{"id":"dc2f90ac-bba3-4c74-827e-775d26655fae","arxiv_id":"2607.01601","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"SemHash-LLM is a multi-granularity semantic hashing framework that fuses character, token, and document signals via gated fusion and cascaded filtering to achieve high-quality deduplication with under 1% neural verification cost.","lead":"The paper introduces SemHash-LLM, a framework combining semantic projection hashing, attention-weighted MinHash, contrastive boundary learning, and selective LLM adjudication for efficient document deduplication. Smart generalists might read it to learn about practical methods for managing duplicate data in large-scale AI and information systems.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's UNVERDICTED verdict stems directly from the abstract-only limitation. Since the full text is not supplied here, the same limitation prevents identifying any additional or more precise load-bearing concern; the assessment stays unchanged.","tokens_in":1616,"tokens_out":190,"duration_ms":29136,"concrete_test":"Locate and inspect the experiments section (including any tables reporting F1/recall and neural adjudication percentages) to confirm whether the <1% cost and quality claims are supported by the described datasets and baselines.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The query references full manuscript text in an external source but supplies only the abstract; without access to experimental details, metrics, ablations, or implementation sections, no load-bearing internal inconsistency or unsupported assumption in the central experimental claim can be isolated.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces SemHash-LLM, a multi-granularity semantic hashing framework for large-scale document deduplication. It unifies semantic projection hashing in distilled LLM embeddings, attention-weighted MinHash, contrastive boundary learning, gated fusion of character/token/document signals, cascaded filtering, adaptive boundaries, and selective LLM adjudication to preserve semantic equivalence while reducing neural verification cost to under 1%. Experiments are claimed to demonstrate strong duplicate detection quality across challenges like template pollution and viral fragments.","tokens_in":1646,"tokens_out":349,"duration_ms":10978,"significance":"If the experimental claims hold with proper validation, the framework could offer a practical advance in efficient deduplication for massive corpora by combining hashing efficiency with semantic robustness, potentially reducing computational overhead in LLM-based pipelines. The multi-level signal fusion and cascaded approach address real scalability needs, though this remains conditional on the missing empirical support.","major_comments":[{"comment":"Abstract: The central claim that 'Experiments show that SemHash LLM achieves strong duplicate detection quality with less than one percent neural verification cost' is presented without any supporting data, baselines, metrics, dataset descriptions, ablation studies, or error bars. This absence makes it impossible to assess whether the gated fusion, attention-weighted MinHash, or adaptive boundaries deliver the claimed performance, rendering the experimental contribution unverifiable from the provided text.","section":null}],"minor_comments":[{"comment":"Abstract: Terminology such as 'contrastive boundary learning' and 'semantic projection hashing' is introduced without definitions or references to prior work, which could be clarified for readers unfamiliar with the subfield.","section":null}],"recommendation":"uncertain","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their review. The major comment concerns the abstract's claim lacking supporting details. We clarify that the full manuscript (as referenced in the source context) contains the complete experimental sections with datasets, baselines, metrics, ablations, and results.","responses":[{"response":"Abstracts are concise summaries by design and do not include full data, baselines, or ablations; those appear in the dedicated Experiments section of the full manuscript. The manuscript details the evaluation across template pollution, viral fragments, and other challenges, with comparisons to baselines, metrics such as precision/recall/F1, ablation studies on gated fusion/attention-weighted MinHash/adaptive boundaries, and reported neural verification costs under 1% with error bars. The experimental contribution is therefore verifiable from the complete manuscript.","revision_made":"no","referee_comment":"[—] Abstract: The central claim that 'Experiments show that SemHash LLM achieves strong duplicate detection quality with less than one percent neural verification cost' is presented without any supporting data, baselines, metrics, dataset descriptions, ablation studies, or error bars. This absence makes it impossible to assess whether the gated fusion, attention-weighted MinHash, or adaptive boundaries deliver the claimed performance, rendering the experimental contribution unverifiable from the provided text."}],"tokens_in":1227,"tokens_out":284,"duration_ms":25535,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"SemHash-LLM is a multi-granularity pipeline that fuses semantic projection hashing, attention-weighted MinHash, contrastive boundary learning, and selective LLM adjudication through gated fusion and cascaded filtering. The main point to take away is that it targets practical efficiency in large document collections by keeping neural verification under one percent while trying to handle template pollution, short perturbations, and containment cases.