SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2504.13630 · EMNLP · 2025

ReMedy: Learning Machine Translation Evaluation from Human Preferences with Reward Modeling

Christof Monz, Shaomu Tan

arXiv · PDF · Open in the Atlas

Code that ran

13 of 13 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 13 have no contract verdict; 0 failures of our instrument; 0 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/2504.13630.json · badge: /api/ran/2504.13630.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
Smu-Tan/Remedy canonical (classifier label)
classifier basis not recorded on the link
13 ran · 0 no recorded run
FunctionStatusWhere it lives
collect_all_scores Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/calibration.py
code served (permissive licence) · get_code("2f9cee62218224c4")
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 os
import tempfile
import shutil

# Create a temporary directory structure with .seg.score files
temp_dir = tempfile.mkdtemp()

# Create subdirectories for language pairs
lp1_dir = os.path.join(temp_dir, 'en-fr')
lp2_dir = os.path.join(temp_dir, 'de-es')
os.makedirs(lp1_dir, exist_ok=True)
os.makedirs(lp2_dir, exist_ok=True)

# Create .seg.score files with sample data
score_file_1 = os.path.join(lp1_dir, 'scores_1.seg.score')
with open(score_file_1, 'w') as f:
    f.write('segment1\t0.85\n')
    f.write('segment2\t0.92\n')
    f.write('segment3\t0.78\n')
    f.write('segment4\t0.88\n')
    f.write('segment5\t0.95\n')

score_file_2 = os.path.join(lp2_dir, 'scores_2.seg.score')
with open(score_file_2, 'w') as f:
    f.write('segment1\t0.75\n')
    f.write('segment2\t0.81\n')
    f.write('segment3\t0.79\n')
    f.write('segment4\t0.84\n')
    f.write('segment5\t0.90\n')
    f.write('segment6\t0.76\n')
    f.write('segment7\t0.82\n')
    f.write('segment8\t0.87\n')
    f.write('segment9\t0.89\n')
    f.write('segment10\t0.77\n')
    f.write('segment11\t0.91\n')
    f.write('segment12\t0.83\n')

# Create plot directory
plot_dir = tempfile.mkdtemp()

ARGS = {
    'raw_score_dir': temp_dir,
    'save_plot': True,
    'plot_dir': plot_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib — 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
create_dataframe Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt22.py
code served (permissive licence) · get_code("d8cb40f7ba05a108")
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 pandas as pd
import numpy as np

# Define the arguments for create_dataframe
ARGS = {
    'seg_results': {
        'en-de': {
            'metricx_xxl_MQM_2020': 0.85,
            'COMET-22': 0.82,
            'COMETKiwi[noref]': 0.80,
            'test_system': 0.88
        },
        'en-fr': {
            'metricx_xxl_MQM_2020': 0.90,
            'COMET-22': 0.87,
            'COMETKiwi[noref]': 0.84,
            'test_system': 0.91
        },
        'de-en': {
            'metricx_xxl_MQM_2020': 0.78,
            'COMET-22': 0.75,
            'COMETKiwi[noref]': 0.72,
            'test_system': 0.79
        }
    },
    'file_order': ['en-de', 'en-fr', 'de-en'],
    'system_name': 'test_system',
    'sys_results': {
        'metricx_xxl_MQM_2020': 0.84,
        'COMET-22': 0.81,
        'COMETKiwi[noref]': 0.79,
        'test_system': 0.86
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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
create_dataframe Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt23.py
code served (permissive licence) · get_code("09f21f989aaacf1e")
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 pandas as pd

# Construct seg_results: a nested dict mapping lang_pair -> system -> score
seg_results = {
    'en-de': {
        'GEMBA-MQM[noref]': 0.75,
        'XCOMET-Ensemble': 0.80,
        'MetricX-23': 0.78,
        'CustomSystem': 0.82,
    },
    'fr-en': {
        'GEMBA-MQM[noref]': 0.72,
        'XCOMET-Ensemble': 0.77,
        'MetricX-23': 0.76,
        'CustomSystem': 0.79,
    },
    'de-fr': {
        'GEMBA-MQM[noref]': 0.70,
        'XCOMET-Ensemble': 0.75,
        'MetricX-23': 0.74,
        'CustomSystem': 0.80,
    },
}

