SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.03563 · 2024

100 instances is all you need: predicting the success of a new LLM on unseen data by testing on a few instances

arXiv · PDF · Open in the Atlas

Code that ran

9 of 9 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; 2 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 6 failures of our instrument; 0 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/2409.03563.json · badge: /api/ran/2409.03563.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
lorypack/referenceinstancespredictability canonical (classifier label)
classifier basis: census
9 ran · 0 no recorded run
FunctionStatusWhere it lives
brierDecomp Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lorypack/referenceinstancespredictability/src/classification_utils.py
code served (permissive licence) · get_code("9007762633c4e575")
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 numpy as np

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

# Create small example data
# preds: predicted probabilities (values between 0 and 1)
# outs: ground truth binary labels (0 or 1)

preds = np.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95], dtype=np.float32)
outs = np.array([0, 0, 0, 1, 0, 1, 1, 1, 1, 1], dtype=np.float32)

ARGS = {
    "preds": preds,
    "outs": outs,
    "return_brier_average": True
}
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, 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
brierScore Ran
ran, violated the contract we drafted
lorypack/referenceinstancespredictability/src/classification_utils.py
code served (permissive licence) · get_code("a63e76c1c5dc2ff8")
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 (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)

# The function brierScore takes two parameters: preds and outs
# It computes: 1 / len(preds) * sum((preds - outs) ** 2)
# So both preds and outs need to support:
# - len() operation
# - subtraction
# - exponentiation (**2)
# - sum() operation

# Create small tensors that can be used as predictions and outputs
preds = torch.tensor([[0.9, 0.1, 0.2, 0.8],
                       [0.7, 0.3, 0.4, 0.6]], dtype=torch.float32)

outs = torch.tensor([[1.0, 0.0, 0.0, 1.0],
                      [1.0, 0.0, 1.0, 0.0]], dtype=torch.float32)

ARGS = {
    "preds": preds,
    "outs": outs
}
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 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
evaluate_predictive_method Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lorypack/referenceinstancespredictability/src/classification_utils.py
code served (permissive licence) · get_code("860217ea71787cbd")
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
import pandas as pd
from sklearn.linear_model import LogisticRegression

# Create small sample data for training
np.random.seed(0)
n_train = 10
n_test = 5
n_features = 4

# Generate random features and binary labels
features_train_data = np.random.randn(n_train, n_features)
features_test_data = np.random.randn(n_test, n_features)
labels_train_data = np.random.randint(0, 2, n_train)
labels_test_data = np.random.randint(0, 2, n_test)

# Create DataFrames
df_train = pd.DataFrame(features_train_data, columns=[f'feature_{i}' for i in range(n_features)])
df_train['target'] = labels_train_data

df_test = pd.DataFrame(features_test_data, columns=[f'feature_{i}' for i in range(n_features)])
df_test['target'] = labels_test_data

# Define feature columns and response column
features_cols = [f'feature_{i}' for i in range(n_features)]
response_col = 'target'

ARGS = {
    'df_train': df_train,
    'df_test': df_test,
    'features_cols': features_cols,
    'response_col': response_col,
    'predictive_method': LogisticRegression,
    'return_trained_method': False,
    'compute_accuracy': True,
    'trained_method': None,
    'binary': True,
    'return_brier_average': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, sklearn — 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
item_curve Ran
ran, honoured the contract we drafted
lorypack/referenceinstancespredictability/src/irt.py
code served (permissive licence) · get_code("a3462b9e34f9705d")
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

# Analyze the function:
# - theta: ability parameter, used in computation a * theta - b
#   The function does .sum(axis=1), so theta should be 2D (batch_size, feature_dim)
# - a: discrimination parameter, shape should match theta for broadcasting
# - b: difficulty parameter, shape should match theta for broadcasting

# Create small test data
batch_size = 2
feature_dim = 4

# theta: shape (2, 4) - ability parameters for 2 subjects across 4 features
theta = np.random.randn(batch_size, feature_dim).astype(np.float64)

# a: discrimination parameter, shape (2, 4) to match theta
a = np.random.rand(batch_size, feature_dim).astype(np.float64) + 0.5

# b: difficulty parameter, shape (2, 4) to match theta
b = np.random.randn(batch_size, feature_dim).astype(np.float64)

ARGS = {
    "theta": theta,
    "a": a,
    "b": b,
}
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
load_irt_parameters Ran
ran; our contract draft was wrong, not the code
lorypack/referenceinstancespredictability/src/irt.py
code served (permissive licence) · get_code("428840006b0d2c7b")
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 json
import tempfile
import os
import numpy as np

# Create a temporary directory and a mock JSON file with IRT parameters
temp_dir = tempfile.mkdtemp()
model_name = os.path.join(temp_dir, "model_")

# Create sample IRT parameters
params = {
    "disc": [[1.2, 1.5, 0.9, 1.1], [1.3, 1.4, 1.0, 1.2], [1.1, 1.6, 0.8, 1.3]],
    "diff": [[0.5, -0.3, 0.2, -0.1], [0.1, 0.4, -0.2, 0.3], [-0.1, 0.6, 0.0, 0.2]],
    "ability": [[0.1, 0.2, 0.3, 0.4], [-0.1, -0.2, 0.1, 0.2], [0.3, 0.5, -0.1, 0.0], [0.2, 0.1, 0.2, 0.3]]
}

# Write the JSON file
with open(model_name + "best_parameters.json", "w") as f:
    json.dump(params, f)

ARGS = {
    "model_name": model_name
}
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
load_with_conditions Ran
ran; our contract draft was wrong, not the code
lorypack/referenceinstancespredictability/src/utils.py
code served (permissive licence) · get_code("10d7f56db0a5521d")
How it ran

