SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2304.09836 · ICML · 2023

Regions of Reliability in the Evaluation of Multivariate Probabilistic Forecasts

Alexandre Drouin, Valentina Zantedeschi, Étienne Marcotte, Nicolas Chapados

arXiv · PDF · Open in the Atlas

Code that ran

17 of 19 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 17 have no contract verdict; 0 failures of our instrument; 2 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/2304.09836.json · badge: /api/ran/2304.09836.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
ServiceNow/regions-of-reliability canonical (classifier label)
classifier basis: census
17 ran · 2 no recorded run
FunctionStatusWhere it lives
crps_quantile Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/crps.py
code served (permissive licence) · get_code("653bfa7827b0a9c9")
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 float64, ndim 0, 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

# Analyze the function signature and body:
# - target: np.array with shape (batch_size, spatial_dim)
# - samples: np.array with shape (num_samples, batch_size, spatial_dim)
# - quantiles: np.array of quantile values (default provided)

# From assertions:
# target.shape[0] == samples.shape[1]  (batch dimension matches)
# target.shape[1] == samples.shape[2]  (spatial dimension matches)

# Set up small distinct dimensions
batch_size = 2
spatial_dim = 4
num_samples = 5

# Create target array with shape (batch_size, spatial_dim)
target = np.random.randn(batch_size, spatial_dim).astype(np.float32)

# Create samples array with shape (num_samples, batch_size, spatial_dim)
samples = np.random.randn(num_samples, batch_size, spatial_dim).astype(np.float32)

# Create quantiles array (can use default or custom)
quantiles = (np.arange(19) / 19.0)[1:]  # excludes 0, includes values up to ~1.0

ARGS = {
    "target": target,
    "samples": samples,
    "quantiles": quantiles,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
crps_slow Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/crps.py
code served (permissive licence) · get_code("683b8dfe9350f6f8")
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 float32, 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 numpy as np

# Construct valid arguments for crps_slow function
# The function expects:
# - target: np.array with shape (batch_size, dim)
# - samples: np.array with shape (num_samples, batch_size, dim)

batch_size = 4
dim = 3
num_samples = 5

# Create target array with shape (batch_size, dim)
target = np.random.randn(batch_size, dim).astype(np.float32)

# Create samples array with shape (num_samples, batch_size, dim)
samples = np.random.randn(num_samples, batch_size, dim).astype(np.float32)

ARGS = {
    "target": target,
    "samples": samples,
}
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 float32, 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
dawid_sebastiani_score Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/dawid_sebastiani.py
code served (permissive licence) · get_code("58cab579ed6bb634")
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 float64, ndim 0, 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

# Set random seed for reproducibility
np.random.seed(42)

# Create small valid arrays
# target shape: (num_features_0, num_features_1) = (4, 3)
# samples shape: (num_samples, num_features_0, num_features_1) = (2, 4, 3)
target = np.random.randn(4, 3).astype(np.float32)
samples = np.random.randn(2, 4, 3).astype(np.float32)

ARGS = {
    "target": target,
    "samples": samples,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
energy_score Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/energy.py
code served (permissive licence) · get_code("ec34b47822b2cf32")
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 float32, 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 numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Create small test arrays
# target shape: (height, width) = (3, 4)
target = np.random.randn(3, 4).astype(np.float32)

# samples shape: (num_samples, height, width) = (2, 3, 4)
samples = np.random.randn(2, 3, 4).astype(np.float32)

# beta parameter
beta = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "beta": beta,
}
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 float32, 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
energy_score_fast Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/energy.py
code served (permissive licence) · get_code("086d8758cc92c2f9")
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 float32, 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 numpy as np

# Construct arguments for energy_score_fast
# target: np.array with shape (d1, d2)
# samples: np.array with shape (num_samples, d1, d2)
# beta: float between 0 and 2

