The Cognitive Penalty: Ablating System 1 and System 2 Reasoning in Edge-Native SLMs for Decentralized Consensus
Summary
Empirical study shows small language models achieve 100% adversarial robustness with System 1 intuition but collapse under System 2 reasoning when used as edge-native governance firewalls in decentralized autonomous organizations.
View Cached Full Text
Cached at: 04/22/26, 10:35 AM
Paper page - The Cognitive Penalty: Ablating System 1 and System 2 Reasoning in Edge-Native SLMs for Decentralized Consensus
Source: https://huggingface.co/papers/2604.16913
Abstract
Small Language Models deployed in decentralized autonomous organizations face challenges when using inference-time computation for adversarial governance scenarios, with simpler models showing better robustness and efficiency.
Decentralized Autonomous Organizations(DAOs) are inclined exploreSmall Language Models(SLMs) as edge-native constitutional firewalls to vet proposals and mitigate semantic social engineering. While scalinginference-time compute(System 2) enhances formal logic, its efficacy in highly adversarial, cryptoeconomic governance environments remains underexplored. To address this, we introduce Sentinel-Bench, an 840-inference empirical framework executing a strict intra-model ablation on Qwen-3.5-9B. By togglinglatent reasoningacross frozen weights, we isolate the impact ofinference-time computeagainst anadversarial Optimism DAO dataset. Our findings reveal a severe compute-accuracy inversion. Theautoregressive baseline(System 1) achieved 100% adversarial robustness, 100% juridical consistency, and state finality in under 13 seconds. Conversely,System 2reasoning introduced catastrophic instability, fundamentally driven by a 26.7%Reasoning Non-Convergence(cognitive collapse) rate. This collapse degraded trial-to-trial consensus stability to 72.6% and imposed a 17x latency overhead, introducing critical vulnerabilities toGovernance Extractable Value(GEV) and hardware centralization. While rare (1.5% of adversarial trials), we empirically captured “Reasoning-Induced Sycophancy,” where the model generated significantly longer internal monologues (averaging 25,750 characters) to rationalize failing the adversarial trap. We conclude that for edge-native SLMs operating underByzantine Fault Tolerance(BFT) constraints,System 1parameterized intuitionis structurally and economically superior toSystem 2iterative deliberationfor decentralized consensus. Code and Dataset: https://github.com/smarizvi110/sentinel-bench
View arXiv pageView PDFGitHub1Add to collection
Get this paper in your agent:
hf papers read 2604\.16913
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2604.16913 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2604.16913 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2604.16913 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Hidden Language Consistency Phenomena in Reasoning LLMs
This paper studies multilingual reasoning models and reveals that language consistency in outputs can degrade or collapse with task difficulty, especially for less represented languages. It argues that evaluating multilingual capability requires jointly considering accuracy, language consistency, and task difficulty.
Consensus is Strategically Insufficient: Reasoning-Trace Disagreement as a Knowledge-Representation Signal
This paper argues that consensus-seeking in multi-agent LLM systems is insufficient for value-laden tasks, proposing a knowledge-representation layer that classifies agent reasoning-trace disagreements into four symbolic states to enable strategic routing in systems like content moderation.
How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures
This paper characterizes two distinct processes by which language models fail in reasoning—committed failure and persistent uncertainty—using token-level uncertainty signals, and demonstrates implications for self-consistency and failure detection strategies.
Context, Reasoning, and Hierarchy: A Cost-Performance Study of Compound LLM Agent Design in an Adversarial POMDP
A controlled study of compound LLM agent design in an adversarial POMDP (CybORG CAGE-2), systematically varying context, reasoning, and hierarchy across five model families. Key findings: programmatic state abstraction yields large returns per token, hierarchy without deliberation tools achieves best absolute performance, and context engineering is more cost-effective than deeper reasoning.
Reasoning Can Be Restored by Correcting a Few Decision Tokens
This paper shows that the reasoning gap between base LLMs and large reasoning models is concentrated on a small set of early planning tokens. It introduces disagreement-guided token intervention, where replacing only those critical tokens with a reasoning model's outputs allows a base model to nearly match the reasoning model's performance.