SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2307.09302 · 2023

Conformal prediction under ambiguous ground truth

arXiv · PDF · Open in the Atlas

Code that ran

15 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; 1 honoured the contract we drafted, 0 violated it and 9 have no contract verdict; 5 failures of our instrument; 4 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/2307.09302.json · badge: /api/ran/2307.09302.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
google-deepmind/uncertain_ground_truth canonical (classifier label)
classifier basis: classifier_2025
6 ran · 0 no recorded run
alaalab/webcp community (classifier label)
classifier basis not recorded on the link
9 ran · 4 no recorded run
FunctionStatusWhere it lives
calibrate_mc_threshold Ran
ran; our contract draft was wrong, not the code
google-deepmind/uncertain_ground_truth/monte_carlo.py
code served (permissive licence) · get_code("9454afd876c4a7c7")
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 jax
import jax.numpy as jnp
import jax.random

# Set seed for reproducibility
jax.config.update('jax_default_prng_impl', 'threefry2x32')

# Create a JAX random key
rng = jax.random.PRNGKey(0)

# Create small tensors with distinct sizes
# num_examples = 2, num_classes = 4
num_examples = 2
num_classes = 4

# Conformity scores: shape (num_examples, num_classes) = (2, 4)
conformity_scores = jnp.array([
    [0.1, 0.2, 0.3, 0.4],
    [0.5, 0.6, 0.7, 0.8]
], dtype=jnp.float32)

# Smooth labels: shape (num_examples, num_classes) = (2, 4)
# Must be valid probability distributions (sum to 1 across classes)
smooth_labels = jnp.array([
    [0.1, 0.2, 0.3, 0.4],
    [0.25, 0.25, 0.25, 0.25]
], dtype=jnp.float32)

# Alpha: confidence level in [0, 1]
alpha = 0.1

# num_samples: number of Monte Carlo samples
num_samples = 5

ARGS = {
    "rng": rng,
    "conformity_scores": conformity_scores,
    "smooth_labels": smooth_labels,
    "alpha": alpha,
    "num_samples": num_samples
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — 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
calibrate_threshold Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
google-deepmind/uncertain_ground_truth/conformal_prediction.py
code served (permissive licence) · get_code("9dbcb314e6aad51b")
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 jax.numpy as jnp
import numpy as np

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

# Create small conformity scores: shape (num_examples, num_classes)
# Using shape (2, 4) - 2 examples, 4 classes
conformity_scores = jnp.array([
    [0.1, 0.5, 0.3, 0.2],
    [0.4, 0.2, 0.6, 0.1]
], dtype=jnp.float32)

# Create labels: shape (num_examples,)
# Labels should be valid indices into the num_classes dimension
labels = jnp.array([1, 2], dtype=jnp.int32)

# Create alpha: confidence level in [0, 1]
alpha = 0.1

ARGS = {
    'conformity_scores': conformity_scores,
    'labels': labels,
    'alpha': alpha
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — 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
class_stratified_coverage Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/metrics.py
code served (permissive licence) · get_code("981bb17491bde067")
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 1, dtype family float, shape (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

# Construct arguments for class_stratified_coverage
# The function computes coverage per class in conformal prediction

# conformal_set_arr: boolean array of shape (n_samples, n_classes)
# indicating whether each class is in the conformal set
conformal_set_arr = np.array([
    [True, False, True, False],
    [False, True, True, True]
], dtype=bool)

# true_class_arr: one-hot encoded true classes of shape (n_samples, n_classes)
true_class_arr = np.array([
    [1, 0, 0, 0],
    [0, 1, 0, 0]
], dtype=bool)

ARGS = {
    "conformal_set_arr": conformal_set_arr,
    "true_class_arr": true_class_arr
}
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 1, dtype family float, shape (4,) — shape (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
compute_prediction_sets Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/conformal_risk_control.py
code served (permissive licence) · get_code("3776f69137111df4")
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
import numpy as np

torch.manual_seed(0)

# Create a small probability array
# Using a 2D array with shape (2, 4) - batch size 2, 4 classes
probability_arr = torch.tensor([
    [0.1, 0.3, 0.5, 0.9],
    [0.2, 0.4, 0.6, 0.8]
], dtype=torch.float32)