# file_order: list of lang pairs in order
file_order = ['en-de', 'fr-en', 'de-fr']

# system_name: the custom system being evaluated
system_name = 'CustomSystem'

# sys_results: optional dict mapping system -> system-level score
sys_results = {
    'GEMBA-MQM[noref]': 0.72,
    'XCOMET-Ensemble': 0.77,
    'MetricX-23': 0.76,
    'CustomSystem': 0.80,
}

ARGS = {
    'seg_results': seg_results,
    'file_order': file_order,
    'system_name': system_name,
    'sys_results': sys_results,
}
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
extract_metrics Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt22.py
code served (permissive licence) · get_code("c2108c8a06f7aff7")
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 os
import tempfile
from collections import OrderedDict

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

# Create sample log files matching the pattern '[a-z]+-[a-z]+_results\.log'
log_files_data = {
    'en-de_results.log': """system rank accuracy
metricx_xxl_MQM_2020 1 0.85
COMET-22 2 0.82
COMETKiwi[noref] 3 0.80
my_system 4 0.78
""",
    'fr-en_results.log': """system rank accuracy
metricx_xxl_MQM_2020 1 0.87
COMET-22 2 0.84
COMETKiwi[noref] 3 0.81
my_system 4 0.79
""",
}

# Write log files to temp directory
for filename, content in log_files_data.items():
    filepath = os.path.join(temp_dir, filename)
    with open(filepath, 'w', encoding='utf-8') as f:
        f.write(content)

ARGS = {
    'result_dir': temp_dir,
    'system_name': 'my_system',
}
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 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
extract_metrics Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt23.py
code served (permissive licence) · get_code("78e2a60b74bbece6")
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 os
import tempfile
from collections import OrderedDict

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

# Create sample log files matching the pattern '[a-z]+-[a-z]+_results\.log'
log_files_data = {
    'en-de_results.log': """System Rank Score
GEMBA-MQM[noref] 1 0.85
XCOMET-Ensemble 2 0.82
MetricX-23 3 0.80
MyCustomMetric 4 0.78
""",
    'fr-en_results.log': """System Rank Score
GEMBA-MQM[noref] 1 0.88
XCOMET-Ensemble 2 0.84
MetricX-23 3 0.81
MyCustomMetric 4 0.79
""",
}

# Write the log files to the temp directory
for filename, content in log_files_data.items():
    filepath = os.path.join(temp_dir, filename)
    with open(filepath, 'w', encoding='utf-8') as f:
        f.write(content)

ARGS = {
    'result_dir': temp_dir,
    'system_name': 'MyCustomMetric'
}
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 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
extract_sys_metrics Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt22.py
code served (permissive licence) · get_code("49ac166cd19925c7")
How it ran

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

import os
import tempfile

# Create a temporary file with test data
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt', encoding='utf-8') as f:
    temp_file = f.name
    f.write("System\tRank\tScore\n")
    f.write("metricx_xxl_MQM_2020\t1\t0.85\n")
    f.write("COMET-22\t2\t0.82\n")
    f.write("COMETKiwi[noref]\t3\t0.80\n")
    f.write("test_system_name\t4\t0.78\n")
    f.write("other_system\t5\t0.75\n")

ARGS = {
    'result_file_path': temp_file,
    'system_name': 'test_system_name'
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_sys_metrics Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_result_wmt23.py
code served (permissive licence) · get_code("37ec69329a14a927")
How it ran

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

import os
import tempfile

# Create a temporary result file with the expected format
temp_dir = tempfile.mkdtemp()
result_file_path = os.path.join(temp_dir, "results.txt")

