SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2604.13552 · 2026

Training-Free Test-Time Contrastive Learning for Large Language Models

Fei Liu, Kai Zhou, Kaiwen Zheng, Jinwu Hu, Te Gu, Mingkai Peng

arXiv · PDF · Open in the Atlas

Code that ran

12 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (5 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 1 violated it and 5 have no contract verdict; 6 failures of our instrument; 2 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/2604.13552.json · badge: /api/ran/2604.13552.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
KevinSCUTer/TF-TTCL community
Papers with Code archive
12 ran · 2 no recorded run
FunctionStatusWhere it lives
ActorConfig Ran
ran on a synthesized input · constructed an object; no result was computed
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("6332a246a17c603a")
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, ActorConfig: 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

# Set seed for reproducibility
torch.manual_seed(0)

# Build ARGS dictionary with all parameters for ActorConfig
ARGS = {
    'role_id': 'actor_001',
    'role_type': 'base',
    'temperature': 0.7,
    'top_p': 0.9,
    'top_k': 50,
    'max_tokens': 512,
    'system_prompt': 'You are a helpful assistant that provides detailed responses to user queries and maintains context throughout conversations.'
}
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 ActorConfig — 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 ActorConfig 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
GenerationResult Ran
ran on a synthesized input · constructed an object; no result was computed
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("99ec0840d518cb5d")
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, GenerationResult: 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 typing import List, Dict, Optional, Tuple, Any
from dataclasses import dataclass, field

torch.manual_seed(0)

@dataclass
class ActorConfig:
    """Actor Configuration"""
    role_id: str
    role_type: str = 'base'
    temperature: float = 0.0
    top_p: Optional[float] = None
    top_k: Optional[int] = None
    max_tokens: int = 512
    system_prompt: str = ''

    def to_dict(self) -> Dict[str, Any]:
        """Convert to dictionary"""
        return {'role_id': self.role_id, 'role_type': self.role_type, 'temperature': self.temperature, 'top_p': self.top_p, 'top_k': self.top_k, 'max_tokens': self.max_tokens, 'system_prompt': self.system_prompt[:50] + '...' if len(self.system_prompt) > 50 else self.system_prompt}

@dataclass
class LLMResponse:
    """LLM Response Data Structure"""
    content: str
    logprobs: List[float]
    tokens: List[str]
    total_tokens: int
    prompt_tokens: int
    completion_tokens: int

@dataclass
class PPLResult:
    """PPL Calculation Result"""
    ppl: float
    average_logprob: float
    token_count: int
    logprobs: List[float]

@dataclass
class GenerationResult:
    """Generation Result"""
    role_id: str
    content: str
    ppl_result: PPLResult
    llm_response: LLMResponse
    config: ActorConfig
    question_variant: Optional[str] = None
    system_prompt: Optional[str] = None

    @property
    def ppl(self) -> float:
        """Shortcut to get PPL value"""
        return self.ppl_result.ppl


# Create ActorConfig instance
actor_config = ActorConfig(
    role_id="actor_001",
    role_type="base",
    temperature=0.7,
    top_p=0.9,
    top_k=50,
    max_tokens=512,
    system_prompt="You are a helpful assistant."
)

# Create LLMResponse instance
llm_response = LLMResponse(
    content="This is a test response from the LLM model.",
    logprobs=[-0.5, -0.3, -0.2, -0.1],
    tokens=["This", "is", "a", "test"],
    total_tokens=10,
    prompt_tokens=3,
    completion_tokens=7
)

# Create PPLResult instance
ppl_result = PPLResult(
    ppl=45.2,
    average_logprob=-0.275,
    token_count=4,
    logprobs=[-0.5, -0.3, -0.2, -0.1]
)

# Create GenerationResult with all parameters
ARGS = {
    "role_id": "actor_001",
    "content": "Generated content from the model.",
    "ppl_result": ppl_result,
    "llm_response": llm_response,
    "config": actor_config,
    "question_variant": "variant_a",
    "system_prompt": "You are a helpful assistant for testing purposes."
}
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 GenerationResult — 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 GenerationResult 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
LLMResponse Ran
ran on a synthesized input · constructed an object; no result was computed
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("096987a022c9dd64")
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, LLMResponse: 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
from typing import List, Dict, Optional, Tuple, Any
from dataclasses import dataclass, field

torch.manual_seed(0)

ARGS = {
    "content": "The quick brown fox jumps over the lazy dog",
    "logprobs": [-0.1, -0.2, -0.15, -0.3, -0.25],
    "tokens": ["The", "quick", "brown", "fox", "jumps"],
    "total_tokens": 15,
    "prompt_tokens": 10,
    "completion_tokens": 5,
}
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 LLMResponse — 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 LLMResponse 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
PPLResult Ran
ran on a synthesized input · constructed an object; no result was computed
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("c4154dd7eaebfadc")
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, PPLResult: 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 typing import List, Dict, Optional, Tuple, Any

torch.manual_seed(0)