# Using small dimensions: d1=4, d2=3, num_samples=4
d1 = 4
d2 = 3
num_samples = 4

# Create target array of shape (d1, d2)
target = np.random.randn(d1, d2).astype(np.float32)

# Create samples array of shape (num_samples, d1, d2)
samples = np.random.randn(num_samples, d1, d2).astype(np.float32)

# Create beta value between 0 and 2
beta = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "beta": beta
}
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 float32, 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
exponential_cross_entropy Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/kl_divergence.py
code served (permissive licence) · get_code("622a08d2aedbaeaa")
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 float64, ndim 0, 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

# Create small numpy arrays for the function parameters
scale1 = np.array([1.0, 2.0, 3.0, 0.5])
scale2 = np.array([2.0, 1.5, 4.0, 1.0])

ARGS = {
    "scale1": scale1,
    "scale2": scale2,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
gaussian_cross_entropy Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/kl_divergence.py
code served (permissive licence) · get_code("42eb1613132e5bd3")
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 float64, ndim 0, 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

# Create small covariance matrices and mean vectors
# Using distinct dimensions: 4 for the dimension of the Gaussian

np.random.seed(0)

# Mean vectors (dimension 4)
mu1 = np.array([0.1, 0.2, 0.3, 0.4])
mu2 = np.array([0.5, 0.6, 0.7, 0.8])

# Covariance matrices (4x4, positive definite)
# Create positive definite matrices using A @ A.T
A1 = np.random.randn(4, 4)
cov1 = A1 @ A1.T

A2 = np.random.randn(4, 4)
cov2 = A2 @ A2.T

ARGS = {
    "mu1": mu1,
    "cov1": cov1,
    "mu2": mu2,
    "cov2": cov2
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
gaussian_kl_divergence Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/kl_divergence.py
code served (permissive licence) · get_code("5b1713c369009179")
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 float64, ndim 0, 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

# Set random seed for reproducibility
np.random.seed(0)

# Define small Gaussian parameters
# mu: mean vector (dimension 4)
# cov: covariance matrix (4x4, symmetric positive definite)

dim = 4

# First Gaussian (ground truth)
mu_gt = np.array([0.0, 0.1, 0.2, 0.3])
cov_gt = np.array([
    [1.0, 0.1, 0.0, 0.0],
    [0.1, 1.2, 0.1, 0.0],
    [0.0, 0.1, 1.1, 0.05],
    [0.0, 0.0, 0.05, 1.0]
])

# Second Gaussian (forecast)
mu_fcst = np.array([0.05, 0.15, 0.25, 0.35])
cov_fcst = np.array([
    [1.1, 0.05, 0.0, 0.0],
    [0.05, 1.3, 0.08, 0.0],
    [0.0, 0.08, 1.2, 0.03],
    [0.0, 0.0, 0.03, 1.05]
])

ARGS = {
    "mu_gt": mu_gt,
    "cov_gt": cov_gt,
    "mu_fcst": mu_fcst,
    "cov_fcst": cov_fcst
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
gaussian_mixture_wasserstein_distance Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/wasserstein.py
code served (permissive licence) · get_code("e9707f076eb0fd4d")
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 float, 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 numpy as np
from scipy.linalg import sqrtm

# Set random seed for reproducibility
np.random.seed(0)

# Define dimensions
dim = 4
n_components_1 = 2
n_components_2 = 3

# Create weights for first Gaussian mixture
weights1 = [0.3, 0.7]

# Create weights for second Gaussian mixture
weights2 = [0.2, 0.5, 0.3]

# Create means (mu) for first mixture
mu1 = [
    np.random.randn(dim),
    np.random.randn(dim)
]

# Create covariances for first mixture (must be positive definite)
cov1 = []
for _ in range(n_components_1):
    A = np.random.randn(dim, dim)
    cov1.append(A @ A.T + np.eye(dim))

# Create means (mu) for second mixture
mu2 = [
    np.random.randn(dim),
    np.random.randn(dim),
    np.random.randn(dim)
]

# Create covariances for second mixture (must be positive definite)
cov2 = []
for _ in range(n_components_2):
    A = np.random.randn(dim, dim)
    cov2.append(A @ A.T + np.eye(dim))

