SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.10191 · ICLR · 2024

Divide and not forget: Ensemble of selectively trained experts in Continual Learning

Tomasz Trzci Ński, Bartłomiej Twardowski, Bartosz Zieli Ński, Grzegorz Rypeść, Sebastian Cygert, Valeriya Khan

arXiv · PDF · Open in the Atlas

Code that ran

9 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 4 failures of our instrument; 1 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/2401.10191.json · badge: /api/ran/2401.10191.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
grypesc/seed canonical (classifier label)
classifier basis: classifier_2025
9 ran · 1 no recorded run
FunctionStatusWhere it lives
ClassDirectoryDataset Ran
ran on a synthesized input · constructed an object; no result was computed
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("911cc14e7fa1b63a")
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 an instance of its own class, ClassDirectoryDataset: the run constructed the object and did not call it, so no result was computed. 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 PIL import Image
import tempfile
import os

torch.manual_seed(0)

# Create temporary directory with dummy image files
temp_dir = tempfile.mkdtemp()

# Create 2 small dummy PNG images
dummy_images = []
for i in range(2):
    img = Image.new('RGB', (4, 3), color=(73, 109, 137))
    img_path = os.path.join(temp_dir, f'image_{i}.png')
    img.save(img_path)
    dummy_images.append(img_path)

# Define a minimal transforms object that has __call__ method
class MinimalTransforms:
    def __call__(self, image):
        # Return the image as-is (PIL Image)
        return image

transforms = MinimalTransforms()

ARGS = {
    'images': dummy_images,
    'transforms': transforms
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ClassDirectoryDataset — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ClassDirectoryDataset output the recorded ndim 4 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
ClassMemoryDataset Ran
ran on a synthesized input · constructed an object; no result was computed
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("e49ad10d768d006a")
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 an instance of its own class, ClassMemoryDataset: the run constructed the object and did not call it, so no result was computed. 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 torchvision import transforms
from PIL import Image

torch.manual_seed(0)

# Create synthetic image data as numpy arrays
# Images need to be in format that PIL.Image.fromarray() can handle
# Shape: (num_images, height, width, channels) for RGB or (num_images, height, width) for grayscale
images = np.random.randint(0, 256, size=(2, 4, 3, 3), dtype=np.uint8)  # 2 images, 4x3 pixels, RGB

# Create a simple transforms pipeline
transforms_pipeline = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

ARGS = {
    "images": images,
    "transforms": transforms_pipeline
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ClassMemoryDataset — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ClassMemoryDataset output the recorded ndim 4 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
ExemplarsDataset Ran
ran on a synthesized input
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("7dbb544669ebb01d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned NoneType. 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 torchvision import transforms
from torch.utils.data import DataLoader, TensorDataset
import sys
from unittest.mock import Mock

# Mock the exemplars_selection module since we can't import it
mock_selector_module = Mock()

class RandomExemplarsSelector:
    def __init__(self, dataset):
        self.dataset = dataset
    
    def __call__(self, model, trn_loader, selection_transform):
        # Return empty lists for this test
        return [], []

mock_selector_module.RandomExemplarsSelector = RandomExemplarsSelector
sys.modules['datasets.exemplars_selection'] = mock_selector_module

# Create a simple transform
transform = transforms.Compose([
    transforms.ToTensor(),
])

# Create class indices
class_indices = [0, 1, 2]

# INIT_ARGS for ExemplarsDataset.__init__
INIT_ARGS = {
    'transform': transform,
    'class_indices': class_indices,
    'num_exemplars': 0,
    'num_exemplars_per_class': 10,
    'exemplar_selection': 'random'
}

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

model = SimpleModel()

# Create a simple training loader with numpy arrays as images
train_images = np.random.randint(0, 255, size=(6, 8, 8, 3), dtype=np.uint8)
train_labels = torch.tensor([0, 0, 1, 1, 2, 2])
train_dataset = TensorDataset(torch.from_numpy(train_images), train_labels)
trn_loader = DataLoader(train_dataset, batch_size=2)