# PPLResult takes: ppl (float), average_logprob (float), token_count (int), logprobs (List[float])
ARGS = {
    "ppl": 42.5,
    "average_logprob": -2.3,
    "token_count": 150,
    "logprobs": [-1.2, -2.1, -1.5, -2.8, -1.9]
}
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 PPLResult — 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 PPLResult 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
SelectionCandidate Ran
ran on a synthesized input · constructed an object; no result was computed
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("ae93c8176d4e4432")
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, SelectionCandidate: 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
from typing import List, Dict, Optional, Tuple, Any

@dataclass
class ActorConfig:
    """Actor Configuration"""
    role_id: str
    role_type: str = 'base'
    temperature: float = 0.0
    top_p: Optional[float] = None
    top_k: Optional[int] = None
    max_tokens: int = 512
    system_prompt: str = ''

    def to_dict(self) -> Dict[str, Any]:
        """Convert to dictionary"""
        return {'role_id': self.role_id, 'role_type': self.role_type, 'temperature': self.temperature, 'top_p': self.top_p, 'top_k': self.top_k, 'max_tokens': self.max_tokens, 'system_prompt': self.system_prompt[:50] + '...' if len(self.system_prompt) > 50 else self.system_prompt}

@dataclass
class LLMResponse:
    """LLM Response Data Structure"""
    content: str
    logprobs: List[float]
    tokens: List[str]
    total_tokens: int
    prompt_tokens: int
    completion_tokens: int

@dataclass
class PPLResult:
    """PPL Calculation Result"""
    ppl: float
    average_logprob: float
    token_count: int
    logprobs: List[float]

@dataclass
class GenerationResult:
    """Generation Result"""
    role_id: str
    content: str
    ppl_result: PPLResult
    llm_response: LLMResponse
    config: ActorConfig
    question_variant: Optional[str] = None
    system_prompt: Optional[str] = None

    @property
    def ppl(self) -> float:
        """Shortcut to get PPL value"""
        return self.ppl_result.ppl

@dataclass
class SelectionCandidate:
    """Selection Candidate"""
    result: GenerationResult
    extracted_answer: Optional[str]
    similarity: Optional[float] = None

    @property
    def ppl(self) -> float:
        return self.result.ppl_result.ppl

    @property
    def role_id(self) -> str:
        return self.result.role_id

    @property
    def content(self) -> str:
        return self.result.content


torch.manual_seed(0)

ppl_result = PPLResult(
    ppl=2.5,
    average_logprob=-0.8,
    token_count=10,
    logprobs=[-0.7, -0.8, -0.9, -0.75]
)

llm_response = LLMResponse(
    content="This is a test response",
    logprobs=[-0.7, -0.8, -0.9, -0.75],
    tokens=["This", "is", "a", "test"],
    total_tokens=14,
    prompt_tokens=10,
    completion_tokens=4
)

config = ActorConfig(
    role_id="actor_001",
    role_type="base",
    temperature=0.7,
    top_p=0.9,
    top_k=50,
    max_tokens=512,
    system_prompt="You are a helpful assistant"
)

generation_result = GenerationResult(
    role_id="actor_001",
    content="This is a test response",
    ppl_result=ppl_result,
    llm_response=llm_response,
    config=config,
    question_variant="v1",
    system_prompt="You are a helpful assistant"
)

ARGS = {
    "result": generation_result,
    "extracted_answer": "test answer",
    "similarity": 0.85
}
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 SelectionCandidate — 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 SelectionCandidate 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_gsm8k_answer_number Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("abd8904317e723b5")
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 re

def extract_gsm8k_answer_number(completion):
    if '<answer>' in completion and '</answer>' in completion:
        completion = completion.split('<answer>')[-1].split('</answer>')[0]
    elif '\\boxed' in completion:
        idx = completion.rfind('\\boxed')
        if idx != -1:
            open_brace = completion.find('{', idx)
            if open_brace != -1:
                balance = 1
                close_brace = -1
                for i in range(open_brace + 1, len(completion)):
                    if completion[i] == '{':
                        balance += 1
                    elif completion[i] == '}':
                        balance -= 1
                    if balance == 0:
                        close_brace = i
                        break
                if close_brace != -1:
                    completion = completion[open_brace + 1:close_brace]
    tokens = re.split('[\\s\\n\\n]+', completion)
    tokens_with_numbers = [token for token in tokens if re.search('-?\\d', token)]
    cleaned_numbers = [re.sub('[^\\d,\\.-]', '', token) for token in tokens_with_numbers]
    if cleaned_numbers:
        extracted_number = cleaned_numbers[-1].replace(',', '').strip('.')
        if extracted_number.count('.') > 1 or extracted_number.count('-') > 1 or (not re.match('^-?\\d*\\.?\\d*$', extracted_number)):
            return None
        try:
            return str(round(float(extracted_number)))
        except:
            print(extracted_number)
            return None
    return None

ARGS = {
    "completion": "The answer is <answer>42</answer> which is correct."
}
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
fix_a_slash_b Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("5c034529f48184d7")
How it ran

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

import torch

# Analyze the function: fix_a_slash_b takes a single parameter 'string'
# The function expects a string that may contain a '/' character
# It tries to split it, convert to integers, and format as LaTeX fraction