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

import os
import pandas as pd
import tempfile

# Create a temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create a test CSV file
test_csv_path = os.path.join(temp_dir, "test_data.csv")
test_df = pd.DataFrame({'col1': [1, 2, 3], 'col2': ['a', 'b', 'c']})
test_df.to_csv(test_csv_path, index=False)

# Create a test pickle file
test_pkl_path = os.path.join(temp_dir, "test_data.pkl")
test_df.to_pickle(test_pkl_path)

# Define ARGS with a valid filename that exists
ARGS = {
    'filename': test_csv_path,
    'overwrite_res': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ngram_vectorize_new Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lorypack/referenceinstancespredictability/src/results_loaders.py
code served (permissive licence) · get_code("04db8a4ce4d86767")
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
from sklearn.feature_extraction.text import TfidfVectorizer

# Create sample text data
train_texts = [
    "this is a sample training text",
    "another training example with words",
    "machine learning is great",
    "text vectorization works well"
]

val_texts = [
    "validation text sample",
    "another validation example"
]

test_texts = [
    "test text here",
    "testing the vectorizer"
]

# Define the ARGS dict with all parameters
ARGS = {
    'train_texts': train_texts,
    'val_texts': val_texts,
    'test_texts': test_texts,
    'ngram_range': (1, 2),
    'token_mode': 'word',
    'min_document_frequency': 2,
    'system_prompt': None
}
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 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
select_features Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lorypack/referenceinstancespredictability/src/results_loaders.py
code served (permissive licence) · get_code("69e94d28f9f47b97")
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 small numpy arrays for the function
np.random.seed(0)

# x_train: vectorized training texts (2 samples, 10 features)
x_train = np.random.rand(2, 10).astype('float32')

# train_labels: training labels (2 samples)
train_labels = np.array([0, 1])

# x_val: vectorized validation texts (2 samples, 10 features)
x_val = np.random.rand(2, 10).astype('float32')

# x_test: vectorized test texts (2 samples, 10 features)
x_test = np.random.rand(2, 10).astype('float32')

# top_k: number of features to select
top_k = 5

ARGS = {
    'x_train': x_train,
    'train_labels': train_labels,
    'x_val': x_val,
    'x_test': x_test,
    'top_k': top_k
}
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 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
sigmoid Ran
ran, honoured the contract we drafted
lorypack/referenceinstancespredictability/src/irt.py
code served (permissive licence) · get_code("a01687b77f001026")
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

# The sigmoid function takes a single parameter 'z' which can be a numeric value or numpy array
# We'll provide a few test cases

# Test case 1: scalar value
scalar_input = 0.0

# Test case 2: numpy array (1D)
array_1d = np.array([0.0, 1.0, -1.0, 2.0])

# Test case 3: numpy array (2D)
array_2d = np.array([[0.0, 1.0], [-1.0, 2.0]])

# Use the scalar input as the main test case
ARGS = {
    "z": scalar_input
}
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

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 2 later) and 6 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
sigmoid 7 other papers
4 dated earlier, 3 later
2006.06887 (2020-06)
Stochastic Optimization for Performative Prediction
2007.13819 (2020-07)
MULTI-LEVEL LOCAL SGD: DISTRIBUTED SGD FOR HETEROGENEOUS HIERARCHICAL NETWORKS
2207.09176 (2022-07)
Self-Supervision Can Be a Good Few-Shot Learner
and 4 more not named here
item_curve 3 other papers
1 dated earlier, 2 later
2402.14992 (2024-02)
tinyBenchmarks: evaluating LLMs with fewer examples
2502.10436 (2025-02)
MERGE 3 : Efficient Evolutionary Merging on Consumer-grade GPUs
2506.07673 (2025-06)
How Benchmark Prediction from Fewer Data Misses the Mark
load_irt_parameters 1 other paper
1 dated earlier
2402.14992 (2024-02)
tinyBenchmarks: evaluating LLMs with fewer examples

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

Predicting the performance of LLMs on individual task instances is essential to ensure their reliability in high-stakes applications. To do so, a possibility is to evaluate the considered LLM on a set of task instances and train an assessor to predict its performance based on features of the instances. However, this approach requires evaluating each new LLM on a sufficiently large set of task instances to train an assessor specific to it. In this work, we leverage the evaluation results of previously tested LLMs to reduce the number of evaluations required to predict the performance of a new LLM. In practice, we propose to test the new LLM on a small set of reference instances and train a generic assessor which predicts the performance of the LLM on an instance based on the performance of the former on the reference set and features of the instance of interest. We conduct empirical studies on HELM-Lite and KindsOfReasoning, a collection of existing reasoning datasets that we introduce, where we evaluate all instruction-fine-tuned OpenAI models until the January 2024 version of GPT4. When predicting performance on instances with the same distribution as those used to train the generic assessor, we find this achieves performance comparable to the LLM-specific assessors trained on the full set of instances. Additionally, we find that randomly selecting the reference instances performs as well as some advanced selection methods we tested. For out of distribution, however, no clear winner emerges and the overall performance is worse, suggesting that the inherent predictability of LLMs is low.

For agents

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

get_harvested_code_for_paper("2409.03563")
get_code_for_paper("2409.03563")
have("2409.03563")

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

curl https://syntology.ai/api/ran/2409.03563.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.03563.svg)](https://syntology.ai/paper/2409.03563)

Connect an agent — have() is free.