SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2106.02225 · 2021

Materials Representation and Transfer Learning for Multi-Property Prediction

arXiv · PDF · Open in the Atlas

Code that ran

9 of 9 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; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 1 failure 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/2106.02225.json · badge: /api/ran/2106.02225.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
CompRhys/roost canonical (classifier label)
classifier basis: census
8 ran · 0 no recorded run
gomes-lab/H-CLMP canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
RobustL1Loss Ran
ran on a synthesized input
CompRhys/roost/roost/core.py
code served (permissive licence) · get_code("d6aab70127704b1d")
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 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

torch.manual_seed(0)

# Create small tensors for the function
batch_size = 2
dim = 4

output = torch.randn(batch_size, dim)
log_std = torch.randn(batch_size, dim)
target = torch.randn(batch_size, dim)

ARGS = {
    "output": output,
    "log_std": log_std,
    "target": target,
}
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 Tensor, 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
RobustL2Loss Ran
ran on a synthesized input
CompRhys/roost/roost/core.py
code served (permissive licence) · get_code("3b454f31a71ab600")
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 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

torch.manual_seed(0)

# Create small tensors for the function parameters
# output: model predictions, shape (batch_size, output_dim)
output = torch.randn(2, 4)

# log_std: log of standard deviation estimates, same shape as output
log_std = torch.randn(2, 4)

# target: ground truth values, same shape as output
target = torch.randn(2, 4)

ARGS = {
    "output": output,
    "log_std": log_std,
    "target": target,
}
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 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
collate_batch Ran
ran on a synthesized input
CompRhys/roost/roost/cgcnn/data.py
code served (permissive licence) · get_code("dd79bba4183b8a50")
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

torch.manual_seed(0)

# Build a minimal dataset_list with the expected structure
# Each item is (inputs, target, comp, cif_id)
# where inputs = (atom_fea, nbr_dist, self_idx, nbr_idx)

# Crystal 1
atom_fea_1 = torch.randn(3, 4)  # 3 atoms, 4 features
nbr_dist_1 = torch.randn(3, 5, 6)  # 3 atoms, 5 neighbors, 6 distance features
self_idx_1 = torch.LongTensor([0, 1, 2])  # self indices
nbr_idx_1 = torch.LongTensor([[0, 1, 2, 1, 0], [1, 0, 2, 0, 1], [2, 1, 0, 1, 2]])  # neighbor indices (3 atoms, 5 neighbors each)
target_1 = (torch.tensor([1.5]),)  # tuple with single target
comp_1 = "NaCl"
cif_id_1 = "cif_001"

inputs_1 = (atom_fea_1, nbr_dist_1, self_idx_1, nbr_idx_1)
item_1 = (inputs_1, target_1, comp_1, cif_id_1)

# Crystal 2
atom_fea_2 = torch.randn(2, 4)  # 2 atoms, 4 features
nbr_dist_2 = torch.randn(2, 5, 6)  # 2 atoms, 5 neighbors, 6 distance features
self_idx_2 = torch.LongTensor([0, 1])  # self indices
nbr_idx_2 = torch.LongTensor([[0, 1, 0, 1, 0], [1, 0, 1, 0, 1]])  # neighbor indices (2 atoms, 5 neighbors each)
target_2 = (torch.tensor([2.3]),)  # tuple with single target
comp_2 = "MgO"
cif_id_2 = "cif_002"

inputs_2 = (atom_fea_2, nbr_dist_2, self_idx_2, nbr_idx_2)
item_2 = (inputs_2, target_2, comp_2, cif_id_2)

dataset_list = [item_1, item_2]

ARGS = {
    "dataset_list": dataset_list
}
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 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
collate_batch Ran
ran on a synthesized input
CompRhys/roost/roost/pretrain/dist_data.py
code served (permissive licence) · get_code("560cdc7d3a9d7b34")
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

torch.manual_seed(0)

# Construct a dataset_list with 2 data points
# Each data point is a tuple: (inputs, target, comp, cif_id)
# where inputs = (atom_fea, nbr_dist, self_idx, nbr_idx, mask_idx)

# Data point 1
atom_fea_1 = torch.randn(3, 4)  # 3 atoms, 4 features each
nbr_dist_1 = torch.randn(3, 5, 6)  # 3 atoms, 5 neighbors, 6 distance features
self_idx_1 = torch.LongTensor([0, 1, 2])  # 3 bonds
nbr_idx_1 = torch.LongTensor([1, 2, 0, 1, 2])  # neighbor indices (5 total)
mask_idx_1 = torch.LongTensor([0, 1, 2, 3, 4])  # 5 bonds
inputs_1 = (atom_fea_1, nbr_dist_1, self_idx_1, nbr_idx_1, mask_idx_1)
target_1 = (torch.tensor([1.5]),)  # tuple of tensors for target
comp_1 = "NaCl"
cif_id_1 = "cif_001"

