SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.16331 · 2026

Decoding the Critique Mechanism in Large Reasoning Models

Hung Le, Heng Ji, Hoang Phan, Khoa Doan, Quang Nguyen, Xiusi Chen

arXiv · PDF · Open in the Atlas

Code that ran

1 of 1 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; 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/2603.16331.json · badge: /api/ran/2603.16331.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
mail-research/lrm-critique-vectors community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
get_activations Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mail-research/lrm-critique-vectors/src/activation/activation_extraction.py
code served (permissive licence) · get_code("c83e792b55372de4")
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 dict. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch.nn as nn
from typing import Dict, List, Optional, Tuple

torch.manual_seed(0)

# Create a minimal stand-in tokenizer
class MinimalTokenizer:
    def __call__(self, text, return_tensors=None, add_special_tokens=False):
        # Return a minimal encoding with input_ids
        # For simplicity, create a tensor of token ids
        token_ids = torch.tensor([[1, 2, 3, 4, 5]], dtype=torch.long)
        return {'input_ids': token_ids}

# Create a minimal stand-in config
class MinimalConfig:
    def __init__(self):
        self.num_hidden_layers = 2

# Create a minimal pretrained model
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.config = MinimalConfig()
        # Create embeddings and hidden states
        self.embedding = nn.Embedding(100, 4)
        self.linear = nn.Linear(4, 4)
        
    def forward(self, input_ids, output_hidden_states=False, use_cache=False):
        # Embed the input
        x = self.embedding(input_ids)  # [batch, seq_len, 4]
        
        # Create hidden states for each layer (including embedding layer at index 0)
        hidden_states = [x]  # hidden_states[0] is embedding
        
        # Generate hidden states for each transformer layer
        for _ in range(self.config.num_hidden_layers):
            x = self.linear(x)
            hidden_states.append(x)
        
        # Create a minimal output object
        class Output:
            pass
        
        output = Output()
        output.hidden_states = hidden_states
        return output

# Instantiate the model and tokenizer
model = MinimalModel()
tokenizer = MinimalTokenizer()

# Define the arguments
ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'texts': ['hello world', 'foo bar'],
    'layers': [0, 1],
    'spans': [(0, 2), (1, 3)]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch, transformers — 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

Also present in other papers' code

Of 1 samples lifted from this paper's linked repositories, 1 carry a function identity and 0 do not. Of the 1 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 1 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

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 Reasoning Models (LRMs) exhibit backtracking and self-verification mechanisms that enable them to revise intermediate steps and reach correct solutions, yielding strong performance on complex logical benchmarks. We hypothesize that such behaviors are beneficial only when the model has sufficiently strong "critique" ability to detect its own mistakes. This work systematically investigates how current LRMs recover from errors by inserting arithmetic mistakes in their intermediate reasoning steps. Notably, we discover a peculiar yet important phenomenon: despite the error propagating throughout the entire chainof-thought (CoT) without any verbalized correction, the model still reaches the correct final answer after the thinking process finishes. This recovery implies the existence of an internal mechanism helping the model to detect errors and trigger self-correction, which we refer to as the hidden critique ability. Building on feature space analysis, we identify a highly interpretable critique vector representing this behavior. Extensive experiments across multiple model scales and families demonstrate that steering latent representations with this vector improves the model's error detection capability and enhances the performance of test-time scaling at no extra training cost. Our findings provide a valuable understanding of LRMs' critique behavior, suggesting a promising direction to control and improve their self-verification mechanism. Our code is available at: https://github.com/mail-research/lrm-critique-vectors.

For agents

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

get_harvested_code_for_paper("2603.16331")
get_code_for_paper("2603.16331")
have("2603.16331")

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

curl https://syntology.ai/api/ran/2603.16331.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.16331.svg)](https://syntology.ai/paper/2603.16331)

Connect an agent — have() is free.