ARGS = {
    "weights1": weights1,
    "weights2": weights2,
    "mu1": mu1,
    "cov1": cov1,
    "mu2": mu2,
    "cov2": cov2
}
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 float, 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
gaussian_wasserstein_distance Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/wasserstein.py
code served (permissive licence) · get_code("86b1b5e8b1f99a49")
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 float64, ndim 0, 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.linalg import sqrtm

np.random.seed(0)

# Create two sets of parameters for two Gaussian distributions
# mu1, cov1 for first Gaussian
mu1 = np.array([1.0, 2.0, 3.0, 4.0])
cov1 = np.array([
    [2.0, 0.1, 0.0, 0.0],
    [0.1, 2.5, 0.1, 0.0],
    [0.0, 0.1, 2.0, 0.1],
    [0.0, 0.0, 0.1, 2.2]
])

# mu2, cov2 for second Gaussian
mu2 = np.array([1.5, 2.5, 3.5, 4.5])
cov2 = np.array([
    [1.8, 0.05, 0.0, 0.0],
    [0.05, 2.3, 0.05, 0.0],
    [0.0, 0.05, 1.9, 0.05],
    [0.0, 0.0, 0.05, 2.1]
])

ARGS = {
    "mu1": mu1,
    "cov1": cov1,
    "mu2": mu2,
    "cov2": cov2
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
kernel_laplace Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/kernel.py
code served (permissive licence) · get_code("a89c62ddaf8564cf")
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 float32, 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 numpy as np

# Construct valid arguments for kernel_laplace
# The function expects:
# - target: np.array with shape (height, width)
# - samples: np.array with shape (num_samples, height, width)
# - gamma: float (default=1.0)

# Looking at the assertions:
# target.shape[0] == samples.shape[1]  -> target height == samples height
# target.shape[1] == samples.shape[2]  -> target width == samples width

# Create small test arrays with distinct dimensions for clarity
height = 2
width = 3
num_samples = 4

target = np.random.randn(height, width).astype(np.float32)
samples = np.random.randn(num_samples, height, width).astype(np.float32)
gamma = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "gamma": gamma
}
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 float32, 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
kernel_laplace_spherical Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/kernel.py
code served (permissive licence) · get_code("0d2316045e5123c8")
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 float32, 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 numpy as np

# Construct valid arguments for kernel_laplace_spherical
# The function expects:
# - target: np.array with shape (d1, d2)
# - samples: np.array with shape (num_samples, d1, d2)
# - gamma: float (default 1.0)

# Looking at the assertions:
# assert target.shape[0] == samples.shape[1]
# assert target.shape[1] == samples.shape[2]

# So if target is shape (4, 3), samples must be shape (num_samples, 4, 3)

np.random.seed(0)

target = np.random.randn(4, 3).astype(np.float32)
samples = np.random.randn(2, 4, 3).astype(np.float32)
gamma = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "gamma": gamma
}
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 float32, 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
kernel_matern Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/kernel.py
code served (permissive licence) · get_code("cf9541f4fe97a349")
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 float64, ndim 0, 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

# Construct valid arguments for kernel_matern

# The function expects:
# - target: np.array with shape (d1, d2) where d1*d2 = dim
# - samples: np.array with shape (num_samples, d1, d2)
# - nu: float (default 1.5)
# - gamma: float (default 1.0)

# Let's create small, distinct dimensions
# target will be shape (2, 3) -> dim=6
# samples will be shape (4, 2, 3) -> 4 samples of 2x3

target = np.random.randn(2, 3).astype(np.float64)
samples = np.random.randn(4, 2, 3).astype(np.float64)
nu = 1.5
gamma = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "nu": nu,
    "gamma": gamma
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
select_params Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/experiments/h0_vs_h1.py
code served (permissive licence) · get_code("ec4c5581d554fa67")
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 function. 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)

def dummy_func(target, samples, **kwargs):
    """A simple function that uses target and samples"""
    return target + samples.sum()

ARGS = {
    'func': dummy_func,
    'kwargs': {'multiplier': 2, 'offset': 5}
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
variogram_score_iter Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/variogram.py
code served (permissive licence) · get_code("7db8a498674767fb")
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 float32, 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 numpy as np