# Data point 2
atom_fea_2 = torch.randn(2, 4)  # 2 atoms, 4 features each
nbr_dist_2 = torch.randn(2, 5, 6)  # 2 atoms, 5 neighbors, 6 distance features
self_idx_2 = torch.LongTensor([0, 1])  # 2 bonds
nbr_idx_2 = torch.LongTensor([1, 0, 1, 0, 1])  # neighbor indices (5 total)
mask_idx_2 = torch.LongTensor([0, 1, 2, 3, 4])  # 5 bonds
inputs_2 = (atom_fea_2, nbr_dist_2, self_idx_2, nbr_idx_2, mask_idx_2)
target_2 = (torch.tensor([2.3]),)  # tuple of tensors for target
comp_2 = "MgO"
cif_id_2 = "cif_002"

dataset_list = [
    (inputs_1, target_1, comp_1, cif_id_1),
    (inputs_2, target_2, comp_2, cif_id_2),
]

ARGS = {
    "dataset_list": dataset_list,
}
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 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
compute_loss Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
gomes-lab/H-CLMP/HCLMP/HCLMP.py
code served (permissive licence) · get_code("3bd2c9e276e69964")
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, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Determine dimensions from function usage:
# - input_label: used with softmax(dim=1) and compared to outputs
# - output dict contains: 'label_out', 'fe_mu', 'fe_logvar', 'label_emb', 
#                         'feat_out', 'fx_mu', 'fx_logvar', 'feat_emb'
# - fe_out, fx_out: used with softmax(dim=1), shape should match input_label
# - fe_mu, fe_logvar, fx_mu, fx_logvar: used in KL computation with sum(dim=1)
#   so they need a second dimension to sum over

batch_size = 2
num_classes = 4
latent_dim = 3

# Create input_label: batch_size x num_classes
input_label = torch.randn(batch_size, num_classes)

# Create output dict with all required keys
# fe_out, fx_out: should be logits/scores matching num_classes for KL divergence
# fe_mu, fe_logvar, fx_mu, fx_logvar: latent space parameters (batch_size x latent_dim)
# label_emb, feat_emb: embeddings (not directly used in loss computation)

output = {
    'label_out': torch.randn(batch_size, num_classes),
    'fe_mu': torch.randn(batch_size, latent_dim),
    'fe_logvar': torch.randn(batch_size, latent_dim),
    'label_emb': torch.randn(batch_size, 5),
    'feat_out': torch.randn(batch_size, num_classes),
    'fx_mu': torch.randn(batch_size, latent_dim),
    'fx_logvar': torch.randn(batch_size, latent_dim),
    'feat_emb': torch.randn(batch_size, 5),
}

ARGS = {
    'input_label': input_label,
    'output': output,
}
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 tuple, ndim 1, dtype family float — a tuple 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
get_cgcnn_input Ran
ran on a synthesized input
CompRhys/roost/roost/cgcnn/utils.py
code served (permissive licence) · get_code("b04460e73d418552")
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 numpy as np

class MockSpecie:
    def __init__(self, symbol):
        self.symbol = symbol

class MockAtom:
    def __init__(self, symbol, coords):
        self.specie = MockSpecie(symbol)
        self.coords = coords

class MockLattice:
    def __init__(self, matrix):
        self.matrix = matrix

class MockStructure:
    def __init__(self, atoms, lattice_matrix, frac_coords):
        self.atoms = atoms
        self.lattice = MockLattice(lattice_matrix)
        self.frac_coords = frac_coords
    
    def __iter__(self):
        return iter(self.atoms)

# Create a mock structure with 2 atoms
lattice_matrix = np.array([
    [5.0, 0.0, 0.0],
    [0.0, 5.0, 0.0],
    [0.0, 0.0, 5.0]
])

frac_coords = np.array([
    [0.0, 0.0, 0.0],
    [0.5, 0.5, 0.5]
])

atoms = [
    MockAtom('Fe', [0.0, 0.0, 0.0]),
    MockAtom('O', [0.5, 0.5, 0.5])
]

struct = MockStructure(atoms, lattice_matrix, frac_coords)

ARGS = {
    'struct': struct
}
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 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
init_model Ran
ran on a synthesized input
CompRhys/roost/roost/utils.py
code served (permissive licence) · get_code("91e255b034e75fde")
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, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Build a minimal model class that satisfies the function's usage
class DummyModel(nn.Module):
    def __init__(self, device=None, robust=False, n_targets=3, task_dict=None, **kwargs):
        super().__init__()
        self.robust = robust
        self.n_targets = n_targets
        self.task_dict = task_dict or {}
        self.model_params = {
            'robust': robust,
            'n_targets': n_targets,
            'task_dict': task_dict or {}
        }
        self.device = device
        