# Test case 1: valid fraction string
string_valid = "3/4"

# Test case 2: invalid format (not a fraction)
string_invalid = "hello/world"

# Test case 3: single value (no slash)
string_no_slash = "42"

# Create ARGS dict with valid test input
ARGS = {
    "string": "5/8"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
fix_fracs Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("9a469d2bed536a0e")
How it ran

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

import torch

def fix_fracs(string):
    substrs = string.split('\\frac')
    new_str = substrs[0]
    if len(substrs) > 1:
        substrs = substrs[1:]
        for substr in substrs:
            new_str += '\\frac'
            if substr[0] == '{':
                new_str += substr
            else:
                try:
                    assert len(substr) >= 2
                except AssertionError:
                    return string
                a = substr[0]
                b = substr[1]
                if b != '{':
                    if len(substr) > 2:
                        post_substr = substr[2:]
                        new_str += '{' + a + '}{' + b + '}' + post_substr
                    else:
                        new_str += '{' + a + '}{' + b + '}'
                elif len(substr) > 2:
                    post_substr = substr[2:]
                    new_str += '{' + a + '}' + b + post_substr
                else:
                    new_str += '{' + a + '}' + b
    string = new_str
    return string

ARGS = {
    'string': 'x\\fracab + \\frac1{2} + \\fraccd3'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
fix_sqrt Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("2a638542d60b227f")
How it ran

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

# Analyzing the function fix_sqrt:
# - It takes a single parameter 'string' which is a string
# - It performs operations on LaTeX sqrt expressions in the string
# - It splits by '\sqrt', processes parts, and reconstructs the string
# - It needs a string containing '\sqrt' to test the main logic

ARGS = {
    'string': r'\sqrt x + \sqrt{y} + z'
}
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_equiv Ran
ran, violated the contract we drafted
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("5015a8c78052f4f0")
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.

# SETUP CODE

# Analyzing the function is_equiv:
# - It takes two string parameters: str1 and str2
# - It takes one optional boolean parameter: verbose (default False)
# - The function performs string manipulations and comparisons
# - No models, tensors, or complex objects are needed

ARGS = {
    'str1': '1/2',
    'str2': '\\frac{1}{2}',
    'verbose': False
}
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
remove_right_units Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("ed7d142070816568")
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.

ARGS = {
    "string": "x^2 + y^2\\text{ meters"
}
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
strip_string Ran
ran; our contract draft was wrong, not the code
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("2e5a927ea1fd6fb2")
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.

# Analyze the function strip_string
# It takes a single parameter: string (str)
# The function performs various string operations and replacements
# No complex objects needed, just a string input

ARGS = {
    "string": "\\frac{1}{2} \\sqrt{4} \\text{ units}"
}
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
SelectionKernel Not yet run
unverified (harvested, no recorded run)
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("bcd899cebc51b855")
How it ran

no invocation recorded.

SelectionResult Not yet run
unverified (harvested, no recorded run)
KevinSCUTer/TF-TTCL/core/source/compare/selection_kernel.py
code served (permissive licence) · get_code("ffa64a7369d56bf9")
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, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 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.

FunctionAlso present inEarliest-datedOthers named
is_equiv
brick of this paper's assembly: helper
4 other papers
3 dated earlier, 1 later
2505.14625 (2025-05)
TinyV: Reducing False Negatives in Verification Improves RL for LLM Reasoning
2511.13223 (2025-11)
TokenSqueeze: Performance-Preserving Compression for Reasoning LLMs
2602.06462 (2026-02)
Diffusion-State Policy Optimization for Masked Diffusion Language Models
and 1 more not named here

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

Large language models (LLMs) demonstrate strong reasoning capabilities, but their performance often degrades under distribution shift. Existing test-time adaptation (TTA) methods rely on gradient-based updates that require white-box access and need substantial overhead, while training-free alternatives are either static or depend on external guidance. In this paper, we propose Training-Free Test-Time Contrastive Learning (TF-TTCL), a training-free adaptation framework that enables a frozen LLM to improve online by distilling supervision from its own inference experiences. Specifically, TF-TTCL implements a dynamic "Explore-Reflect-Steer" loop through three core modules: 1) Semantic Query Augmentation first diversifies problem views via multi-agent role-playing to generate different reasoning trajectories; 2) Contrastive Experience Distillation then captures the semantic gap between superior and inferior trajectories, distilling them into explicit textual rules; and 3) Contextual Rule Retrieval finally activates these stored rules during inference to dynamically steer the frozen LLM toward robust reasoning patterns while avoiding observed errors. Extensive experiments on closed-ended reasoning tasks and openended evaluation tasks demonstrate that TF-TTCL consistently outperforms strong zeroshot baselines and representative TTA methods under online evaluation. Code is available at https://github.com/KevinSCUTer/TF-TTCL.

For agents

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

get_harvested_code_for_paper("2604.13552")
get_code_for_paper("2604.13552")
have("2604.13552")

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

curl https://syntology.ai/api/ran/2604.13552.json

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

[![Syntology run record](https://syntology.ai/api/ran/2604.13552.svg)](https://syntology.ai/paper/2604.13552)

Connect an agent — have() is free.