SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2510.04142 · ICML · 2025

Turning Drift into Constraint: Robust Reasoning Alignment in Non-Stationary Multi-Stream Environments

Jie Lu, Xiaoyu Yang, En Yu, Wei Duan

arXiv · PDF · Open in the Atlas

Code that ran

11 of 11 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, 1 violated it and 0 have no contract verdict; 6 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/2510.04142.json · badge: /api/ran/2510.04142.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
XiaoyuYoung/APO canonical (classifier label)
classifier basis: census
11 ran · 0 no recorded run
FunctionStatusWhere it lives
MMMU_preproc Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/dataset_utils.py
code served (permissive licence) · get_code("99f950f547d92609")
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

# Create a small test DataFrame that mimics MMMU dataset structure
data = pd.DataFrame({
    'A': ['Answer A1', 'Answer A2', np.nan, 'Answer A4'],
    'B': ['Answer B1', 'Answer B2', 'Answer B3', 'Answer B4'],
    'answer': ['Answer A1', 'Answer A2', 'Ground Truth Answer', 'Answer A4']
})

ARGS = {
    'data': data
}
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
can_infer_option Ran
ran, violated the contract we drafted
XiaoyuYoung/APO/evaluation/mmmu/eval_utils.py
code served (permissive licence) · get_code("ffa21dbf032544b9")
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 copy

# Analyze the function to understand what parameters are needed
# The function takes:
# 1. answer: a string that may contain error messages or an answer with choice letters
# 2. choices: a collection of choice identifiers (strings like 'A', 'B', 'C', etc.)

# Test case 1: Valid answer with a single choice
answer_1 = "The answer is A because of the following reasons."
choices_1 = {'A', 'B', 'C', 'D'}

# Test case 2: Answer with rejection phrase
answer_2 = "Sorry, I can't help with images of people yet."
choices_2 = {'A', 'B', 'C'}

# Test case 3: API failure
answer_3 = "Failed to obtain answer via API"
choices_3 = {'A', 'B'}

# Test case 4: Valid single choice with different format
answer_4 = "Based on analysis, the correct option is B."
choices_4 = {'A', 'B', 'C', 'D', 'E'}

# For the function call, we'll use a straightforward example
ARGS = {
    "answer": "The correct answer is A based on the evidence provided in the text.",
    "choices": {'A', 'B', 'C', 'D'}
}
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 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
can_infer_text Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/eval_utils.py
code served (permissive licence) · get_code("64381f4cdeefe5e4")
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 string

# Construct valid arguments for can_infer_text
answer = "The capital of France is Paris, which is a beautiful city."
choices = {
    'A': 'Paris',
    'B': 'London',
    'C': 'Berlin'
}

ARGS = {
    'answer': answer,
    'choices': choices
}
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
ceil_by_factor Ran
ran, honoured the contract we drafted
XiaoyuYoung/APO/qwen-vl-utils/src/qwen_vl_utils/vision_process.py
code served (permissive licence) · get_code("6e45201fa27cb24a")
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 math

def ceil_by_factor(number: int, factor: int) -> int:
    """Returns the smallest integer greater than or equal to 'number' that is divisible by 'factor'."""
    return math.ceil(number / factor) * factor

ARGS = {
    "number": 10,
    "factor": 3
}
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 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
decode_base64_to_image Ran
ran, honoured the contract we drafted
XiaoyuYoung/APO/evaluation/mmmu/common_utils.py
code served (permissive licence) · get_code("1869394f8a7b66a0")
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 PngImageFile. 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 base64
import io
from PIL import Image
import numpy as np

# Create a small valid image in memory
img_array = np.random.randint(0, 256, (10, 10, 3), dtype=np.uint8)
img = Image.fromarray(img_array)

# Convert image to base64 string
img_bytes = io.BytesIO()
img.save(img_bytes, format='PNG')
img_bytes.seek(0)
base64_string = base64.b64encode(img_bytes.read()).decode('utf-8')

ARGS = {
    'base64_string': base64_string
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PngImageFile — a PngImageFile output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PngImageFile output the recorded ndim 3 and dtype family int 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
encode_image_to_base64 Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/common_utils.py
code served (permissive licence) · get_code("3ed464a6a8e2e0b2")
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.

from PIL import Image
import io

# Create a small test image
img = Image.new('RGB', (100, 80), color='red')

ARGS = {
    'image': img,
    'target_size': 50
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — 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
ensure_image_url Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/qwen2_vl/model.py
code served (permissive licence) · get_code("a5f4263865d5afb8")
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 os
import tempfile

# Create a temporary file to test the file:// case
with tempfile.NamedTemporaryFile(delete=False, suffix='.png') as tmp:
    temp_file_path = tmp.name

try:
    ARGS = {
        'image': 'https://example.com/image.jpg'
    }
finally:
    # Clean up temp file
    if os.path.exists(temp_file_path):
        os.remove(temp_file_path)
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
ensure_video_url Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/qwen2_vl/model.py
code served (permissive licence) · get_code("eabaa9c5d46a1a4f")
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 os
import tempfile

# Create a temporary file to use as a valid file path
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, "test_video.mp4")