# Define threshold
threshold = 0.5

ARGS = {
    "probability_arr": probability_arr,
    "threshold": threshold
}
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 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
compute_prediction_sets_threshold Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/conformal_prediction_methods.py
code served (permissive licence) · get_code("d57df6bb388ac156")
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
import numpy as np

torch.manual_seed(0)

# Create a small probability array (2x4)
probability_arr = torch.tensor([
    [0.1, 0.5, 0.9, 0.3],
    [0.2, 0.8, 0.4, 0.7]
], dtype=torch.float32)

# Define a threshold value
threshold = 0.5

ARGS = {
    "probability_arr": probability_arr,
    "threshold": threshold
}
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 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
compute_simscores Ran
ran on a synthesized input
alaalab/webcp/compute_alignments/generate_context_alignments.py
code served (permissive licence) · get_code("1a976aa6e853fa6b")
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 torch.nn as nn

torch.manual_seed(0)

# Build a minimal context encoder that returns an object with last_hidden_state
class MinimalContextEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(8, 4)
    
    def forward(self, input_ids=None, attention_mask=None, **kwargs):
        # Create a minimal output object with last_hidden_state attribute
        batch_size = input_ids.shape[0] if input_ids is not None else 2
        hidden_state = torch.randn(batch_size, 5, 4)  # (batch=2, seq_len=5, hidden_dim=4)
        
        class Output:
            def __init__(self, lhs):
                self.last_hidden_state = lhs
        
        return Output(hidden_state)

context_encoder = MinimalContextEncoder()

# Build query_embedding_dict with embeddings of shape (3, 4) for different classes
query_embedding_dict = {
    'class_a': torch.randn(3, 4),  # (3, 4) - 3 queries, 4-dim embeddings
    'class_b': torch.randn(3, 4),  # (3, 4)
}

# Build tokenized_context - what gets passed to context_encoder
tokenized_context = {
    'input_ids': torch.randint(0, 100, (2, 10)),      # (batch=2, seq_len=10)
    'attention_mask': torch.ones(2, 10, dtype=torch.long),
}

ARGS = {
    'context_encoder': context_encoder,
    'query_embedding_dict': query_embedding_dict,
    'tokenized_context': tokenized_context,
}
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 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
compute_threshold Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/conformal_prediction_methods.py
code served (permissive licence) · get_code("9b4f33f1fb0e29e7")
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.optimize import brentq

np.random.seed(0)

# Create sample data for the function
# probability_arr: shape (N, num_classes) - probability scores for each sample and class
# true_class_arr: shape (N, num_classes) - one-hot encoded true classes

N = 10  # number of samples
num_classes = 4  # number of classes

# Create probability array with random values between 0 and 1
probability_arr = np.random.rand(N, num_classes)

# Create true class array (one-hot encoded)
true_class_arr = np.zeros((N, num_classes))
for i in range(N):
    true_class_arr[i, np.random.randint(0, num_classes)] = 1

# Set alpha (miscoverage level)
alpha = 0.1

ARGS = {
    "alpha": alpha,
    "probability_arr": probability_arr,
    "true_class_arr": true_class_arr,
}
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
conformal_quantile Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
google-deepmind/uncertain_ground_truth/conformal_prediction.py
code served (permissive licence) · get_code("c293ce6b6a5d4d3d")
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 jax.numpy as jnp
import numpy as np

# Create valid arguments for conformal_quantile function

# scores: a JAX array of conformity scores
scores = jnp.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8])

# alpha: confidence level in [0, 1]
alpha = 0.9

ARGS = {
    "scores": scores,
    "alpha": alpha,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — 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
get_fileidx_list Ran
ran on a synthesized input
alaalab/webcp/compute_alignments/generate_context_alignments.py
code served (permissive licence) · get_code("d9ea461f5d5de794")
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 list, ndim 1, 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 tempfile
from pathlib import Path

# Create a temporary directory with mock files
temp_dir = tempfile.TemporaryDirectory()
dataset_subfolder = Path(temp_dir.name)

# Create mock files with the naming pattern: something_INDEX.ext
# The function extracts the index between the last '_' and the last '.'
Path(dataset_subfolder / "data_0.txt").touch()
Path(dataset_subfolder / "data_1.txt").touch()
Path(dataset_subfolder / "data_5.txt").touch()
Path(dataset_subfolder / "sample_10.csv").touch()
Path(dataset_subfolder / "file_3.dat").touch()