# Selection transform
selection_transform = transforms.Compose([
    transforms.ToTensor(),
])

# ARGS for collect_exemplars method
METHOD = 'collect_exemplars'
ARGS = {
    'model': model,
    'trn_loader': trn_loader,
    'selection_transform': selection_transform
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type NoneType — a NoneType output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a NoneType 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GaussianMixture Ran
ran on a synthesized input · constructed an object; no result was computed
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("ee35cf16bfbb70b6")
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 an instance of its own class, GaussianMixture: the run constructed the object and did not call it, so no result was computed. 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)

ARGS = {
    'n_components': 3,
    'n_features': 4,
    'covariance_type': 'full',
    'eps': 1e-6,
    'init_params': 'kmeans',
    'mu_init': None,
    'var_init': None,
}
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 GaussianMixture — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GaussianMixture 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
Inc_Learning_Appr Ran
ran on a synthesized input · constructed an object; no result was computed
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("6af6e3f22725346c")
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 an instance of its own class, Inc_Learning_Appr: the run constructed the object and did not call it, so no result was computed. 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
from argparse import ArgumentParser
from datetime import datetime
import tempfile
import os

torch.manual_seed(0)

class ExperimentLogger:
    """Mock experiment logger"""
    def __init__(self, log_path, exp_name, begin_time=None):
        self.log_path = log_path
        self.exp_name = exp_name
        self.exp_path = os.path.join(log_path, exp_name)
        if begin_time is None:
            self.begin_time = datetime.now()
        else:
            self.begin_time = begin_time

    def log_scalar(self, task, iter, name, value, group=None, curtime=None):
        pass

    def log_args(self, args):
        pass

    def log_result(self, array, name, step):
        pass

    def log_figure(self, name, iter, figure, curtime=None):
        pass

    def save_model(self, state_dict, task):
        pass


class ExemplarsDataset:
    """Mock exemplars dataset"""
    def __init__(self, transform=None, class_indices=None, num_exemplars=0, num_exemplars_per_class=0, exemplar_selection='random'):
        self.transform = transform
        self.class_indices = class_indices
        self.max_num_exemplars = num_exemplars
        self.max_num_exemplars_per_class = num_exemplars_per_class
        self.exemplar_selection = exemplar_selection
        self.images = []
        self.labels = []

    def _is_active(self):
        return self.max_num_exemplars_per_class > 0 or self.max_num_exemplars > 0

    def collect_exemplars(self, model, trn_loader, selection_transform):
        pass


class SimpleModel(nn.Module):
    """Minimal model for testing"""
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(4, 2)
        self.bn = nn.BatchNorm1d(4)

    def forward(self, x):
        x = self.bn(x)
        x = self.fc(x)
        return x


# Create a temporary directory for logging
temp_log_dir = tempfile.mkdtemp()

# Build arguments dict for Inc_Learning_Appr.__init__
ARGS = {
    'model': SimpleModel(),
    'device': torch.device('cpu'),
    'nepochs': 10,
    'lr': 0.05,
    'lr_min': 0.0001,
    'lr_factor': 3,
    'lr_patience': 5,
    'clipgrad': 10000,
    'momentum': 0.9,
    'wd': 0.0005,
    'multi_softmax': False,
    'wu_nepochs': 0,
    'wu_lr_factor': 1,
    'fix_bn': False,
    'eval_on_train': False,
    'logger': ExperimentLogger(temp_log_dir, 'test_experiment'),
    'exemplars_dataset': ExemplarsDataset(transform=None, class_indices=None, num_exemplars=0, num_exemplars_per_class=0, exemplar_selection='random'),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Inc_Learning_Appr — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Inc_Learning_Appr 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
calculate_matmul Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("4451964134c9d216")
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 3, dtype family float, shape (2, 3, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct mat_a with shape (n, k, 1, d)
# n=2 (batch), k=3 (sequence), 1 (fixed), d=4 (feature dimension)
mat_a = torch.randn(2, 3, 1, 4)

# Construct mat_b with shape (n, k, d, 1)
# n=2 (batch), k=3 (sequence), d=4 (feature dimension), 1 (fixed)
mat_b = torch.randn(2, 3, 4, 1)

ARGS = {
    "mat_a": mat_a,
    "mat_b": mat_b,
}
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 3, dtype family float, shape (2, 3, 1) — shape (2, 3, 1) 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
calculate_matmul_n_times Ran
ran; our contract draft was wrong, not the code
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("fa6374ff7f7939f4")
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 4, dtype family float, shape (2, 3, 1, 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

torch.manual_seed(0)

