SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2601.18533 · 2026

From Verifiable Dot to Reward Chain: Harnessing Verifiable Reference-based Rewards for Reinforcement Learning of Open-ended Generation

Lifeng Shang, Yufei Wang, Chaofan Tao, Yuxin Jiang, Xingshan Zeng, Liangyou Li, Haoli Bai, Jierun Chen, Qiyuan Zhang, 𝑲 𝒛

arXiv · PDF · Open in the Atlas

Code that ran

5 of 5 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, 0 violated it and 0 have no contract verdict; 1 failure 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/2601.18533.json · badge: /api/ran/2601.18533.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
YJiangcm/RLVRR community
Papers with Code archive
5 ran · 0 no recorded run
FunctionStatusWhere it lives
find_matched_keep_order_num Ran
ran; our contract draft was wrong, not the code
YJiangcm/RLVRR/openrlhf/reward_tools/reward_fn.py
code served (permissive licence) · get_code("1bfcdc1accc5eea6")
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, 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 re

# Analyzing the function:
# - keywords: a list/iterable of strings that will be used in a regex pattern
# - s: a string to search in
# The function finds all keywords in the string (case-insensitive, word-boundary aware)
# and returns them in lowercase, in order of appearance

# Setup test data
keywords = ["hello", "world", "python", "test"]
s = "Hello there! This is a Python test. Hello WORLD, python is great!"

ARGS = {
    "keywords": keywords,
    "s": s
}
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 list, ndim 1, dtype family float — a list 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
keyword_reward_func Ran
ran, honoured the contract we drafted
YJiangcm/RLVRR/openrlhf/reward_tools/reward_fn.py
code served (permissive licence) · get_code("205e0888cd89b84e")
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, 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 re
import ast
import numpy as np

# Analyze the function signature and body
# keyword_reward_func takes:
# - ai_answer: string (used with split() and passed to find_matched_keep_order_num)
# - keypoint_keywords: can be string or list (parsed with ast.literal_eval if string)
# - reference_answer: string (used with split() and passed to find_matched_keep_order_num)

# keypoint_keywords structure from function logic:
# - It's iterable (for multi_ref_keywords in keypoint_keywords)
# - Each multi_ref_keywords is iterable (for keywords in multi_ref_keywords)
# - Each keywords is passed to find_matched_keep_order_num which expects
#   a list/iterable of phrases to search for

# Build valid arguments:

# Simple keyword lists for testing
keyword_list_1 = ["algorithm", "data"]
keyword_list_2 = ["structure", "complexity"]

# keypoint_keywords structure: list of (list of keyword lists)
keypoint_keywords = [
    [keyword_list_1, keyword_list_2],
    [["performance", "optimization"]],
]

# Sample AI answer with some keywords
ai_answer = "The algorithm uses a hash data structure for efficient lookup with good performance characteristics."

# Sample reference answer
reference_answer = "We analyze the algorithm and data structure, checking complexity and performance optimization."

ARGS = {
    "ai_answer": ai_answer,
    "keypoint_keywords": keypoint_keywords,
    "reference_answer": reference_answer,
}
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 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
lcs_length Ran
ran, honoured the contract we drafted
YJiangcm/RLVRR/openrlhf/reward_tools/reward_fn.py
code served (permissive licence) · get_code("30cd2f1030b42901")
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

ARGS = {
    "seq1": "hello",
    "seq2": "hallo"
}
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 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
reward_func Ran
ran, honoured the contract we drafted
YJiangcm/RLVRR/openrlhf/reward_tools/reward_fn.py
code served (permissive licence) · get_code("6e9c69dc6dfeb4ae")
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 Tensor, ndim 1, dtype family float, shape (2,) 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
import re
import json
import ast

torch.manual_seed(0)

# The function reward_func expects:
# - queries: list of strings (full text including prompt)
# - prompts: list of strings (prompt text)
# - labels: list of strings (ground truth labels with specific format)

