SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.04536 · ICLR · 2024

Evaluating Language Model Agency through Negotiations

Martin Josifoski, Antoine Bosselut, Tim Davidson, Veniamin Veselovsky, Maxime Peyrard, Michal Kosinski, Robert West

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2401.04536.json · badge: /api/ran/2401.04536.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
epfl-dlab/LAMEN canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
check_if_model_instantiated Ran
ran; our contract draft was wrong, not the code
epfl-dlab/LAMEN/src/utils.py
code served (permissive licence) · get_code("ce662efe0ac1fbb1")
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 AgentConfig. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple model class
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create instantiated models dictionary with a cached model
instantiated_models = {
    "gpt2": SimpleModel(),
    "bert": SimpleModel(),
}

# Create a simple agent config class
class AgentConfig:
    def __init__(self, model_name):
        self.model_name = model_name
        self.model = None

# Create agent config that references a model in instantiated_models
agent_config = AgentConfig("gpt2")

ARGS = {
    "instantiated_models": instantiated_models,
    "agent_config": agent_config,
}
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 AgentConfig — a AgentConfig output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a AgentConfig 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_dictionary Ran
ran; our contract draft was wrong, not the code
epfl-dlab/LAMEN/src/utils.py
code served (permissive licence) · get_code("3bb422770073a45e")
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 re

def extract_dictionary(x):
    if isinstance(x, str):
        regex = '{.*?}'
        match = re.search(regex, x, re.MULTILINE | re.DOTALL)
        if match:
            try:
                json_str = match.group()
                json_str = json_str.replace("'", '"')
                dict_ = json.loads(json_str)
                return dict_
            except Exception as e:
                print(f'unable to extract dictionary - {e}')
                return None
        else:
            return None
    else:
        return None

ARGS = {
    'x': 'some text before {"key": "value", "number": 42} some text after'
}
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
fuzzy_index_matching Ran
ran, honoured the contract we drafted
epfl-dlab/LAMEN/src/utils.py
code served (permissive licence) · get_code("d6877bfa047c0336")
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 int64, 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 numpy as np

# Construct valid arguments for fuzzy_index_matching
lst = ['item1', 'item2_42', 'item3_100', 'item4']
value = 'item2_45'

ARGS = {
    'lst': lst,
    'value': value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int64, 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
load_game Ran
ran; our contract draft was wrong, not the code
epfl-dlab/LAMEN/src/games.py
code served (permissive licence) · get_code("1e05e137fdae446b")
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 MockGame. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os
import yaml

# Create temporary YAML files for testing
temp_dir = tempfile.mkdtemp()

# Create a game file
game_data = {
    'name': 'TestGame',
    'description': 'A test game'
}
game_path = os.path.join(temp_dir, 'game.yaml')
with open(game_path, 'w') as f:
    yaml.dump(game_data, f)

# Create a general rules file
general_rules_data = {
    'rules_prompt': 'These are the general rules.',
    'rules': ['Rule 1', 'Rule 2']
}
general_rules_path = os.path.join(temp_dir, 'general_rules.yaml')
with open(general_rules_path, 'w') as f:
    yaml.dump(general_rules_data, f)

# Create a mock game object class that supports the operations in load_game
class MockGame(dict):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.rules_prompt = None
        self.rules = None
    
    def add_general_rules(self):
        """Mock method that would combine general rules with game rules"""
        pass

# Patch yaml.safe_load to return our mock game object
original_safe_load = yaml.safe_load

def patched_safe_load(f):
    content = original_safe_load(f)
    if content is not None:
        # If it's the game file, wrap it in MockGame
        if isinstance(content, dict) and 'name' in content:
            game_obj = MockGame(content)
            return game_obj
        # If it's the rules file, return as-is
        return content
    return content

yaml.safe_load = patched_safe_load

ARGS = {
    'game_path': game_path,
    'general_rules': general_rules_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MockGame — a MockGame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MockGame output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 4 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We introduce an approach to evaluate language model (LM) agency using negotiation games. This approach better reflects real-world use cases and addresses some of the shortcomings of alternative LM benchmarks. Negotiation games enable us to study multi-turn, and cross-model interactions, modulate complexity, and side-step accidental evaluation data leakage. We use our approach to test six widely used and publicly accessible LMs, evaluating performance and alignment in both self-play and cross-play settings. Noteworthy findings include: (i) only closed-source models tested here were able to complete these tasks; (ii) cooperative bargaining games proved to be most challenging to the models; and (iii) even the most powerful models sometimes "lose" to weaker opponents. 1

For agents

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

get_harvested_code_for_paper("2401.04536")
get_code_for_paper("2401.04536")
have("2401.04536")

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

curl https://syntology.ai/api/ran/2401.04536.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.04536.svg)](https://syntology.ai/paper/2401.04536)

Connect an agent — have() is free.