SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2109.14285 · NeurIPS · 2021

Be Confident! Towards Trustworthy Graph Neural Networks via Confidence Calibration

Xiao Wang, Hongrui Liu, Chuan Shi, Cheng Yang

arXiv · PDF · Open in the Atlas

Code that ran

12 of 15 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, 1 violated it and 7 have no contract verdict; 4 failures of our instrument; 3 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/2109.14285.json · badge: /api/ran/2109.14285.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
Diego999/pyGAT canonical (classifier label)
classifier basis not recorded on the link
3 ran · 1 no recorded run
tkipf/pygcn canonical (classifier label)
classifier basis not recorded on the link
3 ran · 0 no recorded run
BUPT-GAMMA/CaGCN community (classifier label)
classifier basis not recorded on the link
6 ran · 1 no recorded run
TaurusTaurus-Rui/CaGCN community (classifier label)
classifier basis not recorded on the link
0 ran · 1 no recorded run
FunctionStatusWhere it lives
brier_score_criterion Ran
ran on a synthesized input
BUPT-GAMMA/CaGCN/util_calibration.py
code served (permissive licence) · get_code("49985109b7aff28c")
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
from torch import nn, optim
from torch.nn import functional as F

torch.manual_seed(0)

# Create small tensors for testing
batch_size = 2
num_classes = 4

# logits: shape (batch_size, num_classes) - raw model outputs
logits = torch.randn(batch_size, num_classes)

# labels: shape (batch_size,) - class indices from 0 to num_classes-1
labels = torch.tensor([0, 2])

# nclass: number of classes
nclass = num_classes

ARGS = {
    "logits": logits,
    "labels": labels,
    "nclass": nclass,
}
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
encode_onehot Ran
ran; our contract draft was wrong, not the code
tkipf/pygcn/pygcn/utils.py
code served (permissive licence) · get_code("3ab88254add4c192")
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 2, dtype family int, shape (6, 3) 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 numpy as np

# The function encode_onehot takes labels as input
# labels should be an array-like of class labels (integers, strings, etc.)
# The function creates a one-hot encoding mapping for unique classes

# Create sample labels with multiple distinct classes
labels = np.array([0, 1, 2, 1, 0, 2], dtype=int)

ARGS = {
    "labels": labels
}
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 2, dtype family int, shape (6, 3) — shape (6, 3) 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
encode_onehot Ran
ran on a synthesized input
Diego999/pyGAT/utils.py
code served (permissive licence) · get_code("f9445ddf8653224c")
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 2, dtype family int, shape (6, 3) 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 numpy as np

# encode_onehot takes labels as input
# labels should be an iterable of class labels that can be converted to a set
# The function will create one-hot encoded vectors for each label

ARGS = {
    "labels": [0, 1, 2, 1, 0, 2]
}
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 2, dtype family int, shape (6, 3) — shape (6, 3) 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
generate_pesudo_label Ran
ran on a synthesized input
BUPT-GAMMA/CaGCN/baseline/co_training.py
code served (permissive licence) · get_code("a7da17a36ae1d68d")
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 2, 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 torch

torch.manual_seed(0)

# Number of samples and classes
n_samples = 10
nclass = 3

# Create boolean tensors for train/test/val splits
idx_train = torch.zeros(n_samples, dtype=torch.bool)
idx_test = torch.zeros(n_samples, dtype=torch.bool)
idx_val = torch.zeros(n_samples, dtype=torch.bool)

# Mark some samples as training data
idx_train[0:3] = True
idx_test[3:6] = True
idx_val[6:8] = True

# Create pseudo labels (initialized with class assignments for training samples)
pseudo_labels = torch.full((n_samples,), -1, dtype=torch.long)
pseudo_labels[0] = 0
pseudo_labels[1] = 1
pseudo_labels[2] = 2

# Create probability matrix P (n_samples x nclass)
# Rows are samples, columns are class probabilities
P = torch.randn(n_samples, nclass)
P = torch.nn.functional.softmax(P, dim=1)  # Normalize to probabilities

# Number of pseudo labels to add per class
topk_pc = 2

ARGS = {
    "idx_train": idx_train,
    "pseudo_labels": pseudo_labels,
    "idx_test": idx_test,
    "idx_val": idx_val,
    "nclass": nclass,
    "topk_pc": topk_pc,
    "P": P
}
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 2, dtype family int — 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_confidence Ran
ran on a synthesized input
BUPT-GAMMA/CaGCN/util_calibration.py
code served (permissive licence) · get_code("f653820785535682")
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 2, 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

torch.manual_seed(0)