# Write a sample result file with the expected format
# Format: system_name rank accuracy
with open(result_file_path, 'w', encoding='utf-8') as f:
    f.write("System Rank Accuracy\n")
    f.write("GEMBA-MQM[noref] 1 0.85\n")
    f.write("XCOMET-Ensemble 2 0.88\n")
    f.write("MetricX-23 3 0.82\n")
    f.write("TestSystem-XYZ 4 0.79\n")

system_name = "TestSystem-XYZ"

ARGS = {
    "result_file_path": result_file_path,
    "system_name": system_name
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
final_optimized_entropy Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/calibration.py
code served (permissive licence) · get_code("f9e4cf195b0484ff")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from scipy import stats, signal

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

# Create sample scores with different distribution characteristics
# Mix of normal and high-reward scores to test the function
scores_normal = np.random.normal(loc=0.5, scale=1.5, size=50)
scores_high_reward = np.random.normal(loc=5.5, scale=0.8, size=30)
scores_combined = np.concatenate([scores_normal, scores_high_reward])

# Also create a simple unimodal distribution
scores_simple = np.random.uniform(-2, 3, size=40)

ARGS = {
    "scores": scores_simple,
    "force_normal_dist": False,
    "force_high_reward_dist": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_full_lang_name Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/languages.py
code served (permissive licence) · get_code("d1291ad7e7513a2b")
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.

LANG_MAP = {'en': 'English', 'zh': 'Chinese', 'fr': 'French', 'de': 'German', 'es': 'Spanish', 'ru': 'Russian', 'ar': 'Arabic', 'ko': 'Korean', 'ja': 'Japanese', 'cs': 'Czech', 'lv': 'Latvian', 'fi': 'Finnish', 'tr': 'Turkish', 'et': 'Estonian', 'lt': 'Lithuanian', 'kk': 'Kazakh', 'pl': 'Polish', 'ta': 'Tamil', 'bn': 'Bengali', 'hi': 'Hindi', 'mr': 'Marathi', 'ne': 'Nepali', 'ro': 'Romanian', 'si': 'Sinhala', 'uk': 'Ukrainian', 'hr': 'croatian', 'liv': 'Livonian', 'sah': 'Yakut', 'he': 'hebrew', 'gu': 'Gujarati', 'km': 'Khmer', 'ps': 'Pushto', 'ha': 'Hausa', 'is': 'Icelandic', 'xh': 'Xhosa', 'zu': 'Zulu'}

def get_full_lang_name(abbr):
    """Convert language abbreviation to full name."""
    if abbr not in LANG_MAP:
        supported_langs = ', '.join(sorted(LANG_MAP.keys()))
        raise ValueError(f"Language code '{abbr}' is not supported. Supported language codes are: {supported_langs}. Please add '{abbr}' to the LANG_MAP dictionary if needed.")
    return LANG_MAP[abbr]

ARGS = {
    'abbr': 'en'
}
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
is_supported_language Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/languages.py
code served (permissive licence) · get_code("56c142f70a3ca3fe")
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 bool, 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 torch
import torch.nn as nn

# The function is_supported_language only takes one parameter: lang_code
# It checks if lang_code is in the LANG_MAP dictionary
# So we need to provide a valid language code string

LANG_MAP = {'en': 'English', 'zh': 'Chinese', 'fr': 'French', 'de': 'German', 'es': 'Spanish', 'ru': 'Russian', 'ar': 'Arabic', 'ko': 'Korean', 'ja': 'Japanese', 'cs': 'Czech', 'lv': 'Latvian', 'fi': 'Finnish', 'tr': 'Turkish', 'et': 'Estonian', 'lt': 'Lithuanian', 'kk': 'Kazakh', 'pl': 'Polish', 'ta': 'Tamil', 'bn': 'Bengali', 'hi': 'Hindi', 'mr': 'Marathi', 'ne': 'Nepali', 'ro': 'Romanian', 'si': 'Sinhala', 'uk': 'Ukrainian', 'hr': 'croatian', 'liv': 'Livonian', 'sah': 'Yakut', 'he': 'hebrew', 'gu': 'Gujarati', 'km': 'Khmer', 'ps': 'Pushto', 'ha': 'Hausa', 'is': 'Icelandic', 'xh': 'Xhosa', 'zu': 'Zulu'}

def is_supported_language(lang_code):
    """Check if a language code is supported."""
    return lang_code in LANG_MAP

