contract we drafted
Da1yuqin/SEAD/for_evaluation/get_metrics_from_jsonl.py
code served (permissive licence) · get_code("fa77461d2cd2929c")
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 re

def extract_user_final_status(dialogue_item):
    """Extract user's final status from dialogue"""
    ground_truth_history = dialogue_item.get('ground_truth_history', [])
    if not ground_truth_history:
        return None
    for msg in reversed(ground_truth_history):
        if msg['role'] == 'user':
            content = msg['content']
            match = re.search('<user_status>\\s*(-?\\d+)\\s*</user_status>', content, re.IGNORECASE)
            if match:
                return int(match.group(1))
            break
    return None


# Setup code
dialogue_item = {
    'ground_truth_history': [
        {
            'role': 'assistant',
            'content': 'Hello, how are you?'
        },
        {
            'role': 'user',
            'content': 'I am doing well. <user_status> 42 </user_status>'
        }
    ]
}

ARGS = {
    'dialogue_item': dialogue_item
}
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
fill_user_params Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/SEAD/vllm_service/vllm_server.py
code served (permissive licence) · get_code("c8a9c51bedd5ee37")
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 re
import logging
from typing import Dict, Any

logging.basicConfig(level=logging.INFO)

# Define ARGS with valid arguments for fill_user_params
ARGS = {
    "template": """
    Please create a character with the following traits:
    - Cooperation level: ${cooperation}
    - Emotional state: ${emotion}
    - Trust level: ${trust}
    - Required behaviors: {behaviors}
    
    Additional notes:
    Some text with ${initial_cooperation} and {initial_emotion} placeholders.
    """,
    "user_params": {
        "cooperation": "high",
        "emotion": "optimistic",
        "trust": "moderate",
        "behaviors": ["be helpful", "ask clarifying questions", "maintain professionalism"]
    }
}
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
fill_user_params Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/for_evaluation/Baseline_test_local_models_vllm.py
code served (permissive licence) · get_code("cf0a4a3e934a7c1d")
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 re
from typing import Dict, List, Tuple, Optional, Any

def fill_user_params(template: str, user_params: Dict[str, Any]) -> str:
    """
    Fill user_params placeholders.
    Only removes ${xxx} format placeholders, not {xxx}.
    """
    try:
        filled = template
        field_mapping = {'cooperation': 'initial_cooperation', 'emotion': 'initial_emotion', 'trust': 'initial_trust', 'behaviors': 'specific_behaviors'}
        for key, value in user_params.items():
            if key == 'behaviors' and isinstance(value, list):
                if value:
                    value_str = '\n'.join([f'- {b}' for b in value])
                else:
                    value_str = '- No specific behavior requirements'
            else:
                value_str = str(value)
            placeholders_to_try = []
            placeholders_to_try.append(f'${{{key}}}')
            mapped_key = field_mapping.get(key)
            if mapped_key:
                placeholders_to_try.append(f'${{{mapped_key}}}')
            for placeholder in placeholders_to_try:
                if placeholder in filled:
                    filled = filled.replace(placeholder, value_str)
                    break
        filled = re.sub('\\$\\{[^}]+\\}', '', filled)
        return filled
    except Exception as e:
        print(f'Error filling user_params: {e}')
        return template


ARGS = {
    'template': 'The user has ${cooperation} cooperation level, ${emotion} emotion, and ${trust} trust level. Behaviors: ${behaviors}. Extra placeholder: ${extra}.',
    'user_params': {
        'cooperation': 'high',
        'emotion': 'positive',
        'trust': 'medium',
        'behaviors': ['be friendly', 'listen carefully', 'respond promptly']
    }
}
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
load_analysis Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/generate_profiles_parallel.py
code served (permissive licence) · get_code("d56b142dd29f350c")
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

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

# Create a valid JSON analysis file
analysis_data = {
    "model": "test_model",
    "accuracy": 0.95,
    "metrics": ["precision", "recall", "f1"],
    "results": {"train": 0.92, "val": 0.90}
}

analysis_file = os.path.join(temp_dir, "analysis.json")
with open(analysis_file, 'w', encoding='utf-8') as f:
    json.dump(analysis_data, f)

ARGS = {
    "analysis_file": analysis_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 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_behavior_library Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/create_prompt_data.py
code served (permissive licence) · get_code("afee7667b68ea938")
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
from pathlib import Path
import tempfile
import os

# Create a temporary directory and file for the behavior library
temp_dir = tempfile.mkdtemp()
library_file = Path(temp_dir) / "behavior_library.json"