# The label format is: "reference_answer[[[SPLIT]]]keypoint_keywords[[[SPLIT]]]style_json"
# where keypoint_keywords is a string representation of a nested list structure
# and style_json is a JSON string with key_points containing verification_code and weight

# Build minimal style validation code
style_validation_code = """
def validate_style(response):
    return len(response) > 0
"""

# Build the style JSON structure
style_json = json.dumps({
    "key_points": [
        {
            "verification_code": style_validation_code,
            "weight": 0.5
        }
    ]
})

# Build keypoint keywords structure - nested list of lists
keypoint_keywords = [
    [
        ["important", "key", "main"],
        ["critical", "essential"]
    ],
    [
        ["example", "instance"],
        ["case", "scenario"]
    ]
]

# Convert to string representation
keypoint_keywords_str = str(keypoint_keywords)

# Build reference answer with some keywords
reference_answer = "The important key main point is that critical essential concepts matter. For example instance, this case scenario demonstrates the principle."

# Build label with all three parts separated by [[[SPLIT]]]
label = f"{reference_answer}[[[SPLIT]]]{keypoint_keywords_str}[[[SPLIT]]]{style_json}"

# Build queries and prompts
prompt1 = "What is the main concept? "
response1 = "The important key main point is critical and essential. For example, this case demonstrates it."
query1 = prompt1 + response1

prompt2 = "Explain the scenario. "
response2 = "The key important main idea is that critical essential factors apply. For instance, this scenario shows the example."
query2 = prompt2 + response2

queries = [query1, query2]
prompts = [prompt1, prompt2]
labels = [label, label]

ARGS = {
    "queries": queries,
    "prompts": prompts,
    "labels": labels
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) 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
style_reward_func Ran
ran, honoured the contract we drafted
YJiangcm/RLVRR/openrlhf/reward_tools/reward_fn.py
code served (permissive licence) · get_code("484f597dab4695dc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re
import json
import ast

ARGS = {
    'answer': 'The implementation uses efficient algorithms and follows best practices for code style.',
    'json_str': '''{
        "key_points": [
            {
                "verification_code": "def check_efficiency(response):\\n    return 'efficient' in response.lower()",
                "weight": 0.5
            },
            {
                "verification_code": "def check_practices(response):\\n    return 'best practices' in response.lower()",
                "weight": 0.5
            }
        ]
    }'''
}
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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 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 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.

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

Reinforcement learning with verifiable rewards (RLVR) succeeds in reasoning tasks (e.g., math and code) by checking the final verifiable answer (i.e., a verifiable dot signal). However, extending this paradigm to open-ended generation is challenging because there is no unambiguous ground truth. Relying on single-dot supervision often leads to inefficiency and reward hacking. To address these issues, we propose reinforcement learning with verifiable reference-based rewards (RLVRR). Instead of checking the final answer, RLVRR extracts an ordered linguistic signal from high-quality references (i.e, reward chain). Specifically, RLVRR decomposes rewards into two dimensions: content, which preserves deterministic core concepts (e.g., keywords), and style, which evaluates adherence to stylistic properties through LLM-based verification. In this way, RLVRR combines the exploratory strength of RL with the efficiency and reliability of supervised fine-tuning (SFT). Extensive experiments on more than 10 benchmarks with Qwen and Llama models confirm the advantages of our approach. RLVRR (1) substantially outperforms SFT trained with ten times more data and advanced reward models, (2) unifies the training of structured reasoning and open-ended generation, and (3) generalizes more effectively while preserving output diversity. These results establish RLVRR as a principled and efficient path toward verifiable reinforcement learning for general-purpose LLM alignment. We release our code and data at https://github.com/YJiangcm/RLVRR.

For agents

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

get_harvested_code_for_paper("2601.18533")
get_code_for_paper("2601.18533")
have("2601.18533")

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

curl https://syntology.ai/api/ran/2601.18533.json

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

[![Syntology run record](https://syntology.ai/api/ran/2601.18533.svg)](https://syntology.ai/paper/2601.18533)

Connect an agent — have() is free.