# Create small test arrays
# target shape: (height, width) - the spatial dimensions
# samples shape: (num_samples, height, width) - ensemble of predictions

height = 3
width = 4
num_samples = 2

# Create target array with shape (3, 4)
target = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0],
    [9.0, 10.0, 11.0, 12.0]
], dtype=np.float32)

# Create samples array with shape (2, 3, 4)
# 2 samples, each with shape (3, 4)
samples = np.array([
    [[1.1, 2.1, 3.1, 4.1],
     [5.1, 6.1, 7.1, 8.1],
     [9.1, 10.1, 11.1, 12.1]],
    [[0.9, 1.9, 2.9, 3.9],
     [4.9, 5.9, 6.9, 7.9],
     [8.9, 9.9, 10.9, 11.9]]
], dtype=np.float32)

p = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "p": p
}
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 float32, 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
variogram_score_mixed Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/variogram.py
code served (permissive licence) · get_code("7a7e624a0059daf1")
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 float32, 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 numpy as np

# Construct arguments for variogram_score_mixed
# target.shape[0] == samples.shape[1]
# target.shape[1] == samples.shape[2]
# samples.shape[0] is the number of samples (first dimension)

# Let's use:
# target: shape (3, 4) - 3 rows, 4 columns
# samples: shape (2, 3, 4) - 2 samples, 3 rows, 4 columns

target = np.random.randn(3, 4).astype(np.float32)
samples = np.random.randn(2, 3, 4).astype(np.float32)
p = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "p": p
}
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 float32, 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
variogram_score_tensor Ran
ran on a synthesized input
ServiceNow/regions-of-reliability/ror/metrics/variogram.py
code served (permissive licence) · get_code("b76e853f4dce8b7b")
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 float32, 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 numpy as np

# Construct valid arguments for variogram_score_tensor
# The function expects:
# - target: np.array with shape (H, W)
# - samples: np.array with shape (N, H, W) where N is number of samples
# - p: float > 0

# Create target array with shape (4, 3)
target = np.random.randn(4, 3).astype(np.float32)

# Create samples array with shape (2, 4, 3)
# First dimension is number of samples (2)
# Second and third dimensions match target shape (4, 3)
samples = np.random.randn(2, 4, 3).astype(np.float32)

# Create p parameter
p = 1.0

ARGS = {
    "target": target,
    "samples": samples,
    "p": p
}
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 float32, 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
compute_metrics_from_benchmark Not yet run
unverified (harvested, no recorded run)
ServiceNow/regions-of-reliability/ror/benchmark_data.py
code served (permissive licence) · get_code("571e637ac68c5e1d")
How it ran

no invocation recorded.

compute_metrics_single_experiment Not yet run
unverified (harvested, no recorded run)
ServiceNow/regions-of-reliability/ror/benchmark_data.py
code served (permissive licence) · get_code("c764476c7f3110be")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 19 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

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

Repositories linked to this paper

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

Abstract

Multivariate probabilistic time series forecasts are commonly evaluated via proper scoring rules, i.e., functions that are minimal in expectation for the ground-truth distribution. However, this property is not sufficient to guarantee good discrimination in the non-asymptotic regime. In this paper, we provide the first systematic finite-sample study of proper scoring rules for time-series forecasting evaluation. Through a power analysis, we identify the "region of reliability" of a scoring rule, i.e., the set of practical conditions where it can be relied on to identify forecasting errors. We carry out our analysis on a comprehensive synthetic benchmark, specifically designed to test several key discrepancies between ground-truth and forecast distributions, and we gauge the generalizability of our findings to real-world tasks with an application to an electricity production problem. Our results reveal critical shortcomings in the evaluation of multivariate probabilistic forecasts as commonly performed in the literature.

For agents

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

get_harvested_code_for_paper("2304.09836")
get_code_for_paper("2304.09836")
have("2304.09836")

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

curl https://syntology.ai/api/ran/2304.09836.json

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

[![Syntology run record](https://syntology.ai/api/ran/2304.09836.svg)](https://syntology.ai/paper/2304.09836)

Connect an agent — have() is free.