ARGS = {
    "dataset_subfolder": dataset_subfolder
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family int — a list 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
mc_conformal_quantile Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
google-deepmind/uncertain_ground_truth/monte_carlo.py
code served (permissive licence) · get_code("2436571bd53cdc72")
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 jax.numpy as jnp
import numpy as np

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

# Create small conformity scores
# Shape should be num_examples * num_samples
num_examples = 5
num_samples = 3
scores = jnp.array(np.random.uniform(0, 1, size=num_examples * num_samples))

# Alpha parameter for confidence level
alpha = 0.9

ARGS = {
    "scores": scores,
    "num_examples": num_examples,
    "num_samples": num_samples,
    "alpha": alpha,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — 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
openclip_process Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/plausibility_generation.py
code served (permissive licence) · get_code("4692dedab0106411")
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 1, dtype family float, shape (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 torch

torch.manual_seed(0)

# Create small tensors for image_logits and text_logits
# image_logits shape: (batch, embedding_dim) - (2, 4)
# text_logits shape: (num_texts, embedding_dim) - (3, 4)
image_logits = torch.randn(2, 4)
text_logits = torch.randn(3, 4)

ARGS = {
    "image_logits": image_logits,
    "text_logits": text_logits,
    "temp": 100.0,
}
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 1, dtype family float, shape (3,) — shape (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
overall_coverage Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/metrics.py
code served (permissive licence) · get_code("829b99fa5448188b")
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 test arrays for conformal sets and true classes
# conformal_set_arr should be a binary array indicating which classes are in the conformal set
# true_class_arr should be a binary array indicating which classes are true
conformal_set_arr = np.array([
    [1, 0, 1, 0],
    [0, 1, 1, 0]
], dtype=bool)

true_class_arr = np.array([
    [0, 0, 1, 0],
    [0, 1, 1, 0]
], dtype=bool)

ARGS = {
    "conformal_set_arr": conformal_set_arr,
    "true_class_arr": true_class_arr
}
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
predict_threshold Ran
ran, honoured the contract we drafted
google-deepmind/uncertain_ground_truth/conformal_prediction.py
code served (permissive licence) · get_code("792f08759b26f4dc")
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 ArrayImpl, ndim 2, dtype family int, 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 jax.numpy as jnp
import numpy as np

# Create small conformity scores array: shape (2, 4) for 2 examples, 4 classes
conformity_scores = jnp.array([
    [0.1, 0.5, 0.8, 0.3],
    [0.2, 0.7, 0.4, 0.9]
], dtype=jnp.float32)

# Create a calibrated threshold value
threshold = 0.5

ARGS = {
    "conformity_scores": conformity_scores,
    "threshold": threshold
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family int, 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
sample_mc_labels Ran
ran; our contract draft was wrong, not the code
google-deepmind/uncertain_ground_truth/monte_carlo.py
code served (permissive licence) · get_code("63fb282a3dc3fb0a")
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 jax
import jax.numpy as jnp
import jax.random

# Set seed for reproducibility
key = jax.random.PRNGKey(0)

# Create small test data
num_examples = 2
num_classes = 4
num_samples = 10

# Create conformity scores: shape (num_examples, num_classes)
# Values should be reasonable conformity scores (e.g., non-negative)
conformity_scores = jnp.array([
    [0.1, 0.5, 0.3, 0.8],
    [0.2, 0.4, 0.9, 0.1]
], dtype=jnp.float32)

# Create smooth labels: shape (num_examples, num_classes)
# Should be normalized probabilities (sum to 1 along last axis)
smooth_labels = jnp.array([
    [0.1, 0.3, 0.2, 0.4],
    [0.25, 0.25, 0.25, 0.25]
], dtype=jnp.float32)

# Create the rng key
rng = jax.random.PRNGKey(42)

ARGS = {
    "rng": rng,
    "conformity_scores": conformity_scores,
    "smooth_labels": smooth_labels,
    "num_samples": num_samples
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — 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
samplewise_loss Ran
ran on a synthesized input
alaalab/webcp/ambiguous_cp/conformal_prediction_methods.py
code served (permissive licence) · get_code("395fb2dace74ef7e")
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 1, dtype family float, shape (2,) 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 sample data for conformal_set_arr
# This should be a binary array indicating which classes are in the conformal set for each sample
# Shape: (batch_size, num_classes)
conformal_set_arr = np.array([
    [1, 0, 1, 0],
    [1, 1, 0, 1]
], dtype=bool)

