{"id":"d3c9a7b3-07e3-49f2-84c6-5c4350245894","arxiv_id":"2508.17751","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The submitted text is two different papers: the abstract's MANGO claims about hierarchical RL are unsupported by a body about CLIP bias transfer.","lead":"This submission pairs an abstract for MANGO, a hierarchical reinforcement learning framework with nested options, with a full text that is an unrelated paper on social bias transfer in CLIP models. The abstract's claims about sample efficiency and generalization therefore have no supporting content in the manuscript as submitted.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Submission body is a different paper; MANGO claim has no supporting evidence.","rationale":"The reader's verdict correctly identifies the weakest assumption: the body text is the MANGO paper. My analysis agrees. The central claim cannot be evaluated because the supporting algorithm and experiments are absent. No amount of speculation about the method's internal consistency can overcome this. The concern is load-bearing because it invalidates every subsequent step of review. I do not find an additional scientific flaw because there is no scientific content to examine. Given this, the reader's UNVERDICTED status is appropriate. A REJECT would be too strong, as the underlying paper might exist and be sound; an ACCEPT is impossible without the text. Therefore I recommend keeping the verdict UNCHANGED. The proposed test—comparing the submission to the official arXiv PDF—will definitively resolve whether the submission is a corrupted upload or an authentic mismatch.","tokens_in":103,"tokens_out":2747,"duration_ms":32427,"concrete_test":"Download the official PDF for arXiv:2508.17751 from arXiv and compare its content with the submitted full text. Specifically, check the title, author list, and first sections. If the official PDF matches the CLIP-bias paper provided here, the MANGO abstract is orphaned; if it instead contains the MANGO manuscript, the mismatch is a submission error and the actual paper should be re-reviewed. Also query arXiv's API for the abstract and confirm whether it matches the MANGO abstract.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that MANGO's nested-option architecture improves sample efficiency and generalization in grid-world tasks—appears only in the abstract. The body text is an entirely different paper: 'From Global to Local: Social Bias Transfer in CLIP' by Ramos et al., with its own title, author list, and footer 'arXiv:2508.17750v1 [cs.CV]'. None of the promised content—the multi-layer abstraction formalism, intra-layer policies, task actions, or the procedurally-generated grid experiments—is present in the submitted manuscript. Consequently, every assertion in the abstract is unsupported by any derivable evidence within the document. This is not a scientific error in the proposed method (which cannot be evaluated) but a structural failure: the submission is not self-consistent, and the manuscript as provided cannot be reviewed for correctness, novelty, or reproducibility. The load-bearing premise—that the text corresponds to the MANGO paper—is false on its face.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript is titled \"Multi-layer Abstraction for Nested Generation of Options (MANGO) in Hierarchical Reinforcement Learning\" and its abstract claims a new HRL framework with nested options, intra-layer policies, task actions, and experiments in procedurally-generated grid environments that improve sample efficiency and generalization. However, the full text is an entirely different paper: \"From Global to Local: Social Bias Transfer in CLIP\" by Ramos et al., with its own abstract, introduction, references, and page footer reading arXiv:2508.17750v1 [cs.CV]. None of the promised MANGO formalism, algorithm definitions, environment descriptions, baseline comparisons, or experimental results appears anywhere in the body. As submitted, the document is not self-consistent and cannot support the abstract's claims.","tokens_in":5326,"tokens_out":1701,"duration_ms":21819,"significance":"If MANGO were actually specified and evaluated as described, the claimed improvements in sample efficiency and generalization for hierarchical RL would be a useful contribution to the field, particularly for sparse-reward and safety-critical applications. However, the submission provides no content that can be assessed for correctness, novelty, or reproducibility. The abstract alone is not a scientific paper, and the body text addresses a disjoint topic. The significance of the work therefore cannot be evaluated on the evidence presented.","major_comments":[{"comment":"The body of the submission is not the MANGO paper. The title, author list, abstract, introduction, and references all correspond to \"From Global to Local: Social Bias Transfer in CLIP\" (Ramos et al.), and the page footer reads arXiv:2508.17750v1 [cs.CV]. None of the MANGO framework—multi-layer abstraction, nested options, intra-layer policies, task actions—is defined in the text, and no grid-environment experiments, baselines, or result tables are reported. The central claim of the abstract is therefore entirely unsupported by the submitted manuscript.","section":"Full text (entire body)"},{"comment":"Even taken on its own terms, the body text is a paper about bias transfer in CLIP models, with no connection to hierarchical reinforcement learning or option-based methods. There is no equation, algorithm, environment definition, or experimental protocol for MANGO anywhere in the document. The submission is internally inconsistent at the most basic level, making it impossible to review the proposed method for scientific soundness.","section":"Abstract vs. body consistency"},{"comment":"The abstract claims \"substantial improvements in both sample efficiency and generalization capabilities compared to standard RL methods.