# Create valid behavior library data
library_data = {
    "User_Persona_Classification": {
        "category_1": ["action_1", "action_2", "action_3"],
        "category_2": ["action_4", "action_5"],
        "category_3": ["action_6", "action_7", "action_8", "action_9"]
    }
}

# Write the library file
with open(library_file, 'w', encoding='utf-8') as f:
    json.dump(library_data, f)

# Define ARGS with the library_file parameter
ARGS = {
    "library_file": library_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 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_behavior_library Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/generate_profiles_parallel.py
code served (permissive licence) · get_code("a528e94a3ea5d552")
How it ran

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

import json
import tempfile
from pathlib import Path

# Create a temporary directory and file for testing
temp_dir = tempfile.mkdtemp()
library_path = Path(temp_dir) / "behavior_library.json"

# Create a sample behavior library JSON file
behavior_data = {
    "User_Persona_Classification": {
        "category_1": ["action_1", "action_2", "action_3"],
        "category_2": ["action_2", "action_4"],
        "category_3": ["action_5"]
    }
}

with open(library_path, 'w', encoding='utf-8') as f:
    json.dump(behavior_data, f)

ARGS = {
    "library_file": str(library_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
load_profiles_from_file Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/create_prompt_data.py
code served (permissive licence) · get_code("475f74719b2c7a8f")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
from pathlib import Path

# Create a temporary file with valid profile data
temp_file = tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False, encoding='utf-8')
temp_file_path = temp_file.name

# Write sample profiles to the temporary file
profiles_data = [
    {'cooperation': 0, 'emotion': 1, 'trust': 2},
    {'cooperation': 1, 'emotion': 0, 'trust': 1},
    {'cooperation': 0, 'emotion': 1, 'trust': 0},
    {'cooperation': 1, 'emotion': 1, 'trust': 2},
    {'cooperation': 0, 'emotion': 0, 'trust': 1},
    {'cooperation': 1, 'emotion': 2, 'trust': 0},
    {'cooperation': 2, 'emotion': 1, 'trust': 2},
    {'cooperation': 0, 'emotion': 2, 'trust': 1},
    {'cooperation': 1, 'emotion': 0, 'trust': 2},
    {'cooperation': 2, 'emotion': 0, 'trust': 0},
]

for profile in profiles_data:
    temp_file.write(json.dumps(profile) + '\n')

temp_file.close()

ARGS = {
    'file_path': temp_file_path,
    'train_samples': 6,
    'test_samples': 2,
}
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 tuple — a tuple output has no array shape of its own; none is 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_prompt_file Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/SEAD/vllm_service/vllm_server.py
code served (permissive licence) · get_code("c7bcd40345c48edd")
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 os
import tempfile

# Create a temporary directory with prompt files for testing
temp_dir = tempfile.mkdtemp()

# Create sample prompt files
with open(os.path.join(temp_dir, 'call_client.txt'), 'w', encoding='utf-8') as f:
    f.write('You are a client support assistant.')

with open(os.path.join(temp_dir, 'call_chatbot.txt'), 'w', encoding='utf-8') as f:
    f.write('You are a chatbot assistant.')

with open(os.path.join(temp_dir, 'custom.txt'), 'w', encoding='utf-8') as f:
    f.write('You are a custom assistant.')

ARGS = {
    'prompt_dir': temp_dir,
    'mode': 'call_client'
}
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
read_prompt Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/create_prompt_data.py
code served (permissive licence) · get_code("18b5ff46b3ab5402")
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.

from pathlib import Path
import tempfile
import os

# Create a temporary directory and file for testing
temp_dir = tempfile.mkdtemp()
temp_file = Path(temp_dir) / "prompt.txt"
temp_file.write_text("This is a sample prompt.\n", encoding='utf-8')

ARGS = {
    'file_path': temp_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 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
restart_vllm_service Ran
ran, violated the contract we drafted
Da1yuqin/SEAD/for_evaluation/Baseline_test_local_models_vllm.py
code served (permissive licence) · get_code("8d1e5517fde1b5f5")
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 os
import tempfile

# Create a temporary directory for the restart script
temp_dir = tempfile.mkdtemp()
restart_script_path = os.path.join(temp_dir, "start_vllm.sh")

# Create a minimal bash script that does nothing (won't actually run)
with open(restart_script_path, 'w') as f:
    f.write("#!/bin/bash\n")
    f.write("# Minimal restart script\n")
    f.write("exit 0\n")

os.chmod(restart_script_path, 0o755)