# Ensure the file exists for testing
with open(temp_file, 'w') as f:
    f.write("test")

ARGS = {
    'video': temp_file
}
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
floor_by_factor Ran
ran, honoured the contract we drafted
XiaoyuYoung/APO/qwen-vl-utils/src/qwen_vl_utils/vision_process.py
code served (permissive licence) · get_code("8155263d7ff19bb3")
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 math

ARGS = {
    "number": 17,
    "factor": 5,
}
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 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
md5 Ran
ran; our contract draft was wrong, not the code
XiaoyuYoung/APO/evaluation/mmmu/common_utils.py
code served (permissive licence) · get_code("ebb7893959c38819")
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 tempfile
import os

# Create a temporary file with some content
temp_file = tempfile.NamedTemporaryFile(mode='wb', delete=False, suffix='.txt')
temp_file.write(b'Hello, this is test content for MD5 hashing.')
temp_file.close()

ARGS = {
    'file_path': temp_file.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 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
round_by_factor Ran
ran, honoured the contract we drafted
XiaoyuYoung/APO/qwen-vl-utils/src/qwen_vl_utils/vision_process.py
code served (permissive licence) · get_code("e252767324188623")
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
import numpy as np

# Analyze the function:
# round_by_factor(number: int, factor: int) -> int
# The function takes two integers and returns an integer.
# It divides number by factor, rounds to nearest integer, then multiplies by factor.

# Set up arguments
number_arg = 17
factor_arg = 5

ARGS = {
    "number": number_arg,
    "factor": factor_arg,
}
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 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

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 10 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 3 the same month, 3 later) and 1 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
round_by_factor 25 other papers
16 dated earlier, 2 the same month, 4 later, 2 same year, order not determinable, 1 undated
2409.12191 (2024-09)
Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
2410.11623 (2024-10)
VidEgoThink: Assessing Egocentric Video Understanding Capabilities for Embodied AI
2412.01720 (2024-12)
LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant
and 22 more not named here
ceil_by_factor 23 other papers
14 dated earlier, 2 the same month, 4 later, 2 same year, order not determinable, 1 undated
2409.12191 (2024-09)
Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
2410.11623 (2024-10)
VidEgoThink: Assessing Egocentric Video Understanding Capabilities for Embodied AI
2412.01720 (2024-12)
LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant
and 20 more not named here
floor_by_factor 23 other papers
14 dated earlier, 2 the same month, 4 later, 2 same year, order not determinable, 1 undated
2409.12191 (2024-09)
Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
2410.11623 (2024-10)
VidEgoThink: Assessing Egocentric Video Understanding Capabilities for Embodied AI
2412.01720 (2024-12)
LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant
and 20 more not named here
can_infer_text 2 other papers
1 dated earlier, 1 same year, order not determinable
2025.findings-acl.964 (2025 (year only))
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models
2506.08011 (2025-06)
Play to Generalize: Learning to Reason Through Game Play
can_infer_option 1 other paper
1 same year, order not determinable
2025.findings-acl.964 (2025 (year only))
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models
decode_base64_to_image 1 other paper
1 same year, order not determinable
2025.findings-acl.964 (2025 (year only))
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models
encode_image_to_base64 1 other paper
1 same year, order not determinable
2025.findings-acl.964 (2025 (year only))
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models
ensure_image_url 1 other paper
1 dated earlier
2507.01949 (2025-07)
Kwai Keye-VL Technical Report
ensure_video_url 1 other paper
1 dated earlier
2507.01949 (2025-07)
Kwai Keye-VL Technical Report
md5 1 other paper
1 same year, order not determinable
2025.findings-acl.964 (2025 (year only))
Multimodal Inconsistency Reasoning (MMIR): A New Benchmark for Multimodal Reasoning Models

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

This paper identifies a critical yet underexplored challenge in reasoning alignment from multiple multi-modal large language models (MLLMs): In non-stationary environments, the diverse reasoning distributions of source models often evolve unpredictably, transmitting systematic biases and drift to the target model. To address this, we formulate multi-source reasoning alignment as a constraint satisfaction problem under concept drift theory. We propose Autonomous Preference Optimization (APO), a novel framework that treats inter-model divergences not as noise, but as dynamic negative constraints. APO operates via a two-stage protocol: first, supervised bootstrapping projects the target model into the capability union of source models; second, constraint-aware optimization synthesizes a consistent consensus manifold by explicitly suppressing drifting trajectories via a multi-negative Plackett-Luce objective. Extensive experiments on chest X-ray interpretation demonstrate that our 7B model achieves superior robustness, outperforming even proprietary source models in average accuracy. Furthermore, we release CXR-MAX, a large-scale benchmark comprising 170,982 reasoning trajectories from seven large-scale MLLMs to facilitate research on reasoning alignment under drift. Code and data are available at: https: //github.com/XiaoyuYoung/APO.

For agents

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

get_harvested_code_for_paper("2510.04142")
get_code_for_paper("2510.04142")
have("2510.04142")

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

curl https://syntology.ai/api/ran/2510.04142.json

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

[![Syntology run record](https://syntology.ai/api/ran/2510.04142.svg)](https://syntology.ai/paper/2510.04142)

Connect an agent — have() is free.