# Parameters for calculate_matmul_n_times
n_components = 3
k = n_components

# mat_a: shape (n, k, 1, d)
# n=2 (batch), k=3 (components), 1 (singleton), d=4 (feature dim)
mat_a = torch.randn(2, 3, 1, 4)

# mat_b: shape (1, k, d, d)
# 1 (singleton), k=3 (components), d=4, d=4
mat_b = torch.randn(1, 3, 4, 4)

ARGS = {
    "n_components": n_components,
    "mat_a": mat_a,
    "mat_b": mat_b,
}
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 4, dtype family float, shape (2, 3, 1, 4) — shape (2, 3, 1, 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
softmax_temperature Ran
ran; our contract draft was wrong, not the code
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("c1cbed26c18dd1f2")
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 Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a small tensor for x parameter
x = torch.randn(2, 4)

# dim parameter - softmax dimension
dim = 1

# tau parameter - temperature
tau = 0.5

# Construct ARGS dict with all parameters
ARGS = {
    'x': x,
    'dim': dim,
    'tau': tau
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
softmax_temperature Ran
ran; our contract draft was wrong, not the code
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("56730f951de5fd44")
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 (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small input tensor (batch_size=2, feature_dim=4)
x = torch.randn(2, 4)

# dim parameter specifies which dimension to apply softmax over
dim = 1

# tau is the temperature parameter (default is 1.0, but we can set it explicitly)
tau = 0.5

ARGS = {
    "x": x,
    "dim": dim,
    "tau": tau
}
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 (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Appr Not yet run
unverified (harvested, no recorded run)
grypesc/seed/src/approach/seed.py
code served (permissive licence) · get_code("9a8759c93a1ea172")
How it ran

no invocation recorded.

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 9 distinct functions with an identity, 5 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 5 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.

FunctionAlso present inEarliest-datedOthers named
ExemplarsDataset
brick of this paper's assembly: helper
4 other papers
3 dated earlier, 1 later · 2 only as an assembly brick
1606.09282 (2016-06)
Learning without Forgetting
1611.07725 (2016-11)
iCaRL: Incremental Classifier and Representation Learning
1612.00796 (2016-12)
Overcoming catastrophic forgetting in neural networks
and 1 more not named here
calculate_matmul
brick of this paper's assembly: helper
2 other papers
2 later
2407.12200 (2024-07)
This Probably Looks Exactly Like That: An Invertible Prototypical Network
2606.25347 (2026-06)
Geometry-Anchored Transport Framework for Exemplar-Free Class-Incremental Learning
ClassDirectoryDataset
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2606.05675 (2026-06)
Two-Way Is Better Than One: Bidirectional Alignment with Cycle Consistency for Exemplar-Free Class-Incremental Learning
ClassMemoryDataset
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2606.05675 (2026-06)
Two-Way Is Better Than One: Bidirectional Alignment with Cycle Consistency for Exemplar-Free Class-Incremental Learning
calculate_matmul_n_times
brick of this paper's assembly: helper
1 other paper
1 later
2606.25347 (2026-06)
Geometry-Anchored Transport Framework for Exemplar-Free Class-Incremental Learning

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

Class-incremental learning is becoming more popular as it helps models widen their applicability while not forgetting what they already know. A trend in this area is to use a mixture-of-expert technique, where different models work together to solve the task. However, the experts are usually trained all at once using whole task data, which makes them all prone to forgetting and increasing computational burden. To address this limitation, we introduce a novel approach named SEED. SEED selects only one, the most optimal expert for a considered task, and uses data from this task to fine-tune only this expert. For this purpose, each expert represents each class with a Gaussian distribution, and the optimal expert is selected based on the similarity of those distributions. Consequently, SEED increases diversity and heterogeneity within the experts while maintaining the high stability of this ensemble method. The extensive experiments demonstrate that SEED achieves state-of-the-art performance in exemplar-free settings across various scenarios, showing the potential of expert diversification through data in continual learning.

For agents

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

get_harvested_code_for_paper("2401.10191")
get_code_for_paper("2401.10191")
have("2401.10191")

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

curl https://syntology.ai/api/ran/2401.10191.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.10191.svg)](https://syntology.ai/paper/2401.10191)

Connect an agent — have() is free.