# Create a small output tensor with shape (batch_size=2, num_classes=4)
# The function applies softmax and takes max over dim=1 (class dimension)
output = torch.randn(2, 4)

# Define ARGS dict with all parameters
ARGS = {
    "output": output,
    "with_softmax": False
}
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 2, 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
irova_calibrate Ran
ran on a synthesized input
BUPT-GAMMA/CaGCN/util_calibration.py
code served (permissive licence) · get_code("d35fed350a45f9fc")
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 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 numpy as np
from sklearn.isotonic import IsotonicRegression

np.random.seed(0)

# Create small dummy data matching the function's expectations
# logit: (batch_size, num_classes) - raw logits before softmax
# label: (batch_size, num_classes) - binary labels for each class
# logit_eval: (batch_size, num_classes) - logits for evaluation set

batch_size = 2
num_classes = 4

logit = np.random.randn(batch_size, num_classes).astype(np.float32)
label = np.random.randint(0, 2, (batch_size, num_classes)).astype(np.float32)
logit_eval = np.random.randn(batch_size, num_classes).astype(np.float32)

ARGS = {
    "logit": logit,
    "label": label,
    "logit_eval": logit_eval,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, 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
load_data Ran
ran on a synthesized input
tkipf/pygcn/pygcn/utils.py
code served (permissive licence) · get_code("0503d0e1f92b51c3")
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
import scipy.sparse as sp
import torch
import tempfile
import os

torch.manual_seed(0)

# Create a temporary directory and minimal dataset files
temp_dir = tempfile.mkdtemp()
dataset_name = 'cora'

# Create .content file: node_id feature1 feature2 ... featureN label
num_nodes = 100
num_features = 10
num_classes = 7

content_data = []
for i in range(num_nodes):
    node_id = i
    features = [str(np.random.randint(0, 2)) for _ in range(num_features)]
    label = str(np.random.randint(0, num_classes))
    content_data.append([str(node_id)] + features + [label])

content_path = os.path.join(temp_dir, f'{dataset_name}.content')
np.savetxt(content_path, content_data, fmt='%s')

# Create .cites file: source_node target_node
num_edges = 150
cites_data = []
for _ in range(num_edges):
    source = np.random.randint(0, num_nodes)
    target = np.random.randint(0, num_nodes)
    if source != target:
        cites_data.append([source, target])

cites_path = os.path.join(temp_dir, f'{dataset_name}.cites')
np.savetxt(cites_path, cites_data, fmt='%d')

ARGS = {
    'path': temp_dir + '/',
    'dataset': dataset_name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, 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
make_dot Ran
ran; our contract draft was wrong, not the code
Diego999/pyGAT/visualize_graph.py
code served (permissive licence) · get_code("bca7f86b2fb5bf67")
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 Digraph. 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 computation graph
x = torch.randn(2, 4, requires_grad=True)
w = torch.randn(4, 3, requires_grad=True)
b = torch.randn(3, requires_grad=True)

# Forward pass to build autograd graph
y = torch.matmul(x, w) + b
loss = y.sum()
loss.backward()

# Create params dict with named parameters
params = {
    'x': x,
    'w': w,
    'b': b
}

# var should be the output variable (with grad_fn) from backward pass
var = loss

ARGS = {
    'var': var,
    'params': params
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
graphviz, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Digraph — a Digraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Digraph 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
normalize Ran
ran, violated the contract we drafted
tkipf/pygcn/pygcn/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from songw-sw/x-fnc, is pointer only (licence: NONE) · get_code("aa9c29936dee40a7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from songw-sw/x-fnc, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

normalize_adj Ran
ran; our contract draft was wrong, not the code
Diego999/pyGAT/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from triplej0079/DeepIM, is pointer only (licence: NONE) · get_code("d45f434d25f2e870")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from triplej0079/DeepIM, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

normalize_adj Ran
ran; our contract draft was wrong, not the code
BUPT-GAMMA/CaGCN/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from songw-sw/x-fnc, is pointer only (licence: NONE) · get_code("21b0199200507ff0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from songw-sw/x-fnc, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

preprocess_adj Ran
ran on a synthesized input
BUPT-GAMMA/CaGCN/utils.py
code served (permissive licence) · get_code("934ca60e9bd12c56")
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 coo_matrix. 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
import scipy.sparse as sp