\" No experimental setup, hyperparameters, environment generator, or code is provided. Even if the body text were the correct MANGO paper, the absence of any empirical artifact would prevent verification of the claimed results. As it stands, the claim is not merely unverified but unverifiable from the submitted text.","section":"Reproducibility and empirical support"}],"minor_comments":[{"comment":"The page footer lists arXiv:2508.17750v1 [cs.CV], which is inconsistent with the manuscript number 2508.17751 stated in the submission title. This should be corrected if the wrong file was uploaded; otherwise it signals a metadata mismatch that needs clarification.","section":"Page footer / metadata"}],"recommendation":"reject","confidential_remarks":"The manuscript as received appears to be an upload or metadata error: the abstract describes one paper and the body is a completely different paper. Under the review rule that the full text is in-scope evidence, the submission cannot be evaluated as the MANGO paper. I recommend desk rejection without further technical review. If the intended submission is the CLIP bias-transfer paper, it should be resubmitted under the appropriate venue with consistent metadata."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here is the quick take: the submission is not a coherent paper. The abstract describes MANGO, a hierarchical RL framework with nested options and claimed sample-efficiency gains in grid environments. The body is a completely different paper, \"From Global to Local: Social Bias Transfer in CLIP,\" with different authors and an arXiv footer pointing to 2508.17750v1. No algorithm, equations, experiments, or results for MANGO appear anywhere. There is nothing to review.\n\nWhat is worth saying in the abstract's favor: the MANGO idea—multiple abstraction layers with nested macro-actions and intra-layer policies—is a reasonable extension of the options framework. It is not field-reshaping, but if it genuinely delivered substantial gains in sparse-reward tasks, it would be a useful contribution to the HRL literature. The CLIP paper in the body might also be a legitimate empirical study, but it is not the paper advertised, and I am not going to credibly assess it here.\n\nThe soft spots are not subtle. The load-bearing claim of the abstract has zero supporting content in the manuscript. This is not a case of missing appendices or a weak related-work section; it is a structural mismatch that makes review impossible. Even the abstract alone would need more detail: no formal definition of the layers, no explanation of how options are nested, no baseline description, no table of results. But the mismatch is disqualifying on its own.\n\nMy recommendation: desk reject as submitted. This is not a comment on the underlying science—the MANGO approach might be fine, or the CLIP paper might be fine—but the submission is not self-consistent. The right move is to return it to the authors to correct the abstract or the body, and require a complete manuscript that actually presents the claimed work. If a proper MANGO manuscript is submitted later, it should get a fair review. As it stands, there is no way to evaluate novelty, soundness, or reproducibility.","headline":"The abstract and body are two different papers; the MANGO claims have no supporting content, so this is a desk reject, not a scientific evaluation.","tokens_in":5870,"tokens_out":2090,"would_cite":false,"duration_ms":24621,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The MANGO abstract claims a nested-options HRL framework, but the submitted body is a different paper on CLIP bias transfer.","keywords":["hierarchical reinforcement learning","multi-layer abstraction","options framework","nested macro-actions","sparse reward tasks","sample efficiency","generalization","manuscript mismatch"],"falsifier":"A concrete check is to inspect the supplied PDF's title page and footer: if the title and identifier there are not the same as those in the abstract, the MANGO claims have no supporting text. A second check is to search the body for the term MANGO; its absence outside the abstract would confirm that no algorithm or experiment is presented.","tokens_in":5027,"feed_emoji":"⚠️","tokens_out":4320,"duration_ms":52508,"temperature":0.7,"pith_summary":"The submission's abstract announces MANGO, a hierarchical reinforcement learning framework that stacks abstraction layers, nests options as macro-actions, and promises better