SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.11639 · 2026

PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning

Yang Song, Dongsheng Wang, Chen Yang, Yifan Wang, Tao Zhang, Shuo Shang, Ke Shi, Shen Gao, Zhenwei An, Guangyue Peng, Zongchao Chen, Yuntao Wen, and 5 more

arXiv · PDF · Open in the Atlas

Code that ran

12 of 12 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; 1 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 9 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/2602.11639.json · badge: /api/ran/2602.11639.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
LiveCodeBench/LiveCodeBench canonical (classifier label)
classifier basis: census
12 ran · 0 no recorded run
FunctionStatusWhere it lives
check_correctness Ran
ran, violated the contract we drafted
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/utils_execute.py
code served (permissive licence) · get_code("d4af0384e850be8d")
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 multiprocessing

# Simple Python program that passes execution
check_program = """
def hello():
    return 42

result = hello()
assert result == 42
"""

timeout = 3

ARGS = {
    "check_program": check_program,
    "timeout": timeout,
}
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
check_testcase_output Ran
ran, violated the contract we drafted
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/compute_test_output_prediction_metrics.py
code served (permissive licence) · get_code("965d069ea8fcbe2c")
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 ast
import json

# Analyze the function check_testcase_output
# Parameters:
# 1. testcase_str: A string containing test case code, possibly multi-line
# 2. expected_output: A string containing JSON-serializable expected output

# The function:
# - Splits testcase_str by lines and looks for 'assert' statements
# - Parses assert statements to extract the right side of '=='
# - Evaluates the extracted expression
# - Parses expected_output as JSON
# - Compares the two evaluated values

# Simple valid test case 1: basic assertion with equality
testcase_str_1 = "assert 2 + 2 == 4"
expected_output_1 = "4"

# Simple valid test case 2: string comparison
testcase_str_2 = "assert 'hello' == 'hello'"
expected_output_2 = '"hello"'

# Simple valid test case 3: list comparison
testcase_str_3 = "assert [1, 2, 3] == [1, 2, 3]"
expected_output_3 = "[1, 2, 3]"

# Multi-line test case with comments and assertion
testcase_str_4 = """# This is a comment
x = 5
# Another comment
assert x + 5 == 10"""
expected_output_4 = "10"

# Test with dict comparison
testcase_str_5 = "assert {'a': 1} == {'a': 1}"
expected_output_5 = '{"a": 1}'

ARGS = {
    "testcase_str": testcase_str_1,
    "expected_output": expected_output_1
}
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
clean_if_name Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/testing_util.py
code served (permissive licence) · get_code("5faad4dd18dd77ef")
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 ast

code_with_if_name = '''
def hello():
    print("Hello")

if __name__ == '__main__':
    hello()
'''

code_without_if_name = '''
def hello():
    print("Hello")
'''

code_invalid = '''
this is not valid python code {{{
'''

ARGS = {
    "code": code_with_if_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
compute_metrics_from_results Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/pass_k_utils.py
code served (permissive licence) · get_code("9188e3d845f5322a")
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 numpy as np

# Construct the results dictionary that compute_metrics_from_results expects
# The function iterates through results.items() where each value is a list of generations
# Each generation is converted to a numpy array and checked if all elements > 0

results = {
    'task_1': [
        [1, 2, 3],      # generation 1: all > 0, so correct
        [0, 1, 2],      # generation 2: not all > 0, so incorrect
        [2, 3, 4],      # generation 3: all > 0, so correct
    ],
    'task_2': [
        [1, 1, 1],      # generation 1: all > 0, so correct
        [1, 2, 3],      # generation 2: all > 0, so correct
    ],
    'task_3': [
        [0, 0, 0],      # generation 1: not all > 0, so incorrect
        [-1, 2, 3],     # generation 2: not all > 0, so incorrect
        [1, 1, 1],      # generation 3: all > 0, so correct
        [2, 2, 2],      # generation 4: all > 0, so correct
        [3, 3, 3],      # generation 5: all > 0, so correct
    ],
}

k_list = [1, 5]