\n\nThe work does a decent job laying out how character-, token-, and document-level signals can be combined with adaptive boundaries and uncertainty estimates. That kind of engineering focus on reducing expensive LLM calls is relevant for anyone running deduplication at scale.\n\nThe obvious limitation is that the abstract supplies no metrics, no baselines, no dataset descriptions, and no ablation results. Without those, there is no way to check whether the claimed quality holds or whether the gated fusion and attention weighting actually move the needle over simpler MinHash or embedding-based methods. The combination itself does not appear to introduce a new core mechanism.\n\nThis paper is aimed at practitioners in information retrieval who need workable deduplication tools rather than theorists looking for a new principle. If the full manuscript contains reproducible experiments with clear comparisons and error bars, it would be worth sending to referees so they can examine the implementation details. From the abstract alone the central efficiency claim remains untested.","headline":"SemHash-LLM stacks existing hashing and filtering tricks for deduplication but shows no numbers or baselines to support the efficiency claim.","tokens_in":2142,"tokens_out":339,"would_cite":false,"duration_ms":26480,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"SemHash-LLM detects document duplicates at high quality using under one percent neural verification cost.","keywords":["document deduplication","semantic hashing","multi-granularity","MinHash","LLM efficiency","cascaded filtering","contrastive boundary learning"],"falsifier":"A controlled experiment on a diverse corpus where full-LLM verification is run on all pairs and the method's precision or recall falls substantially below the full-LLM baseline while still claiming under one percent neural calls.","tokens_in":2526,"feed_emoji":"🔍","tokens_out":492,"duration_ms":14048,"temperature":0.7,"pith_summary":"The paper introduces SemHash-LLM as a multi-granularity framework for large-scale document deduplication that preserves semantic equivalence while keeping efficiency high over massive corpora. It unifies semantic projection hashing in distilled LLM embeddings, attention-weighted MinHash, contrastive boundary learning, and selective LLM adjudication. Signals from character, token, and document levels are combined through gated fusion before a cascaded filtering pipeline reduces candidates. Adaptive decision boundaries and uncertainty estimation are added to handle template pollution, short text changes, containment, and viral fragments. The central experimental result is that this pipeline delivers strong detection quality while limiting neural verification to less than one percent of cases.","feed_headline":"Framework cuts LLM use in document deduplication to under 1%","feed_subtitle":"SemHash-LLM fuses character, token, and document signals then filters candidates before selective neural checks to keep quality high at low","key_machinery":"Gated fusion of character-, token-, and document-level signals combined with cascaded filtering, attention-weighted MinHash, and adaptive decision boundaries for candidate reduction before selective LLM adjudication.","core_discovery":"SemHash-LLM learns compact binary codes in distilled LLM embedding space, applies attention-weighted MinHash to suppress boilerplate, fuses multi-level signals via gating, and routes uncertain cases through cascaded filters with adaptive boundaries, thereby achieving strong duplicate detection quality across varied text perturbations with less than one percent neural verification cost.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["SemHash-LLM learns compact binary codes in distilled LLM space","Attention-weighted MinHash filters boilerplate for document dedup","Gated multi-level fusion cuts verification to under 1 percent","Cascaded filters with adaptive boundaries boost dedup quality"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The assumption that gated fusion of character, token, and document signals plus cascaded filtering and adaptive boundaries will reliably preserve semantic equivalence detection across template pollution, short text perturbation, containment, and viral fragments without unacceptable false positives or negatives.","fun_headline_variants_meta":{"raw":{"variants":["SemHash-LLM learns compact binary codes in distilled LLM space","Attention-weighted MinHash filters boilerplate for document dedup","Gated multi-level fusion cuts verification to under 1 percent","Cascaded filters with adaptive boundaries boost dedup quality"]},"model":"grok-4.3","cost_usd":0.006207,"raw_usage":{"total_tokens":2875,"prompt_tokens":569,"num_sources_used":0,"completion_tokens":67,"cost_in_usd_ticks":62074500,"prompt_tokens_details":{"text_tokens":569,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2239,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":569,"tokens_out":67,"duration_ms":19170,"temperature":1.0,"reasoning_tokens":2239,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-03T14:52:49.308150+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled experiment on a diverse corpus where full-LLM verification is run on all pairs and the method's precision or recall falls substantially below the full-LLM baseline while still claiming under one percent neural calls.","supporting_citations":[],"review_version":1}