ARGS = {
    'lang_code': 'en'
}
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 bool, 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
load_score_from_text Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/eval_wmt/extract_wmt24_esa.py
code served (permissive licence) · get_code("25030bf189a91632")
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 tempfile
import os

# Create a temporary file with score data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("system_a\t0.75\n")
temp_file.write("system_a\t0.82\n")
temp_file.write("system_a\t0.91\n")
temp_file.write("system_b\t0.68\n")
temp_file.write("system_b\t0.77\n")
temp_file.write("system_c\t0.85\n")
temp_file.close()

input_file_path = temp_file.name

ARGS = {
    "input_file": input_file_path,
    "num_segments": None,
}
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
load_translation_data Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/score.py
code served (permissive licence) · get_code("ef92151eb4f55698")
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 tempfile
import os

# Create temporary files with translation data
temp_dir = tempfile.mkdtemp()

src_file = os.path.join(temp_dir, 'source.txt')
mt_file = os.path.join(temp_dir, 'mt.txt')
ref_file = os.path.join(temp_dir, 'reference.txt')

# Write test data to files
with open(src_file, 'w', encoding='utf-8') as f:
    f.write("Hello world\n")
    f.write("How are you\n")

with open(mt_file, 'w', encoding='utf-8') as f:
    f.write("Bonjour le monde\n")
    f.write("Comment allez vous\n")

with open(ref_file, 'w', encoding='utf-8') as f:
    f.write("Bonjour le monde\n")
    f.write("Comment allez-vous\n")

ARGS = {
    'src_file': src_file,
    'mt_file': mt_file,
    'ref_file': ref_file,
    'src_lang': 'en',
    'tgt_lang': 'fr'
}
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
sigmoid Ran
ran on a synthesized input
Smu-Tan/Remedy/remedy/toolbox/calibration.py
code served (permissive licence) · get_code("9da49703a6c63108")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (4,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# The sigmoid function takes two parameters:
# 1. x: Raw score or array of scores (can be scalar, list, or array)
# 2. temperature: Temperature parameter for scaling (float, default=1.0)

# Create a small array of raw scores
raw_scores = np.array([0.5, -1.2, 2.3, -0.8])

# Define temperature parameter
temperature_value = 0.5

# Construct ARGS dict with all parameters
ARGS = {
    "x": raw_scores,
    "temperature": temperature_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (4,) — shape (4,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 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 13 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

A key challenge in MT evaluation is the inherent noise and inconsistency of human ratings. Regression-based neural metrics struggle with this noise, while prompting LLMs shows promise at system-level evaluation but performs poorly at segment level. In this work, we propose ReMedy, a novel MT metric framework that reformulates translation evaluation as a reward modeling task. Instead of regressing on imperfect human ratings directly, ReMedy learns relative translation quality using pairwise preference data, resulting in a more reliable evaluation. In extensive experiments across WMT22-24 shared tasks (39 language pairs, 111 MT systems), ReMedy achieves stateof-the-art performance at both segment-and system-level evaluation. Specifically, ReMedy-9B surpasses larger WMT winners and massive closed LLMs such as MetricX-13B, XCOMET-Ensemble, GEMBA-GPT-4, PaLM-540B, and finetuned PaLM2. Further analyses demonstrate that ReMedy delivers superior capability in detecting translation errors and evaluating low-quality translations. 1

For agents

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

get_harvested_code_for_paper("2504.13630")
get_code_for_paper("2504.13630")
have("2504.13630")

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

curl https://syntology.ai/api/ran/2504.13630.json

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

[![Syntology run record](https://syntology.ai/api/ran/2504.13630.svg)](https://syntology.ai/paper/2504.13630)

Connect an agent — have() is free.