# Create a small adjacency matrix (4x4 sparse matrix)
# Using a simple graph structure
adj_data = np.array([1, 1, 1, 1, 1, 1])
adj_row = np.array([0, 0, 1, 1, 2, 3])
adj_col = np.array([1, 2, 0, 3, 1, 2])
adj = sp.coo_matrix((adj_data, (adj_row, adj_col)), shape=(4, 4))

ARGS = {
    "adj": adj,
    "with_ego": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type coo_matrix — a coo_matrix output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coo_matrix 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
generate_pesudo_label Not yet run
unverified (harvested, no recorded run)
BUPT-GAMMA/CaGCN/CaGCN.py
code served (permissive licence) · get_code("bbb7de90271b2771")
How it ran

no invocation recorded.

generate_pesudo_label Not yet run
unverified (harvested, no recorded run)
TaurusTaurus-Rui/CaGCN/baseline/self_training.py
code served (permissive licence) · get_code("39199ce92db5bcf5")
How it ran

no invocation recorded.

load_data Not yet run
unverified (harvested, no recorded run)
Diego999/pyGAT/utils.py
code served (permissive licence) · get_code("d1ed93364d2ef7cb")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 8 are also present in other papers whose code we hold (8 with at least one paper dated earlier, 1 the same month, 5 later) and 7 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
normalize 24 other papers
8 dated earlier, 16 later
1609.02907 (2016-09)
Semi-Supervised Classification with Graph Convolutional Networks
1803.03735 (2018-03)
Attention-based Graph Neural Network for Semi-supervised Learning
1805.09112 (2018-05)
Hyperbolic Neural Networks
and 21 more not named here
normalize_adj 14 other papers
3 dated earlier, 1 the same month, 10 later
1812.00971 (2018-12)
Learning to Learn How to Learn: Self-Adaptive Visual Navigation Using Meta-Learning
2003.06749 (2020-03)
Learning hierarchical relationships for object-goal navigation
2104.09807 (2021-04)
Visual Navigation with Spatial Attention
and 11 more not named here
encode_onehot 12 other papers
6 dated earlier, 6 later
1609.02907 (2016-09)
Semi-Supervised Classification with Graph Convolutional Networks
1710.10903 (2017-10)
Graph Attention Networks
1803.03735 (2018-03)
Attention-based Graph Neural Network for Semi-supervised Learning
and 9 more not named here
normalize_adj 7 other papers
2 dated earlier, 5 later
1710.10903 (2017-10)
Graph Attention Networks
2009.13603 (2020-09)
Visual Pivoting for (Unsupervised) Entity Alignment
2204.05351 (2022-04)
Graph Ordering Attention Networks
and 4 more not named here
encode_onehot 2 other papers
1 dated earlier, 1 later
1905.12921 (2019-05)
Quantifying the Alignment of Graph and Features in Deep Learning
2204.05351 (2022-04)
Graph Ordering Attention Networks
make_dot 2 other papers
2 dated earlier
1710.10903 (2017-10)
Graph Attention Networks
1910.06922 (2019-10)
Gradient penalty from a maximum margin perspective
load_data 1 other paper
1 dated earlier
1710.10903 (2017-10)
Graph Attention Networks
load_data 1 other paper
1 dated earlier
1609.02907 (2016-09)
Semi-Supervised Classification with Graph Convolutional Networks

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

Despite Graph Neural Networks (GNNs) have achieved remarkable accuracy, whether the results are trustworthy is still unexplored. Previous studies suggest that many modern neural networks are over-confident on the predictions, however, surprisingly, we discover that GNNs are primarily in the opposite direction, i.e., GNNs are under-confident. Therefore, the confidence calibration for GNNs is highly desired. In this paper, we propose a novel trustworthy GNN model by designing a topology-aware post-hoc calibration function. Specifically, we first verify that the confidence distribution in a graph has homophily property, and this finding inspires us to design a calibration GNN model (CaGCN) to learn the calibration function. CaGCN is able to obtain a unique transformation from logits of GNNs to the calibrated confidence for each node, meanwhile, such transformation is able to preserve the order between classes, satisfying the accuracypreserving property. Moreover, we apply the calibration GNN to self-training framework, showing that more trustworthy pseudo labels can be obtained with the calibrated confidence and further improve the performance. Extensive experiments demonstrate the effectiveness of our proposed model in terms of both calibration and accuracy.

For agents

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

get_harvested_code_for_paper("2109.14285")
get_code_for_paper("2109.14285")
have("2109.14285")

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

curl https://syntology.ai/api/ran/2109.14285.json

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

[![Syntology run record](https://syntology.ai/api/ran/2109.14285.svg)](https://syntology.ai/paper/2109.14285)

Connect an agent — have() is free.