sample efficiency and generalization on grid-world tasks. The body of the submitted text, however, is not the MANGO paper: it is a separate article on how social biases transfer from CLIP models to downstream tasks, with a different title, a different author list, and a footer that does not match the abstract's identifier. This means the manuscript, as presented, contains no algorithm, equations, or experiments that would support the MANGO claims. The one coherent claim that can be read in the full text is the CLIP bias-transfer article's empirical finding that bias measurement depends on the data subset considered.","feed_headline":"MANGO abstract promises RL options; body is a different paper","feed_subtitle":"The HRL framework's experiments and algorithm are absent from the supplied text.","key_machinery":"For the intended MANGO framework, the central object is a multi-layer options hierarchy: each layer defines an abstract state space, options serve as macro-actions, intra-layer policies guide transitions within the abstract space, and task actions carry task-specific reward components. This machinery is supposed to enable nesting and reuse of learned movement primitives across layers. In the submitted body text, the corresponding machinery is instead the CLIP bias-analysis setup: measuring pre-training bias on global and local subsets of data, then computing correlations between that bias and downstream bias in tasks such as VQA and captioning. No equations or training procedure for the inte","core_discovery":"The paper intended under this identifier aims to establish that decomposing a reinforcement learning task into multiple abstraction layers, where each layer defines an abstract state space and generates nested options, allows an agent to reuse learned macro-actions and thereby improve sample efficiency, generalization, and interpretability. The author's position would be that this nested-option hierarchy outperforms standard RL in procedurally generated sparse-reward grid environments. That position, however, is not demonstrated in the supplied body text: the body consists of a different manuscript, so the central claim is currently unsupported by any algorithmic description or experimental","pith_inferences":["A reader who wants to evaluate MANGO should seek a corrected manuscript; the submitted text cannot support or refute the framework's claims as it stands.","The CLIP article that actually occupies the body suggests that aggregate bias metrics may be misleading, since bias scores vary substantially between global and local data views.","The discrepancy between abstract and body points to a simple, testable safeguard: a consistency check that the supplied full text matches the abstract's title, author list, and identifier would have caught this mismatch before distribution."],"forward_implications":["If the intended MANGO claim held, agents would solve long-horizon sparse-reward tasks by reusing nested macro-actions across abstraction layers rather than relearning from scratch.","Sample efficiency and generalization on procedurally generated grid worlds would improve relative to standard RL baselines.","Layered options would make the agent's decision process transparent, which would matter for safety-critical and industrial deployments.","The abstract's proposed future work—automated abstraction discovery, continuous or fuzzy environments, and robust multi-layer training—would be the natural next tests of the framework."],"supporting_citations":[],"fun_headline_variants":["MANGO paper: abstract promises, body delivers another study","Abstract describes MANGO, body text is a different paper","MANGO's experiments and algorithm missing from supplied text","HRL framework MANGO: claims unsupported by missing content","MANGO abstract vs body: no algorithm, no experiments"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the body text is the MANGO manuscript; if that is not true, then every claim in the abstract is without evidence in the submission.","fun_headline_variants_meta":{"raw":{"variants":["MANGO paper: abstract promises, body delivers another study","Abstract describes MANGO, body text is a different paper","MANGO's experiments and algorithm missing from supplied text","HRL framework MANGO: claims unsupported by missing content","MANGO abstract vs body: no algorithm, no experiments"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000568,"raw_usage":{"total_tokens":2489,"prompt_tokens":673,"completion_tokens":1816,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":417,"completion_tokens_details":{"reasoning_tokens":1745}},"tokens_in":417,"tokens_out":1816,"duration_ms":13993,"temperature":1.0,"reasoning_tokens":1745,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T16:45:04.237310+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check is to inspect the supplied PDF's title page and footer: if the title and identifier there are not the same as those in the abstract, the MANGO claims have no supporting text. A second check is to search the body for the term MANGO; its absence outside the abstract would confirm that no algorithm or experiment is presented.","supporting_citations":[],"review_version":1}