SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.09714 · NeurIPS · 2024

Large language model validity via enhanced conformal prediction methods

Emmanuel Candès, John Cherian, Isaac Gibbs

arXiv · PDF · Open in the Atlas

Code that ran

23 of 32 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; 4 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 17 failures of our instrument; 9 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/2406.09714.json · badge: /api/ran/2406.09714.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
dmis-lab/olaph canonical (classifier label)
classifier basis: census
10 ran · 2 no recorded run
jjcherian/conformal-safety canonical (classifier label)
classifier basis: census
7 ran · 4 no recorded run
jjcherian/conditional-conformal canonical (classifier label)
classifier basis: census
6 ran · 3 no recorded run
FunctionStatusWhere it lives
ROUGESCORE Ran
ran, honoured the contract we drafted
dmis-lab/olaph/pdata_collection.py
pointer only (licence: NONE) · get_code("0bf59e178503b9e6")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

auto_eval Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/pred_to_sft.py
pointer only (licence: NONE) · get_code("cd195a4090c8d1a8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

best_demos Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/Factscore/factscore/atomic_facts.py
pointer only (licence: NONE) · get_code("510528f4db1006bc")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

best_demos Ran
ran; our contract draft was wrong, not the code
jjcherian/conformal-safety/src/atomizer.py
code served (permissive licence) · get_code("50ffc3534adfd228")
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 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.

from rank_bm25 import BM25Okapi

# Create a simple BM25 object with some sample sentences
demos_sents = [
    "the quick brown fox jumps over the lazy dog",
    "a fast red fox leaps over a sleeping dog",
    "the slow turtle walks under the tall tree",
    "birds fly high in the blue sky",
    "cats sleep on comfortable beds"
]

# Tokenize the sentences for BM25
tokenized_sents = [sent.split(' ') for sent in demos_sents]

# Create BM25 object
bm25 = BM25Okapi(tokenized_sents)

# Define parameters
query = "quick fox jumps dog"
k = 2

ARGS = {
    "query": query,
    "bm25": bm25,
    "demos_sents": demos_sents,
    "k": k
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
rank_bm25 — 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 float — 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
binary_search Ran
ran, honoured the contract we drafted
jjcherian/conditional-conformal/conditionalconformal/condconf.py
code served (permissive licence) · get_code("a9044f6ed0707c0e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# binary_search takes a function and searches for a root between min and max
# The function should return a value that changes sign around the root
# We'll create a simple quadratic function: f(x) = x^2 - 4
# This has roots at x = -2 and x = 2
# We'll search between min=0 and max=3, expecting to find root near 2

def quadratic_func(x):
    """Simple function for testing: f(x) = x^2 - 4"""
    return x**2 - 4

ARGS = {
    "func": quadratic_func,
    "min": 0,
    "max": 3,
    "tol": 0.001
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_conformity_scores Ran
ran, honoured the contract we drafted
jjcherian/conformal-safety/src/conformal.py
code served (permissive licence) · get_code("c5b53b2caa9e7c67")
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 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 typing import List

# Construct the dataset
# Each unit has 'atomic_facts' which is a list of dicts with 'is_supported' key
dataset = [
    {
        'atomic_facts': [
            {'is_supported': True},
            {'is_supported': False},
            {'is_supported': True},
        ]
    },
    {
        'atomic_facts': [
            {'is_supported': False},
            {'is_supported': True},
        ]
    }
]

# Construct scores_list
# Each element is a numpy array of scores corresponding to atomic facts
# First unit has 3 atomic facts, so scores array has 3 elements
# Second unit has 2 atomic facts, so scores array has 2 elements
scores_list = [
    np.array([0.1, 0.9, 0.3]),
    np.array([0.5, 0.2])
]

ARGS = {
    'dataset': dataset,
    'scores_list': scores_list
}
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 list, ndim 1, dtype family float — 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
conformal_filter Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jjcherian/conformal-safety/src/conformal.py
code served (permissive licence) · get_code("08b581e4f2313f0c")
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 float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct dataset: List of dicts containing 'atomic_facts'
dataset = [
    {
        'atomic_facts': ['fact1', 'fact2', 'fact3'],
        'id': 0
    },
    {
        'atomic_facts': ['fact4', 'fact5'],
        'id': 1
    }
]

# Construct scores_list: List of lists of scores (one per atomic fact)
scores_list = [
    [0.9, 0.3, 0.7],  # scores for first unit's 3 facts
    [0.5, 0.8]        # scores for second unit's 2 facts
]

