SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.30188 · 2026

CalArena: A Large-Scale Post-Hoc Calibration Benchmark

Michael Jordan, David Holzmüller, Francis Bach, Eugène Berta

arXiv · PDF · Open in the Atlas

Code that ran

9 of 13 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 4 failures of our instrument; 4 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/2605.30188.json · badge: /api/ran/2605.30188.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
probkit/CalArena canonical (classifier label)
classifier basis: census
5 ran · 2 no recorded run
probkit/probmetrics community
Papers with Code archive
4 ran · 2 no recorded run
FunctionStatusWhere it lives
CategoricalDirac Ran
ran on a synthesized input · constructed an object; no result was computed
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("2cb97643c6c332ce")
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, CategoricalDirac: 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)

# CategoricalDirac expects:
# - labels: torch.Tensor of shape (n_samples,) with integer class indices
# - n_classes: Optional[int] specifying the number of classes

# Create a small labels tensor with shape (2,)
# Values should be valid class indices (0 to n_classes-1)
labels = torch.tensor([0, 1], dtype=torch.long)

# Specify number of classes
n_classes = 3

ARGS = {
    "labels": labels,
    "n_classes": n_classes,
}
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 CategoricalDirac — 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 CategoricalDirac 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
CategoricalProbs Ran
ran on a synthesized input · constructed an object; no result was computed
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("062d7e1b3c4ebf36")
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, CategoricalProbs: 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)

# Create a simple probs tensor
# Shape: (n_samples=2, n_classes=4)
probs = torch.tensor([
    [0.1, 0.2, 0.3, 0.4],
    [0.25, 0.25, 0.25, 0.25]
], dtype=torch.float32)

ARGS = {
    "probs": probs
}
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 CategoricalProbs — 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 CategoricalProbs 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
Distribution Ran
ran on a synthesized input · constructed an object; no result was computed
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("e9b34bb29943118e")
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, Distribution: 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 pathlib import Path
from typing import Union, Optional, List

torch.manual_seed(0)

class Distribution:
    def __init__(self):
        self._metric_names = ["loss", "accuracy"]
        self._n_samples = 100
        self._data = torch.randn(2, 4)
    
    def get_metric_names(self) -> List[str]:
        return self._metric_names
    
    def save(self, folder: Union[str, Path]):
        folder = Path(folder)
        folder.mkdir(parents=True, exist_ok=True)
    
    @staticmethod
    def load(folder: Union[str, Path]) -> 'Distribution':
        return Distribution()
    
    def get_n_samples(self) -> int:
        return self._n_samples
    
    def __getitem__(self, *args) -> 'Distribution':
        return Distribution()

ARGS = {}
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 Distribution — 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 Distribution 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
compute_winrates Ran
ran, honoured the contract we drafted
probkit/CalArena/utils.py
code served (permissive licence) · get_code("76b7f944d5978263")
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 DataFrame. 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 pandas as pd

np.random.seed(0)

# Create sample benchmark results
n_rows = 12
n_methods = 3
methods = ['method_a', 'method_b', 'method_c']
datasets = ['dataset_1', 'dataset_2', 'dataset_3', 'dataset_4']

# Create results DataFrame with metric columns for each method
data = {}
for method in methods:
    data[f'{method}_brier'] = np.random.rand(n_rows)

# Add dataset column - repeat datasets to cover all rows
data['dataset'] = np.tile(datasets, n_rows // len(datasets) + 1)[:n_rows]

results = pd.DataFrame(data)

ARGS = {
    'results': results,
    'methods': methods,
    'metric': 'brier',
    'bootstrap_datasets': True,
    'n_bootstraps': 100,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
convert_cifar10_to_binary Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
probkit/CalArena/calibration_benchmarks/generate_cv_benchmarks.py
code served (permissive licence) · get_code("5fc9711474109d85")
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 numpy as np
from scipy.special import logsumexp

# Create small sample logits and labels for CIFAR-10
# Logits: batch_size=2, num_classes=10
np.random.seed(0)
logits = np.random.randn(2, 10).astype(np.float32)

# Labels: batch_size=2, each label is between 0-9
labels = np.array([1, 5], dtype=np.int64)

ARGS = {
    "logits": logits,
    "labels": labels,
}
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 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
softmax Ran
ran, honoured the contract we drafted
probkit/CalArena/calibration_benchmarks/generate_cv_benchmarks.py
code served (permissive licence) · get_code("743ce59ae7c6d8e9")
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

# Create a small batch of logits with shape (2, 4)
# This represents 2 samples, each with 4 classes
logits = np.array([
    [1.0, 2.0, 3.0, 0.5],
    [0.1, 0.2, 0.3, 0.4]
], dtype=np.float32)

ARGS = {
    "logits": logits
}
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 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
test_calibrator Ran
ran; our contract draft was wrong, not the code
probkit/CalArena/run_benchmark.py
code served (permissive licence) · get_code("be627f1e164c402e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from unittest.mock import Mock

torch.manual_seed(0)

# Create a simple calibrator mock that has fit and predict_proba methods
class SimpleCalibrator:
    def fit(self, p_cal, y_cal):
        self.p_cal = p_cal
        self.y_cal = y_cal
    
    def predict_proba(self, p_test):
        # Return calibrated probabilities (same shape as input)
        return p_test * 0.9 + 0.05  # Simple transformation

