SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2607.04429 · 2026

evalci: A Python Library for Statistically Rigorous Comparison of Language Model Evaluations

Shreyas Chandrahas

arXiv · PDF · Open in the Atlas

Code that ran

13 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 8 failures of our instrument; 1 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2607.04429.json · badge: /api/ran/2607.04429.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
Shreyaskc/evalci community
Papers with Code archive
13 ran · 1 no recorded run
FunctionStatusWhere it lives
CIResult Ran
ran on a synthesized input · constructed an object; no result was computed
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("6504ccb986870a77")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CIResult: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from dataclasses import dataclass, field

torch.manual_seed(0)

ARGS = {
    "estimate": 0.5,
    "lower": 0.45,
    "upper": 0.55,
    "method": "bootstrap",
    "n": 1000,
    "confidence": 0.95
}
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 CIResult — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CIResult 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
CompareResult Ran
ran on a synthesized input · constructed an object; no result was computed
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("891cb21c235fa770")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CompareResult: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from dataclasses import field

# Construct arguments for CompareResult
ARGS = {
    'delta': 0.15,
    'ci': (0.05, 0.25),
    'p_value': 0.032,
    'method': 'welch',
    'paired': False,
    'n': 42,
    'confidence': 0.95,
    'extra': {'notes': 'test comparison'}
}
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 CompareResult — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CompareResult 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
bootstrap_interval Ran
ran; our contract draft was wrong, not the code
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("da9874b0603c50e9")
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
from scipy import stats as _scipy_stats

np.random.seed(0)

# Create sample data for bootstrap analysis
data = np.array([1.2, 1.5, 1.8, 2.1, 2.3, 2.5, 2.8, 3.0, 3.2, 3.5])

# Define a custom statistic function (mean is default but we can use it explicitly)
def custom_statistic(x):
    return np.mean(x)

ARGS = {
    'data': data,
    'statistic': np.mean,
    'method': 'BCa',
    'confidence': 0.95,
    'n_resamples': 9999,
    'random_state': 0,
    'vectorized': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 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
ci Ran
ran; our contract draft was wrong, not the code
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("6219547495fa7208")
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 CIResult. 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

# Test case 1: Binary scores with wilson method
scores_binary = np.array([1.0, 1.0, 0.0, 1.0, 0.0])

ARGS = {
    'scores': scores_binary,
    'method': 'wilson',
    'confidence': 0.95,
    'n_resamples': 9999,
    'random_state': None,
}
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 CIResult — a CIResult output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a CIResult 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
clopper_pearson_interval Ran
ran, honoured the contract we drafted
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("f03ac473a85bde77")
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.

from scipy import stats as _scipy_stats

ARGS = {
    'k': 7,
    'n': 10,
    'confidence': 0.95
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
scipy — 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
mcnemar_test Ran
ran, honoured the contract we drafted
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("177e24430c39228f")
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

# Create test data for McNemar's test
# a and b should be binary arrays (0/1 correctness outcomes)
# We'll create arrays where:
# - some predictions match (both 0 or both 1)
# - some differ (n01: a=0, b=1; n10: a=1, b=0)

np.random.seed(0)

# Create paired binary outcomes
a = np.array([1, 0, 1, 1, 0, 1, 0, 0, 1, 0])
b = np.array([1, 1, 1, 0, 0, 1, 1, 0, 1, 0])

ARGS = {
    'a': a,
    'b': b,
    'correction': True,
    'exact': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 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
paired_bootstrap_ci Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("620b1f42d4971727")
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(42)

# Create a small array of differences for bootstrap CI computation
diffs = np.array([0.1, 0.2, 0.15, 0.25, 0.18, 0.22, 0.19, 0.21, 0.17, 0.23])

# Define arguments for paired_bootstrap_ci
ARGS = {
    'diffs': diffs,
    'confidence': 0.95,
    'n_resamples': 9999,
    'rng': np.random.default_rng(seed=42)
}
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
paired_bootstrap_p Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("6e8bffa3e62ee384")
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

np.random.seed(0)
rng = np.random.default_rng(seed=0)

# Create small paired differences array
diffs = np.array([0.5, -0.3, 0.2, 0.1, -0.15])

ARGS = {
    "diffs": diffs,
    "n_resamples": 9999,
    "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 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
paired_permutation_p Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("04fc13d94e614b4e")
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
import itertools

np.random.seed(42)