# Construct thresholds: List of threshold values (one per unit)
thresholds = [0.6, 0.7]

ARGS = {
    'dataset': dataset,
    'scores_list': scores_list,
    'thresholds': thresholds
}
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 list, ndim 1, dtype family float — 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
eval_score Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/pred_to_preference.py
pointer only (licence: NONE) · get_code("96273073405b5dcc")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

find_unique_element Ran
ran; our contract draft was wrong, not the code
jjcherian/conformal-safety/src/dataset.py
code served (permissive licence) · get_code("f2b316a3259d0a06")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function takes:
# - lst: a list to search through
# - condition: a callable that takes an element and returns True/False
# - approx_index: an integer index

# Create a simple list of integers
lst = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]

# Define a condition function that checks if an element is even
def condition(element):
    return element % 2 == 0

# Approximate index (starting point for search)
approx_index = 3

ARGS = {
    "lst": lst,
    "condition": condition,
    "approx_index": approx_index
}
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 int, ndim 0, dtype family int, 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
generate_annotation_prompt Ran
ran; our contract draft was wrong, not the code
jjcherian/conformal-safety/src/query.py
code served (permissive licence) · get_code("e970ec6b296d3b38")
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 str. 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.

# Analyzing the function to understand what it needs:
# - prompt: str - the original instruction
# - subclaims: List[str] - but looking at _concat_claims, it expects each subclaim to be a dict with a 'message' key

# Looking at _concat_claims:
# It iterates over subclaims and accesses subclaim['message']
# So subclaims should be a list of dicts with 'message' keys

# The function _concat_claims is called with subclaims parameter
# Each subclaim needs to have a 'message' field (accessed via subclaim['message'])

ARGS = {
    'prompt': 'What are the benefits of exercise?',
    'subclaims': [
        {'message': 'Exercise improves cardiovascular health'},
        {'message': 'Exercise makes you feel good'},
        {'message': 'Exercise is a waste of time'}
    ]
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generate_merge_prompt Ran
ran; our contract draft was wrong, not the code
jjcherian/conformal-safety/src/query.py
code served (permissive licence) · get_code("153977b5ca408d0b")
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 str. 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
from typing import Dict, List

MERGE_PROMPT = "You will get an instruction and a set of facts that are true. Construct an answer using ONLY the facts provided, and use ALL of the facts provided. If no facts is given, reply and say that you don't know enough to respond.\n"

def _concat_claims(subclaims: List[str]) -> str:
    return '\n'.join((f"{i}: {subclaim['message']}" for i, subclaim in enumerate(subclaims)))

def generate_merge_prompt(prompt: str, subclaims: List[str]) -> str:
    final_output = MERGE_PROMPT + f'The original instruction was: {prompt}\n'
    final_output += f'The facts are: {_concat_claims(subclaims)}'
    return final_output

prompt = "What is the capital of France?"
subclaims = [
    {"message": "France is a country in Europe"},
    {"message": "Paris is the capital of France"},
    {"message": "The Seine river flows through Paris"}
]

ARGS = {
    "prompt": prompt,
    "subclaims": subclaims
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
generate_subclaim_prompt Ran
ran; our contract draft was wrong, not the code
jjcherian/conformal-safety/src/query.py
code served (permissive licence) · get_code("a305f477743aeac3")
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 str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "prompt": "What is the capital of France?",
    "response": "The capital of France is Paris. It is located in the north-central part of the country. Paris is known for its iconic landmarks like the Eiffel Tower and the Louvre Museum."
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
indicator_matrix Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jjcherian/conditional-conformal/conditionalconformal/synthetic_data.py
code served (permissive licence) · get_code("a43705d36c393243")
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 (4, 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

# Define arguments for indicator_matrix function
scalar_values = [0.5, 1.5, 2.5, 3.5]
disc = [0.0, 1.0, 2.0, 3.0, 4.0]

ARGS = {
    "scalar_values": scalar_values,
    "disc": disc
}
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 (4, 4) — shape (4, 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
is_invalid_paragraph_ppl Ran
ran, violated the contract we drafted
dmis-lab/olaph/Factscore/factscore/abstain_detection.py
pointer only (licence: NONE) · get_code("8eb846a8e76a79b8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

is_invalid_ppl Ran
ran, violated the contract we drafted
dmis-lab/olaph/Factscore/factscore/abstain_detection.py
pointer only (licence: NONE) · get_code("672bd38d830be48d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

make_inst Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/pred_to_preference.py
pointer only (licence: NONE) · get_code("a58480b813525951")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

normalize_answer Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/Factscore/factscore/atomic_facts.py
pointer only (licence: NONE); the stored copy get_code serves, from saptarshi059/cdqa-project, is pointer only (licence: NONE) · get_code("6a96435eba311b08")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

phi_fn_intercept Ran
ran, honoured the contract we drafted
jjcherian/conditional-conformal/experiments/synthetic_trial.py
code served (permissive licence) · get_code("abe7e77347b17747")
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, 1) 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

def phi_fn_intercept(x):
    return np.ones((x.shape[0], 1))