ARGS = {
    'results': results,
    'k_list': k_list,
}
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 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
estimate_pass_at_k Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/pass_k_utils.py
pointer only (licence: MIT); the stored copy get_code serves, from zai-org/glm-simple-evals, is pointer only (licence: NONE) · get_code("9df10eb02582b09f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from zai-org/glm-simple-evals, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

extract_instance_results Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/pass_k_utils.py
code served (permissive licence) · get_code("406597cd2faf5fef")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list. 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

# The function expects:
# - results: a dict where keys are task_ids and values are lists of generations
# - each generation is a list/iterable of numeric values (grades)
# - the function checks if all grades > 0 for each generation

# Create sample results data
results = {
    "task_1": [
        [0.5, 0.8, 0.9],      # generation 1: all > 0, so True
        [0.1, -0.2, 0.3],     # generation 2: has -0.2, so False
    ],
    "task_0": [
        [0.7, 0.6, 0.8, 0.9], # generation 1: all > 0, so True
    ],
    "task_2": [
        [0.2, 0.3],            # generation 1: all > 0, so True
        [0.1, 0.0, 0.5],       # generation 2: has 0.0 which is not > 0, so False
        [-0.5, 0.3],           # generation 3: has -0.5, so False
    ],
}

ARGS = {
    "results": results
}
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 list — a list 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
make_cot_output_prompt Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/prompts/code_execution.py
code served (permissive licence) · get_code("878b793b991d04e2")
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
import numpy as np

# Analyze the function signature and usage
# make_cot_output_prompt(s) takes a single parameter 's'
# The function unpacks s as: code, input = s
# So 's' should be a tuple/list of (code, input)
# 'code' is a string containing Python function definition
# 'input' is a string containing the function call assertion

# Create simple test code and input
code = """def add_numbers(a, b):
    return a + b"""

input_str = "add_numbers(a = 2, b = 3)"

# Construct the tuple
s = (code, input_str)

ARGS = {
    "s": s
}
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 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
make_direct_output_prompt Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/prompts/code_execution.py
code served (permissive licence) · get_code("296f66764dc9eb63")
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

torch.manual_seed(0)

code = """def repeatNumber(number : int) -> int:
    return number"""

input_str = "repeatNumber(number = 17)"

s = (code, input_str)

ARGS = {
    "s": s
}
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
make_function Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/testing_util.py
code served (permissive licence) · get_code("e478b5bb18723734")
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 ast

def make_function(code: str) -> str:
    try:
        import_stmts = []
        all_other_stmts = []
        astree = ast.parse(code)
        for stmt in astree.body:
            if isinstance(stmt, (ast.Import, ast.ImportFrom)):
                import_stmts.append(stmt)
            else:
                all_other_stmts.append(stmt)
        function_ast = ast.FunctionDef(name='wrapped_function', args=ast.arguments(posonlyargs=[], args=[], kwonlyargs=[], kw_defaults=[], defaults=[]), body=all_other_stmts, decorator_list=[], lineno=-1)
        main_code = import_string + '\n' + ast.unparse(import_stmts) + '\n' + ast.unparse(function_ast)
        return main_code
    except Exception as e:
        return code

import_string = 'from string import *\nfrom re import *\nfrom datetime import *\nfrom collections import *\nfrom heapq import *\nfrom bisect import *\nfrom copy import *\nfrom math import *\nfrom random import *\nfrom statistics import *\nfrom itertools import *\nfrom functools import *\nfrom operator import *\nfrom io import *\nfrom sys import *\nfrom json import *\nfrom builtins import *\nfrom typing import *\nimport string\nimport re\nimport datetime\nimport collections\nimport heapq\nimport bisect\nimport copy\nimport math\nimport random\nimport statistics\nimport itertools\nimport functools\nimport operator\nimport io\nimport sys\nimport json\nsys.setrecursionlimit(50000)\n'

# Simple test code that includes imports and other statements
test_code = """
import math
from collections import defaultdict

x = 42
y = math.sqrt(x)
result = y + 1
"""

ARGS = {
    'code': test_code
}
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
parse_assert_statement Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/compute_test_output_prediction_metrics.py
code served (permissive licence) · get_code("ae8f42636ac09e82")
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 ast