# Create a minimal model path
model_path = "/tmp/dummy_model"

# Define ARGS dict with all required parameters
ARGS = {
    "restart_script": restart_script_path,
    "model_path": model_path,
    "gpu_ids": "0,1",
    "mode": "call_client",
    "max_wait_time": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
psutil, requests — 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
sample_random_profile Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/utils/generate_profiles_parallel.py
code served (permissive licence) · get_code("52e90609827ad734")
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. 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 random
from typing import Dict, List, Tuple, Set

# Create a behavior library with some sample actions
behavior_lib = {
    'all_actions': [
        'cooperate',
        'defect',
        'share',
        'punish',
        'reward',
        'negotiate',
        'compromise',
        'escalate'
    ]
}

# Create a random number generator with a fixed seed for reproducibility
rng = random.Random(42)

# Create an analysis dict with recommended weights
analysis = {
    'recommended_weights': {
        'cooperation_weights': [0, 0.33, 0.34, 0.33, 0],
        'emotion_weights': [0, 0.5, 0.5, 0],
        'trust_weights': [0, 0.25, 0.25, 0.25, 0.25, 0]
    }
}

# Set up the arguments dictionary
ARGS = {
    'behavior_lib': behavior_lib,
    'rng': rng,
    'analysis': analysis,
    'min_behaviors': 1,
    'max_behaviors': 3
}
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 tuple — a tuple output has no array shape of its own; none is 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
unpad_dataproto Ran
ran; our contract draft was wrong, not the code
Da1yuqin/SEAD/verl/protocol.py
code served (permissive licence) · get_code("25f0ea3f460f6ce1")
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 DataProto. 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)

class DataProto:
    """Stand-in class for DataProto that supports slicing."""
    def __init__(self, data):
        self.data = data
    
    def __getitem__(self, key):
        """Support slicing to return a new DataProto with sliced data."""
        sliced_data = self.data[key]
        return DataProto(sliced_data)
    
    def __repr__(self):
        return f"DataProto({self.data})"

# Create a small DataProto instance with some data
data_tensor = torch.randn(2, 4)
data_proto = DataProto(data_tensor)

ARGS = {
    "data": data_proto,
    "pad_size": 0
}
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 DataProto — a DataProto output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataProto 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_weight_loader Not yet run
unverified (harvested, no recorded run)
Da1yuqin/SEAD/verl/models/weight_loader_registry.py
code served (permissive licence) · get_code("919cf310ad4c189a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 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 18 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
unpad_dataproto 89 other papers
36 dated earlier, 13 the same month, 37 later, 3 undated
2409.19256 (2024-09)
HybridFlow: A Flexible and Efficient RLHF Framework
2410.21236 (2024-10)
Flaming-hot Initiation with Regular Execution Sampling for Large Language Models
2025.findings-emnlp.1015 (2025 (year only))
MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning
and 86 more not named here
get_weight_loader 20 other papers
15 dated earlier, 2 the same month, 2 later, 1 undated
2410.21236 (2024-10)
Flaming-hot Initiation with Regular Execution Sampling for Large Language Models
2025.findings-emnlp.1015 (2025 (year only))
MT-R1-Zero: Advancing LLM-based Machine Translation via R1-Zero-like Reinforcement Learning
2502.14768 (2025-02)
Logic-RL: Unleashing LLM Reasoning with Rule-Based Reinforcement Learning
and 17 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

Large Language Models have demonstrated remarkable capabilities in open-domain dialogues. However, current methods exhibit suboptimal performance in service dialogues, as they rely on noisy, low-quality human conversation data. This limitation arises from data scarcity and the difficulty of simulating authentic, goal-oriented user behaviors. To address these issues, we propose SEAD (Self-Evolving Agent for Service Dialogue), a framework that enables agents to learn effective strategies without large-scale human annotations. SEAD decouples user modeling into two components: a Profile Controller that generates diverse user states to manage training curriculum, and a User Role-play Model that focuses on realistic role-playing. This design ensures the environment provides adaptive training scenarios rather than acting as an unfair adversary. Experiments demonstrate that SEAD significantly outperforms Open-source Foundation Models and Closed-source Commercial Models, improving task completion rate by 17.6% and dialogue efficiency by 11.1%. Code is available at: https://github.com/Da1yuqin/SEAD.

For agents

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

get_harvested_code_for_paper("2602.03548")
get_code_for_paper("2602.03548")
have("2602.03548")

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

curl https://syntology.ai/api/ran/2602.03548.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.03548.svg)](https://syntology.ai/paper/2602.03548)

Connect an agent — have() is free.