SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.04838 · EMNLP · 2024

Rationale-Aware Answer Verification by Pairwise Self-Evaluation

Saku Sugawara, Akira Kawabata

arXiv · PDF · Open in the Atlas

Code that ran

6 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 6 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2410.04838.json · badge: /api/ran/2410.04838.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
akirakawabata/reps canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
deep_merge Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/utils/config.py
code served (permissive licence) · get_code("1cc4c0a5bee7d783")
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 torch
import numpy as np

torch.manual_seed(0)

# Construct arguments for deep_merge function
# The function takes two dictionaries: base and update
# It recursively merges update into base, with special handling for nested dicts

# Create a base dictionary with nested structure
base_dict = {
    "model_config": {
        "hidden_size": 256,
        "num_layers": 3,
    },
    "training": {
        "lr": 0.001,
        "batch_size": 32,
    },
    "name": "baseline",
}

# Create an update dictionary that will be merged in
update_dict = {
    "model_config": {
        "hidden_size": 512,  # This will override
        "dropout": 0.1,       # This will be added
    },
    "training": {
        "lr": 0.0001,         # This will override
    },
    "version": "v2",          # This will be added at top level
}

ARGS = {
    "base": base_dict,
    "update": update_dict,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_answer Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/reward_training.py
code served (permissive licence) · get_code("747f8df0daf7cfe2")
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

# Construct a valid explanation dict with an 'answer' key
exp = {
    'answer': '  THE ANSWER  ',
    'other_field': 'some value'
}

# Define ARGS mapping every parameter to a value
ARGS = {
    'exp': exp
}
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
format_choices Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/reward_training.py
code served (permissive licence) · get_code("ce94d4b0057e82ea")
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
from typing import List, Dict

torch.manual_seed(0)

# The function expects a dict with 'label' and 'text' keys,
# each mapping to a list of strings
ARGS = {
    'choices': {
        'label': ['A', 'B', 'C'],
        'text': ['option one', 'option two', 'option three']
    }
}
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
load_config Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/utils/config.py
code served (permissive licence) · get_code("4a37023f73009457")
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 tempfile
import yaml
from pathlib import Path

# Create temporary directory and config files for testing
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name)

# Create main config file
main_config = {
    'dataset': 'mnist',
    'model': {
        'type': 'cnn',
        'layers': 3
    },
    'training': {
        'epochs': 10,
        'batch_size': 32
    }
}

main_config_path = temp_path / 'config.yaml'
with open(main_config_path, 'w', encoding='utf-8') as f:
    yaml.dump(main_config, f)

# Create datasets subdirectory and dataset config file
datasets_dir = temp_path / 'datasets'
datasets_dir.mkdir(exist_ok=True)

dataset_config = {
    'model': {
        'layers': 5
    },
    'data': {
        'path': '/data/mnist',
        'train_size': 60000
    }
}

dataset_config_path = datasets_dir / 'mnist.yaml'
with open(dataset_config_path, 'w', encoding='utf-8') as f:
    yaml.dump(dataset_config, f)

ARGS = {
    'config_path': str(main_config_path)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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
load_jsonl Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/utils/io.py
code served (permissive licence) · get_code("599ccf65a05b3aa1")
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 json
from pathlib import Path
import tempfile
import os

# Create a temporary JSONL file for testing
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.jsonl")

# Write sample JSONL data
with open(temp_file, 'w', encoding='utf-8') as f:
    f.write(json.dumps({"id": 1, "text": "hello"}) + "\n")
    f.write(json.dumps({"id": 2, "text": "world"}) + "\n")
    f.write(json.dumps({"id": 3, "data": [1, 2, 3]}) + "\n")

ARGS = {
    "file_path": 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 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
load_qa_dataset Ran
ran; our contract draft was wrong, not the code
akirakawabata/reps/src/cot_generation.py
code served (permissive licence) · get_code("2a63d05f13d1c328")
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 json
from pathlib import Path
from typing import List, Dict
import tempfile
import os

# Create a temporary directory for our stand-in files
temp_dir = tempfile.mkdtemp()

# Create a mock StrategyQA file
strategyqa_file = os.path.join(temp_dir, "train.json")
strategyqa_data = [
    {"qid": "q1", "question": "Is the sky blue?", "answer": True},
    {"qid": "q2", "question": "Do cats fly?", "answer": False},
    {"qid": "q3", "question": "Is water wet?", "answer": True},
]
with open(strategyqa_file, 'w', encoding='utf-8') as f:
    json.dump(strategyqa_data, f)

# Define the config dictionary
config = {
    'cot_generation': {
        'few_shot_samples_ids': [],
        'strategyqa_file': strategyqa_file
    }
}

# Define ARGS for different dataset options
# We'll use 'strategy_qa' since it doesn't require HuggingFace dataset downloads
ARGS = {
    'dataset_name': 'strategy_qa',
    'config': config
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets — 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
prepare_training_data Not yet run
unverified (harvested, no recorded run)
akirakawabata/reps/src/reward_training.py
code served (permissive licence) · get_code("b05eebfb3a5a380f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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

Answer verification identifies correct solutions among candidates generated by large language models (LLMs). Current approaches typically train verifier models by labeling solutions as correct or incorrect based solely on whether the final answer matches the gold answer. However, this approach neglects any flawed rationale in the solution yielding the correct answer, undermining the verifier's ability to distinguish between sound and flawed rationales. We empirically show that in StrategyQA, only 19% of LLM-generated solutions with correct answers have valid rationales. Furthermore, we demonstrate that training a verifier on valid rationales significantly improves its ability to distinguish valid and flawed rationales. To make a better verifier without extra human supervision, we introduce REPS (Rationale Enhancement through Pairwise Selection), a method for selecting valid rationales from candidates by iteratively applying pairwise self-evaluation using the same LLM that generates the solutions. Verifiers trained on solutions selected by REPS outperform those trained using conventional training methods on three reasoning benchmarks (ARC-Challenge, DROP, and StrategyQA). Our results suggest that training reliable verifiers requires ensuring the validity of rationales in addition to the correctness of the final answers, which would be critical for models assisting humans in solving complex reasoning tasks.

For agents

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

get_harvested_code_for_paper("2410.04838")
get_code_for_paper("2410.04838")
have("2410.04838")

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

curl https://syntology.ai/api/ran/2410.04838.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.04838.svg)](https://syntology.ai/paper/2410.04838)

Connect an agent — have() is free.