# Create sample data for true_class_arr
# This should be a one-hot encoded array indicating the true class for each sample
# Shape: (batch_size, num_classes)
true_class_arr = np.array([
    [0, 1, 0, 0],
    [0, 0, 1, 0]
], dtype=bool)

ARGS = {
    "conformal_set_arr": conformal_set_arr,
    "true_class_arr": true_class_arr
}
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 1, dtype family float, shape (2,) — shape (2,) 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
openclip_image_preprocess Not yet run
unverified (harvested, no recorded run)
alaalab/webcp/ambiguous_cp/experiment.py
code served (permissive licence) · get_code("a3b4517e7249c4cf")
How it ran

no invocation recorded.

openclip_process Not yet run
unverified (harvested, no recorded run)
alaalab/webcp/ambiguous_cp/experiment.py
code served (permissive licence) · get_code("33202b9bd3d95c68")
How it ran

no invocation recorded.

openclip_text_preprocess Not yet run
unverified (harvested, no recorded run)
alaalab/webcp/ambiguous_cp/experiment.py
code served (permissive licence) · get_code("cc19e29ec50fcdf5")
How it ran

no invocation recorded.

performance_report Not yet run
unverified (harvested, no recorded run)
alaalab/webcp/ambiguous_cp/analysis.py
code served (permissive licence) · get_code("ad14a67ac002ed26")
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

Conformal Prediction (CP) allows to perform rigorous uncertainty quantification by constructing a prediction set $C(X)$ satisfying $\mathbb{P}(Y \in C(X))\geq 1-α$ for a user-chosen $α\in [0,1]$ by relying on calibration data $(X_1,Y_1),...,(X_n,Y_n)$ from $\mathbb{P}=\mathbb{P}^{X} \otimes \mathbb{P}^{Y|X}$. It is typically implicitly assumed that $\mathbb{P}^{Y|X}$ is the "true" posterior label distribution. However, in many real-world scenarios, the labels $Y_1,...,Y_n$ are obtained by aggregating expert opinions using a voting procedure, resulting in a one-hot distribution $\mathbb{P}_{vote}^{Y|X}$. For such ``voted'' labels, CP guarantees are thus w.r.t. $\mathbb{P}_{vote}=\mathbb{P}^X \otimes \mathbb{P}_{vote}^{Y|X}$ rather than the true distribution $\mathbb{P}$. In cases with unambiguous ground truth labels, the distinction between $\mathbb{P}_{vote}$ and $\mathbb{P}$ is irrelevant. However, when experts do not agree because of ambiguous labels, approximating $\mathbb{P}^{Y|X}$ with a one-hot distribution $\mathbb{P}_{vote}^{Y|X}$ ignores this uncertainty. In this paper, we propose to leverage expert opinions to approximate $\mathbb{P}^{Y|X}$ using a non-degenerate distribution $\mathbb{P}_{agg}^{Y|X}$. We develop Monte Carlo CP procedures which provide guarantees w.r.t. $\mathbb{P}_{agg}=\mathbb{P}^X \otimes \mathbb{P}_{agg}^{Y|X}$ by sampling multiple synthetic pseudo-labels from $\mathbb{P}_{agg}^{Y|X}$ for each calibration example $X_1,...,X_n$. In a case study of skin condition classification with significant disagreement among expert annotators, we show that applying CP w.r.t. $\mathbb{P}_{vote}$ under-covers expert annotations: calibrated for $72\%$ coverage, it falls short by on average $10\%$; our Monte Carlo CP closes this gap both empirically and theoretically.

For agents

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

get_harvested_code_for_paper("2307.09302")
get_code_for_paper("2307.09302")
have("2307.09302")

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

curl https://syntology.ai/api/ran/2307.09302.json

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

[![Syntology run record](https://syntology.ai/api/ran/2307.09302.svg)](https://syntology.ai/paper/2307.09302)

Connect an agent — have() is free.