# Create small paired differences array
# These are the differences between paired observations
diffs = np.array([0.5, -0.3, 0.2, 0.1, -0.4, 0.15], dtype=float)

# Create RNG for reproducibility
rng = np.random.default_rng(seed=42)

ARGS = {
    "diffs": diffs,
    "n_resamples": 9999,
    "rng": rng,
    "exact_threshold": 12
}
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 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
unpaired_bootstrap_ci Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("4b7a577202a14f70")
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(42)

# Create two small numpy arrays with distinct sizes
a = np.array([1.2, 2.3, 3.1, 4.5, 5.2])
b = np.array([2.1, 3.4, 4.2])

# Create RNG
rng = np.random.default_rng(seed=42)

ARGS = {
    "a": a,
    "b": b,
    "confidence": 0.95,
    "n_resamples": 9999,
    "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, 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
unpaired_bootstrap_p Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("a547f79c157fc13e")
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

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

# Create two independent sample arrays with distinct sizes
a = np.array([1.2, 2.3, 3.1, 4.5, 5.2])
b = np.array([2.1, 3.0, 4.2])

# Create an RNG instance
rng = np.random.default_rng(seed=42)

# Define arguments dictionary
ARGS = {
    'a': a,
    'b': b,
    'n_resamples': 9999,
    '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 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
unpaired_permutation_p Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("4555f85b0d6608a2")
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

# Create sample data for the two independent samples
a = np.array([1.2, 2.3, 3.1, 4.5, 5.2])
b = np.array([2.1, 3.2, 4.3, 5.1, 6.2])

# Create a random number generator
rng = np.random.default_rng(seed=42)

ARGS = {
    'a': a,
    'b': b,
    'n_resamples': 9999,
    '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 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
wilson_interval Ran
ran, honoured the contract we drafted
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("53d188c2463155be")
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
from scipy import stats as _scipy_stats

def wilson_interval(k, n, confidence=0.95):
    """Wilson score interval for a binomial proportion."""
    if n <= 0:
        raise ValueError('n must be positive')
    z = _scipy_stats.norm.ppf(0.5 + confidence / 2)
    phat = k / n
    z2 = z * z
    denom = 1 + z2 / n
    center = (phat + z2 / (2 * n)) / denom
    half = z * np.sqrt(phat * (1 - phat) / n + z2 / (4 * n * n)) / denom
    return (max(0.0, center - half), min(1.0, center + half))


# Set up arguments
k = 7  # number of successes
n = 20  # number of trials
confidence = 0.95  # confidence level

ARGS = {
    'k': k,
    'n': n,
    'confidence': confidence
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 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
compare Not yet run
unverified (harvested, no recorded run)
Shreyaskc/evalci/src/evalci/stats.py
code served (permissive licence) · get_code("50c0c4ad02a4c7b8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 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 14 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

The dominant practice in language model evaluation is to report a single accuracy number per model and declare the higher one better, without testing whether the gap could plausibly be sampling noise. On benchmarks of a few thousand items, and under temperature sampling where a model can differ from itself run to run by more than the reported gap between models, this practice routinely overstates confidence in headline claims. The statistical machinery to fix thisconfidence intervals, paired significance tests, power analysis, clustered standard errors, multiplecomparison correction -is well established, but no standard, pip-installable tool packages it in the shape an evaluation actually takes: a per-item results table. We present evalci, a pure-Python library (numpy/scipy/pandas only) that turns a per-item results table into a publicationready claim -e.g., "Model A beats Model B, ∆ = 3.1 pts, 95% CI [1.2, 5.0], paired permutation p = 0.002, n = 1,319" -in one function call, with adapters for lm-evaluation-harness and HELM output. Every routine is validated against an independent reference (statsmodels, or brute-force exact enumeration) rather than only against itself. As a case study, we reanalyze a public comparison of nine language models' MMLU accuracy and find that 3 of the 8 adjacent leaderboard-rank gaps are not statistically significant after correcting for the 36 pairwise comparisons the ranking implies.

For agents

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

get_harvested_code_for_paper("2607.04429")
get_code_for_paper("2607.04429")
have("2607.04429")

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

curl https://syntology.ai/api/ran/2607.04429.json

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

[![Syntology run record](https://syntology.ai/api/ran/2607.04429.svg)](https://syntology.ai/paper/2607.04429)

Connect an agent — have() is free.