# Create a metrics object with compute_all_from_labels_probs method
class SimpleMetrics:
    def compute_all_from_labels_probs(self, labels, probs):
        # Return a dict of metric names to tensors
        return {
            'accuracy': torch.tensor(0.85),
            'auc': torch.tensor(0.92),
            'ece': torch.tensor(0.05)
        }

# Create calibrator instance
cal = SimpleCalibrator()

# Name for the calibrator
name = 'test_cal'

# Metrics object
metrics = SimpleMetrics()

# Results dictionary to accumulate results
results = {}

# Calibration set probabilities (batch=2, classes=2)
p_cal = np.array([[0.3, 0.7], [0.6, 0.4]], dtype=np.float32)

# Calibration set labels
y_cal = np.array([1, 0], dtype=np.int64)

# Test set probabilities (batch=2, classes=2)
p_test = np.array([[0.4, 0.6], [0.5, 0.5]], dtype=np.float32)

# Test set labels
y_test = np.array([1, 0], dtype=np.int64)

ARGS = {
    'cal': cal,
    'name': name,
    'metrics': metrics,
    'results': results,
    'p_cal': p_cal,
    'y_cal': y_cal,
    'p_test': p_test,
    'y_test': y_test
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
to_one_hot Ran
ran; our contract draft was wrong, not the code
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("3836d7ad343a7f8b")
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 (5, 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 tensor of class labels
y = torch.tensor([0, 1, 2, 1, 0], dtype=torch.long)

# Number of classes
n_classes = 4

# Label smoothing epsilon
label_smoothing_eps = 0.1

ARGS = {
    "y": y,
    "n_classes": n_classes,
    "label_smoothing_eps": label_smoothing_eps,
}
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 (5, 4) — shape (5, 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
unpickle_logits Ran
ran; our contract draft was wrong, not the code
probkit/CalArena/calibration_benchmarks/generate_cv_benchmarks.py
code served (permissive licence) · get_code("ae46b0097de7878e")
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 pickle
import tempfile
import torch
import numpy as np

torch.manual_seed(0)

# Create temporary pickle file with valid data
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.pkl')
temp_path = temp_file.name
temp_file.close()

# Create sample data
logits_cal = torch.randn(2, 4)
labels_cal = torch.randint(0, 2, (2,))
logits_test = torch.randn(2, 4)
labels_test = torch.randint(0, 2, (2,))

# Pickle the data
with open(temp_path, 'wb') as f:
    pickle.dump(((logits_cal, labels_cal), (logits_test, labels_test)), f)

ARGS = {
    'file': temp_path
}
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 — 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
Calibrator Not yet run
unverified (harvested, no recorded run)
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("82840aef3816028a")
How it ran

no invocation recorded.

find_best_config_tabarena Not yet run
unverified (harvested, no recorded run)
probkit/CalArena/calibration_benchmarks/generate_tabarena_benchmarks.py
code served (permissive licence) · get_code("251b319086af199e")
How it ran

no invocation recorded.

load_benchmark_results Not yet run
unverified (harvested, no recorded run)
probkit/CalArena/utils.py
code served (permissive licence) · get_code("60e9aa8069a3bf90")
How it ran

no invocation recorded.

validate_probabilities Not yet run
unverified (harvested, no recorded run)
probkit/probmetrics/probmetrics/calibrators/base.py
code served (permissive licence) · get_code("7fcc3c21ebdae872")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 12 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
to_one_hot
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2602.24230 (2026-02)
A Variational Estimator for L p Calibration Errors

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

Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated. Post-hoc calibration provides a simple and widely used solution, but the large number of proposed methods, combined with small-scale and inconsistent evaluations, makes it difficult to determine which approaches are truly effective in practice. We introduce a large-scale, standardized benchmark for post-hoc calibration, covering nearly 2000 experiments across tabular and computer vision tasks, including binary, multiclass, and largescale classification settings. Our benchmark aggregates predictions from a diverse set of classical models, modern deep learning architectures, and foundation models, and provides unified, reproducible implementations of dozens of calibration methods within a common evaluation framework. We argue that Post-Hoc Improvement (PHI) in proper scoring rules offers a principled alternative to traditional calibration error estimators for comparing post-hoc methods, capturing both calibration quality and potential degradation to the model's predictive performance. Using this framework, we conduct the most comprehensive empirical study of post-hoc calibration to date. Our results reveal consistent patterns across domains: smooth calibration functions outperform binning-based approaches, dedicated multiclass methods are essential in high-dimensional settings, and generic machine learning models are not competitive without calibration-specific design. To facilitate future research, we release all data, code, and evaluation tools, providing a plug-and-play benchmark for developing and comparing calibration methods.

For agents

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

get_harvested_code_for_paper("2605.30188")
get_code_for_paper("2605.30188")
have("2605.30188")

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

curl https://syntology.ai/api/ran/2605.30188.json

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

[![Syntology run record](https://syntology.ai/api/ran/2605.30188.svg)](https://syntology.ai/paper/2605.30188)

Connect an agent — have() is free.