SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.18566 · 2026

Benchmarking System Dynamics AI Assistants: Cloud Versus Local Large Language Models on CLD Extraction and Discussion

Terry Leitch

arXiv · PDF · Open in the Atlas

Code that ran

5 of 6 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 2 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/2604.18566.json · badge: /api/ran/2604.18566.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
UB-IAD/sd-ai canonical (classifier label)
classifier basis: census
5 ran · 1 no recorded run
FunctionStatusWhere it lives
build_executable Ran
ran, violated the contract we drafted
UB-IAD/sd-ai/third-party/time-series-behavior-analysis/build.py
code served (permissive licence) · get_code("b4a801090615aafb")
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 sys
import os
from pathlib import Path
import tempfile

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()
root_path = Path(temp_dir)
dist_path = root_path / 'dist'
build_path = root_path / 'build'
init_module_path = root_path / 'module' / '__init__.py'

# Create necessary directories
dist_path.mkdir(exist_ok=True)
build_path.mkdir(exist_ok=True)
init_module_path.parent.mkdir(exist_ok=True)

# Create dummy files
main_script_path = root_path / 'main.py'
main_script_path.write_text('print("test")')
init_module_path.write_text('# init module')

# Create dist_examples directory
dist_examples_path = dist_path / 'examples'
dist_examples_path.mkdir(exist_ok=True)

# Build the paths dictionary
paths = {
    'root': root_path,
    'init_module': init_module_path,
    'main_script': main_script_path,
    'dist': dist_path,
    'build': build_path,
    'dist_examples': dist_examples_path,
}

ARGS = {
    'paths': paths,
    'onefile': True,
}
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
create_spec_file Ran
ran; our contract draft was wrong, not the code
UB-IAD/sd-ai/third-party/time-series-behavior-analysis/build.py
code served (permissive licence) · get_code("7edc44352b84c72f")
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 PosixPath. 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 for testing
temp_dir = tempfile.mkdtemp()
root_path = Path(temp_dir)

# Create necessary dummy files/directories
main_script_path = root_path / "main.py"
main_script_path.write_text("# Main script placeholder\n")

init_module_path = root_path / "__init__.py"
init_module_path.write_text("# Init module placeholder\n")

# Construct the paths dictionary
paths = {
    'main_script': str(main_script_path),
    'root': root_path,
    'init_module': str(init_module_path),
}

# Define ARGS
ARGS = {
    'paths': paths,
    'onefile': True,
}
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 PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath 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_csv_data Ran
ran, honoured the contract we drafted
UB-IAD/sd-ai/third-party/time-series-behavior-analysis/classify_behavior.py
code served (permissive licence) · get_code("4add2d583e034913")
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 ndarray, ndim 1, dtype family float, shape (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 csv
import tempfile
import os
import numpy as np

# Create a temporary CSV file with sample time series data
temp_dir = tempfile.mkdtemp()
filepath = os.path.join(temp_dir, "timeseries.csv")

with open(filepath, 'w', newline='') as f:
    writer = csv.writer(f)
    writer.writerow(['timestamp', 'value', 'other'])
    writer.writerow(['2024-01-01', '10.5', 'a'])
    writer.writerow(['2024-01-02', '20.3', 'b'])
    writer.writerow(['2024-01-03', '15.7', 'c'])
    writer.writerow(['2024-01-04', '25.1', 'd'])

ARGS = {
    'filepath': filepath,
    'column': 'value',
    'skip_header': True
}
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 ndarray, ndim 1, dtype family float, shape (4,) — shape (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_json_data Ran
ran, honoured the contract we drafted
UB-IAD/sd-ai/third-party/time-series-behavior-analysis/classify_behavior.py
code served (permissive licence) · get_code("48a884dfccfdcbd2")
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 ndarray, ndim 1, dtype family float, shape (6,) 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 json
import tempfile
import numpy as np
import os

# Create a temporary JSON file with time series data
temp_dir = tempfile.mkdtemp()
filepath = os.path.join(temp_dir, "timeseries.json")

# Write test data to JSON file
test_data = {
    "values": [1.0, 2.5, 3.7, 4.2, 5.1, 6.3]
}

with open(filepath, 'w') as f:
    json.dump(test_data, f)

ARGS = {
    "filepath": filepath,
    "key": "values"
}
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 ndarray, ndim 1, dtype family float, shape (6,) — shape (6,) 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
parse_relationships Ran
ran; our contract draft was wrong, not the code
UB-IAD/sd-ai/third-party/causal-decoder/inference.py
code served (permissive licence) · get_code("23d4fd5b47fa5084")
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 parse_relationships(text: str):
    rels = []
    for part in re.split('[,\\n;]+', text):
        part = part.strip()
        if not part:
            continue
        m = re.match('(.+?)->\\s*([+-])\\s*(.+)', part)
        if not m:
            continue
        frm, pol, to = (m.group(1).strip(), m.group(2), m.group(3).strip())
        rels.append({'from': frm, 'to': to, 'polarity': pol})
    return rels


ARGS = {
    'text': 'protein A -> + protein B\nprotein B -> - protein C; gene X -> + gene Y, factor Z -> - outcome W'
}
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
generate_targets Not yet run
unverified (harvested, no recorded run)
UB-IAD/sd-ai/third-party/causal-decoder/inference.py
code served (permissive licence) · get_code("345a2d274a9a7f11")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 6 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

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

Repositories linked to this paper

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

Abstract

We present a systematic evaluation of large language model families-spanning both proprietary cloud APIs and locally-hosted open-source models-on two purpose-built benchmarks for System Dynamics AI assistance: the CLD Leaderboard (53 tests, structured causal loop diagram extraction) and the Discussion Leaderboard (interactive model discussion, feedback explanation, and model building coaching). On CLD extraction, cloud models achieve 77-89% overall pass rates; the best local model reaches 77% (Kimi K2.5 GGUF Q3, zero-shot engine), matching mid-tier cloud performance. On Discussion, the best local models achieve 50-100% on model building steps and 47-75% on feedback explanation, but only 0-50% on error fixing-a category dominated by long-context prompts that expose memory limits in local deployments. A central contribution of this paper is a systematic analysis of model type effects on performance: we compare reasoning vs. instruction-tuned architectures, GGUF (llama.cpp) vs. MLX (mlx lm) backends, and quantization levels (Q3 / Q4 K M / MLX-3bit / MLX-4bit / MLX-6bit) across the same underlying model families. We find that backend choice has larger practical impact than quantization level: mlx lm does not enforce JSON schema constraints, requiring explicit prompt-level JSON instructions, while llama.cpp grammar-constrained sampling handles JSON reliably but causes indefinite generation on long-context prompts for dense models. We document the full parameter sweep (t, p, k) for all local models, cleaned timing data (stuck requests excluded), and a practitioner guide for running 671B-123B parameter models on Apple Silicon.

For agents

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

get_harvested_code_for_paper("2604.18566")
get_code_for_paper("2604.18566")
have("2604.18566")

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

curl https://syntology.ai/api/ran/2604.18566.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.18566.svg)](https://syntology.ai/paper/2604.18566)

Connect an agent — have() is free.