        # Add some learnable parameters
        self.linear = nn.Linear(4, 8)
        self.output = nn.Linear(8, n_targets)
        
        # Attributes that the function may access
        self.epoch = 0
        self.best_val_score = 0.0
    
    @property
    def num_params(self):
        return sum(p.numel() for p in self.parameters() if p.requires_grad)

ARGS = {
    'model_class': DummyModel,
    'model_name': 'test_model',
    'model_params': {
        'robust': False,
        'n_targets': 3,
        'task_dict': {'task1': 'regression'}
    },
    'run_id': 'run_001',
    'optim': 'Adam',
    'learning_rate': 0.001,
    'weight_decay': 1e-5,
    'momentum': 0.9,
    'device': 'cpu',
    'milestones': [10, 20],
    'gamma': 0.3,
    'resume': None,
    'fine_tune': None,
    'transfer': None
}
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 tuple, ndim 1, dtype family float — a tuple 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
parse_cgcnn Ran
ran on a synthesized input
CompRhys/roost/roost/cgcnn/data.py
code served (permissive licence) · get_code("ea6012102ca4dc65")
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 ast
import numpy as np

# The function parse_cgcnn takes two string parameters: cell and sites
# cell should be a string representation of a list/array that can be parsed by ast.literal_eval
# sites should be a string representation of a list of strings in format "element @ x y z"

ARGS = {
    "cell": "[[1.0, 0.0, 0.0], [0.0, 2.0, 0.0], [0.0, 0.0, 3.0]]",
    "sites": "['Si @ 0.0 0.0 0.0', 'O @ 0.5 0.5 0.5']"
}
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 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
sampled_softmax Ran
ran on a synthesized input
CompRhys/roost/roost/core.py
code served (permissive licence) · get_code("1a63e7aeda822d81")
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
from torch.nn.functional import softmax

torch.manual_seed(0)

# Construct valid arguments for sampled_softmax
# The function expects:
# - pre_logits: tensor that will be repeated and added to noise
# - log_std: tensor that will be exponentiated and repeated
# - samples: int for number of samples (default 10)

batch_size = 2
num_classes = 4

# pre_logits should be shape (batch_size, num_classes)
pre_logits = torch.randn(batch_size, num_classes)

# log_std should be shape (batch_size, num_classes)
log_std = torch.randn(batch_size, num_classes)

# samples is an integer
samples = 10

ARGS = {
    "pre_logits": pre_logits,
    "log_std": log_std,
    "samples": samples
}
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

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 3 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 3 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.

FunctionAlso present inEarliest-datedOthers named
RobustL1Loss 1 other paper
1 later
2206.04968 (2022-06)
What Information is Necessary and Sufficient to Predict Materials Properties using Machine Learning?
RobustL2Loss 1 other paper
1 later
2206.04968 (2022-06)
What Information is Necessary and Sufficient to Predict Materials Properties using Machine Learning?
sampled_softmax 1 other paper
1 later
2206.04968 (2022-06)
What Information is Necessary and Sufficient to Predict Materials Properties using Machine 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

The adoption of machine learning in materials science has rapidly transformed materials property prediction. Hurdles limiting full capitalization of recent advancements in machine learning include the limited development of methods to learn the underlying interactions of multiple elements, as well as the relationships among multiple properties, to facilitate property prediction in new composition spaces. To address these issues, we introduce the Hierarchical Correlation Learning for Multi-property Prediction (H-CLMP) framework that seamlessly integrates (i) prediction using only a material's composition, (ii) learning and exploitation of correlations among target properties in multi-target regression, and (iii) leveraging training data from tangential domains via generative transfer learning. The model is demonstrated for prediction of spectral optical absorption of complex metal oxides spanning 69 3-cation metal oxide composition spaces. H-CLMP accurately predicts non-linear composition-property relationships in composition spaces for which no training data is available, which broadens the purview of machine learning to the discovery of materials with exceptional properties. This achievement results from the principled integration of latent embedding learning, property correlation learning, generative transfer learning, and attention models. The best performance is obtained using H-CLMP with Transfer learning (H-CLMP(T)) wherein a generative adversarial network is trained on computational density of states data and deployed in the target domain to augment prediction of optical absorption from composition. H-CLMP(T) aggregates multiple knowledge sources with a framework that is well-suited for multi-target regression across the physical sciences.

For agents

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

get_harvested_code_for_paper("2106.02225")
get_code_for_paper("2106.02225")
have("2106.02225")

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

curl https://syntology.ai/api/ran/2106.02225.json

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

[![Syntology run record](https://syntology.ai/api/ran/2106.02225.svg)](https://syntology.ai/paper/2106.02225)

Connect an agent — have() is free.