Decentralized Blockchain-based Robust Multi-agent Multi-armed Bandit

I have several concerns:

1. The first main concern is the writing of this paper. The introduction section is excessively long, but still does not introduce the insight of the problem clearly. In many places, the author lacks examples or explanations. For example, in section 3, the author states a lengthy method, but does not provide intuition or insight, making it easy for people to get lost in the details and not understand what the sub-technique is used for.
2. No numerical results. This paper has no numerical results at all, making it difficult to intuitively assess the effectiveness of the BC-UCB algorithm. In fact, there are a series of works considering the robustness of the algorithms against Byzantine attacks. The authors can consider using them as baselines for experiments.

• Regarding its readability, the paper needs to be significantly streamlined. For instance, the paper spends 2.5 pages on various topics and only lines 151-158 to describe its contributions. I found it very hard to read and understand. Also, I think that paper ends abruptly at the page limit and then simply continues in the appendix (the conclusions are in page 15). Other examples include the caption of Figure 1 which almost overlaps with the text and the storm of references (especially in the first part) which don't really provide a context for the paper but rather confuse the reader.
• Regarding its rigor, there seem to be undefined abbreviations (e.g., UCB), redundant notations, and questionable assumptions, e.g., in line 169 why is B=T and how can we assume that the number of blocks is fixed (and not growing) and each block is valid with some probability? This is an uncommon representation of a blockchain (history). Other examples include Theorem 1 which mentions "existential forgery, and universal composability" but which are only defined much later in the appendix (same for the other theorems). This significantly interrupts the flow.
• The paper mingles several notions, federated learning, blockchains, multi-arm bandits, multi-agent learning etc and to allow for a reasonable assessment, it needs to make a stronger effort to both clarify this combination and to draw clear connections to existing literature and existing real-world (or at least theoretical) problems.
• After reading the problem formulation a couple of times, I still struggle to understand what is the context: what are multi-agent MABs on blockchains? Is there a practical scenario that this setting captures?

• The motivations and literature review for robust multi-agent MAB problem and blockchains as well as their incorporation are not sufficient and hard to follow. There should be a section dedicated to related work for these respective topics, instead of a 3-page introduction with no headings.
• Description and motivations on the considered types of attacks (e.g., Byzantine) from participants are not clear. This information is particularly important given that the proposed framework relies on the assumptions about the structure of malicious behaviors as stated by the authors.
• As a result of the previous two points, the practicality and implications of the proposed framework are not clear.
• Missing citations throughout the paper: Lines 164-165: standard notations for traditional MAB and multi-agent MAB Line 224: "... existing work on Byzantine-resilient multi-agent MAB" Line 230: "This assumption is quite common in blockchain works."

• The authors could better motivate the problem. The described applications/examples do not feel like a good fit. E.g. in many cases the other agents are merely self-interested.
• The paper does not include any experimental/numerical results.
• The paper does not seem to offer any insights that could be of independent technical interest either from MAB or blockchain side but feels more like a collage of largely pre-existing techniques. Although this is not a deal breaker, it makes the lack of sufficient motivation even more pressing as a problem.