# Create a small numpy array with shape (batch_size, features)
# batch_size=2, features=4
x = np.random.randn(2, 4)

ARGS = {
    "x": x
}
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, 1) — shape (2, 1) 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
phi_fn_shifts Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jjcherian/conditional-conformal/experiments/synthetic_trial.py
code served (permissive licence) · get_code("ca62a7e8fcef8799")
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 (4, 6) 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.stats import norm

eval_locs = [1.5, 3.5]

eval_scale = 0.2

other_locs = [0.5, 2.5, 4.5]

other_scale = 1

def phi_fn_shifts(x):
    shifts = [norm.pdf(x, loc=loc, scale=eval_scale).reshape(-1, 1) for loc in eval_locs]
    shifts.extend([norm.pdf(x, loc=loc, scale=other_scale).reshape(-1, 1) for loc in other_locs])
    shifts.append(np.ones((x.shape[0], 1)))
    return np.concatenate(shifts, axis=1)

# Create a small test input array
x = np.array([0.5, 1.0, 1.5, 2.0])

ARGS = {
    "x": x
}
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 ndarray, ndim 2, dtype family float, shape (4, 6) — shape (4, 6) 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
remove_citation Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/Factscore/factscore/abstain_detection.py
pointer only (licence: NONE); the stored copy get_code serves, from rhyang2021/logu, is pointer only (licence: NONE) · get_code("3635046105ae0fce")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

setup_cvx_problem Ran
ran; our contract draft was wrong, not the code
jjcherian/conditional-conformal/conditionalconformal/condconf.py
code served (permissive licence) · get_code("c738d349dd542153")
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 Problem. 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

np.random.seed(0)

# x_calib: calibration features for kernel computation
x_calib = np.random.randn(5, 3)  # 5 samples, 3 features

# scores_calib: calibration scores (must be 1D or will be reshaped)
scores_calib = np.random.randn(5)

# phi_calib: calibration features for linear constraint (can be None)
phi_calib = np.random.randn(5, 2)  # 5 samples, 2 features

# infinite_params: dict with optional kernel, gamma, lambda parameters
infinite_params = {
    'kernel': 'rbf',
    'gamma': 0.5
}

ARGS = {
    'x_calib': x_calib,
    'scores_calib': scores_calib,
    'phi_calib': phi_calib,
    'infinite_params': infinite_params
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cvxpy, numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Problem — a Problem output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Problem 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
split_dataset Ran
ran; our contract draft was wrong, not the code
jjcherian/conditional-conformal/conditionalconformal/experiment_utils.py
code served (permissive licence) · get_code("4fe1f68fa8d32424")
How it ran

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

import numpy as np

# Create a simple dataset with numpy arrays
# X: 10 samples, 4 features
# Y: 10 labels
X = np.random.randn(10, 4)
Y = np.random.randint(0, 2, 10)
dataset = (X, Y)

# Parameters for splitting
n_test = 2
n_calib = 3
# rng should be a numpy random generator with shuffle method
rng = np.random.RandomState(42)

ARGS = {
    "dataset": dataset,
    "n_test": n_test,
    "n_calib": n_calib,
    "rng": rng
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
text_to_sentences Ran
ran; our contract draft was wrong, not the code
dmis-lab/olaph/Factscore/factscore/atomic_facts.py
pointer only (licence: NONE) · get_code("13867f872a481bd3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

BERTSCORE Not yet run
unverified (harvested, no recorded run)
dmis-lab/olaph/pdata_collection.py
pointer only (licence: NONE) · get_code("91e4954df685088b")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

BLEURT Not yet run
unverified (harvested, no recorded run)
dmis-lab/olaph/pdata_collection.py
pointer only (licence: NONE) · get_code("68d4cb0d0d0e65db")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, dmis-lab/olaph when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

calibrate_thresholds Not yet run
unverified (harvested, no recorded run)
jjcherian/conformal-safety/src/conformal.py
code served (permissive licence) · get_code("17beb7b9e7b9c708")
How it ran

no invocation recorded.

finish_dual_setup Not yet run
unverified (harvested, no recorded run)
jjcherian/conditional-conformal/conditionalconformal/condconf.py
code served (permissive licence) · get_code("7040aef1e8bc0075")
How it ran

no invocation recorded.

generate_cqr_data Not yet run
unverified (harvested, no recorded run)
jjcherian/conditional-conformal/conditionalconformal/synthetic_data.py
code served (permissive licence) · get_code("cb73b72629a8bdb8")
How it ran

no invocation recorded.