# Test case 1: Valid assert statement with equality comparison
statement_1 = "assert x == 5"

# Test case 2: Valid assert statement with string comparison
statement_2 = "assert result == 'hello'"

# Test case 3: Valid assert statement with complex expression
statement_3 = "assert func(a, b) == [1, 2, 3]"

ARGS = {
    "statement": "assert output == 42"
}
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
pass_at_k Ran
ran, honoured the contract we drafted
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/compute_code_execution_metrics.py
code served (permissive licence) · get_code("3a7c7ec16769928a")
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) and returned float64, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Analyzing the function:
# - n: appears to be an integer (used in np.arange and arithmetic)
# - c: appears to be an integer (used in arithmetic)
# - k: appears to be an integer or array-like (used in division with np.arange result)
# The function computes pass@k metric, commonly used in code generation evaluation
# where n is total samples, c is correct samples, k is the cutoff

# Valid test case: n=10 total samples, c=3 correct, k=2 cutoff
ARGS = {
    "n": 10,
    "c": 3,
    "k": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float, 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
truncatefn Ran
ran; our contract draft was wrong, not the code
LiveCodeBench/LiveCodeBench/lcb_runner/evaluation/testing_util.py
code served (permissive licence) · get_code("f23d0f0bf67cc7cb")
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 truncatefn(s, length=300):
    if isinstance(s, str):
        pass
    else:
        s = str(s)
    if len(s) <= length:
        return s
    return s[:length // 2] + '...(truncated) ...' + s[-length // 2:]

torch.manual_seed(0)

# The function takes a string (or something convertible to string) and truncates it
# s: can be a string or any object convertible to string
# length: integer specifying the max length before truncation

ARGS = {
    "s": "This is a test string that will be used to verify the truncatefn function works correctly with various inputs.",
    "length": 300
}
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

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 1 the same month, 2 later) and 5 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
pass_at_k 6 other papers
5 dated earlier, 1 later
2206.03865 (2022-06)
Fault-Aware Neural Code Rankers
2306.10763 (2023-06)
Guiding Language Models of Code with Global Context using Monitors
2402.09497 (2024-02)
Instruction Tuning for Secure Code Generation
and 3 more not named here
estimate_pass_at_k 2 other papers
1 dated earlier, 1 later
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language Models
2604.04356 (2026-04)
REAM: Merging Improves Pruning of Experts in LLMs
truncatefn 2 other papers
1 dated earlier, 1 the same month
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language Models
2602.03719 (2026-02)
BranPO: Scalable Contrastive Branch Sampling for Long-Horizon Agentic Reinforcement Learning
clean_if_name 1 other paper
1 dated earlier
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language Models
compute_metrics_from_results 1 other paper
1 dated earlier
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language Models
extract_instance_results 1 other paper
1 dated earlier
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language Models
make_function 1 other paper
1 dated earlier
2309.00071 (2023-09)
YaRN: Efficient Context Window Extension of Large Language 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

Language Reasoning Models (LRMs) achieve strong performance by scaling testtime computation but often suffer from "overthinking", producing excessively long reasoning traces that increase latency and memory usage. Existing LRMs typically enforce conciseness with uniform length penalties, which over-compress crucial early deduction steps at the sequence level and indiscriminately penalize all queries at the group level. To solve these limitations, we propose PACE, a dual-level framework for prefix-protected and difficulty-aware compression under hierarchical supervision. At the sequence level, prefix-protected optimization employs decaying mixed rollouts to maintain valid reasoning paths while promoting conciseness. At the group level, difficulty-aware penalty dynamically scales length constraints based on query complexity, maintaining exploration for harder questions while curbing redundancy on easier ones. Extensive experiments on DeepSeek-R1-Distill-Qwen (1.5B/7B) demonstrate that PACE achieves a substantial reduction in token usage (up to 55.7%) while simultaneously improving accuracy (up to 4.1%) on math benchmarks, with generalization ability to code, science, and general domains ‡ .

For agents

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

get_harvested_code_for_paper("2602.11639")
get_code_for_paper("2602.11639")
have("2602.11639")

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

curl https://syntology.ai/api/ran/2602.11639.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.11639.svg)](https://syntology.ai/paper/2602.11639)

Connect an agent — have() is free.