get_conformal_loss Not yet run
unverified (harvested, no recorded run)
jjcherian/conformal-safety/src/prob_model.py
code served (permissive licence) · get_code("84209d574223e28e")
How it ran

no invocation recorded.

get_prompts Not yet run
unverified (harvested, no recorded run)
jjcherian/conformal-safety/src/dataset.py
code served (permissive licence) · get_code("94a610d9319c1840")
How it ran

no invocation recorded.

s3_open Not yet run
unverified (harvested, no recorded run)
jjcherian/conformal-safety/src/aws_utils.py
code served (permissive licence) · get_code("7ab268a78b3e6254")
How it ran

no invocation recorded.

torch_ts Not yet run
unverified (harvested, no recorded run)
jjcherian/conditional-conformal/experiments/temperatureScaling.py
code served (permissive licence) · get_code("24635b5492ae6ce5")
How it ran

no invocation recorded.

Also present in other papers' code

Of 32 samples lifted from this paper's linked repositories, 32 carry a function identity and 0 do not. Of the 32 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 6 the same month, 4 later) and 25 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

FunctionAlso present inEarliest-datedOthers named
normalize_answer 37 other papers
29 dated earlier, 1 the same month, 7 later · 1 only as an assembly brick
1808.07042 (2018-08)
CoQA: A Conversational Question Answering Challenge
1810.04805 (2018-10)
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
1902.01718 (2019-02)
End-to-End Open-Domain Question Answering with BERTserini
and 34 more not named here
is_invalid_paragraph_ppl 6 other papers
4 dated earlier, 1 the same month, 1 later
2305.14251 (2023-05)
FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
2402.12659 (2024-02)
FinBen: A Holistic Financial Benchmark for Large Language Models
2402.18045 (2024-02)
Multi-FAct: Assessing Factuality of Multilingual LLMs using FActScore
and 3 more not named here
is_invalid_ppl 6 other papers
4 dated earlier, 1 the same month, 1 later
2305.14251 (2023-05)
FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
2402.12659 (2024-02)
FinBen: A Holistic Financial Benchmark for Large Language Models
2402.18045 (2024-02)
Multi-FAct: Assessing Factuality of Multilingual LLMs using FActScore
and 3 more not named here
remove_citation 6 other papers
4 dated earlier, 1 the same month, 1 later
2305.14251 (2023-05)
FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
2402.12659 (2024-02)
FinBen: A Holistic Financial Benchmark for Large Language Models
2402.18045 (2024-02)
Multi-FAct: Assessing Factuality of Multilingual LLMs using FActScore
and 3 more not named here
best_demos 5 other papers
4 dated earlier, 1 the same month
2305.14251 (2023-05)
FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
2402.12659 (2024-02)
FinBen: A Holistic Financial Benchmark for Large Language Models
2402.18045 (2024-02)
Multi-FAct: Assessing Factuality of Multilingual LLMs using FActScore
and 2 more not named here
text_to_sentences 5 other papers
4 dated earlier, 1 the same month
2305.14251 (2023-05)
FACTSCORE: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
2402.12659 (2024-02)
FinBen: A Holistic Financial Benchmark for Large Language Models
2402.18045 (2024-02)
Multi-FAct: Assessing Factuality of Multilingual LLMs using FActScore
and 2 more not named here
binary_search 1 other paper
1 dated earlier
2207.12274 (2022-07)
MAPIE: an open-source library for distribution-free uncertainty quantification

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

We develop new conformal inference methods for obtaining validity guarantees on the output of large language models (LLMs). Prior work in conformal language modeling identifies a subset of the text that satisfies a high-probability guarantee of correctness. These methods work by filtering claims from the LLM's original response if a scoring function evaluated on the claim fails to exceed a threshold calibrated via split conformal prediction. Existing methods in this area suffer from two deficiencies. First, the guarantee stated is not conditionally valid. The trustworthiness of the filtering step may vary based on the topic of the response. Second, because the scoring function is imperfect, the filtering step can remove many valuable and accurate claims. We address both of these challenges via two new conformal methods. First, we generalize the conditional conformal procedure of Gibbs et al. (2023) in order to adaptively issue weaker guarantees when they are required to preserve the utility of the output. Second, we show how to systematically improve the quality of the scoring function via a novel algorithm for differentiating through the conditional conformal procedure. We demonstrate the efficacy of our approach on biography and medical question-answering datasets.

For agents

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

get_harvested_code_for_paper("2406.09714")
get_code_for_paper("2406.09714")
have("2406.09714")

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

curl https://syntology.ai/api/ran/2406.09714.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.09714.svg)](https://syntology.ai/paper/2406.09714)

Connect an agent — have() is free.