{"id":"3559a52d-230d-4254-973d-099dfdd9c788","arxiv_id":"2606.20846","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Multi-level analysis of adverse social interactions on X and Bluesky finds structural negativity more persistently marks subgroup disruption while toxic communication captures broader conflict.","lead":"The paper proposes a multi-level framework analyzing adverse social interactions on X and Bluesky at micro, meso, and macro scales. It finds structural negativity persistently marks subgroup disruption while toxic communication signals broader conflict.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Matched triadic designs and random/recommendation references may leave residual confounding from algorithmic exposure and self-selection","rationale":"The reader's weakest assumption correctly isolates the identification problem that underpins the multi-scale claim. Observational network studies on these platforms routinely face exactly this class of confounder; the paper's design choices address it only partially, so the headline result remains conditional on the controls being sufficient.","tokens_in":1680,"tokens_out":267,"duration_ms":7795,"concrete_test":"Recompute the meso-level triad statistics after adding explicit matching on user activity rate, account age, and recent recommendation exposure (using available metadata or proxy timestamps); if the structural-vs-toxic distinction in disruption persistence shrinks below statistical significance, the complementary-signals interpretation weakens.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim—that structural disconnection and toxic communication supply complementary signals, with the former more persistently marking subgroup disruption—rests on the meso-level matched triadic comparisons and macro-level reference models successfully isolating ASI effects. In X and Bluesky data, however, these controls do not obviously block platform-driven recommendation effects or user self-selection into hostile neighborhoods; any residual correlation between tie formation and unobserved user traits could produce the reported persistence and scope differences without the claimed complementarity.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes a multi-level framework for studying adverse social interactions (ASIs) in online communities, drawing on large-scale X and Bluesky datasets. It examines friend/foe patterns at the micro level, peer influence via matched triadic designs at the meso level, and subgroup disruption at the macro level by comparing observed patterns against random and recommendation-based reference models. The central claim is that structural disconnection and toxic communication act as complementary signals: structural negativity more persistently marks subgroup disruption, while toxic communication reflects broader conflict within and across communities.","tokens_in":1776,"tokens_out":521,"duration_ms":15111,"significance":"If the controls successfully isolate ASI effects, the work would provide a rare multi-scale integration of structural and content-based signals for community evolution, with the public code release aiding reproducibility. The directional findings on persistence and scope differences could inform platform moderation if the meso- and macro-level comparisons hold after addressing potential confounders.","major_comments":[{"comment":"Meso-level matched triadic designs: the claim that these designs isolate peer influence on ASIs (and thereby support the complementarity result) is load-bearing for the central claim, yet the manuscript provides no details on how self-selection into hostile neighborhoods or algorithmic exposure is blocked; residual correlation between tie formation and unobserved user traits could produce the reported persistence differences without the claimed structural vs. toxic distinction.","section":"meso-level analysis (matched triadic designs)"},{"comment":"Macro-level reference models: the comparison of subgroup disruption against random and recommendation-based references is presented as evidence that structural negativity marks disruption more persistently, but without reported robustness checks (e.g., alternative matching criteria or sensitivity to recommendation algorithm parameters), it is unclear whether the scope differences survive plausible alternative reference constructions.","section":"macro-level analysis"}],"minor_comments":[{"comment":"The abstract and methods overview omit any mention of statistical controls, error estimation, data cleaning steps, or robustness checks; these should be added to allow verification of the directional findings.","section":"abstract and methods overview"},{"comment":"Notation for 'structural disconnection' and 'toxic communication' should be defined explicitly with examples from the datasets before the results are interpreted.","section":"results interpretation"}],"recommendation":"major_revision","confidential_remarks":"The low soundness rating in the reader's report aligns with the absence of implementation details; the manuscript may benefit from an expanded methods appendix before resubmission."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback. We address each major comment below, indicating revisions where the manuscript requires clarification or additional checks.","responses":[{"response":"We acknowledge that the manuscript lacks explicit details on how the matched triadic designs address self-selection into hostile neighborhoods or algorithmic exposure. The design matches triads on observable covariates including degree, activity level, and temporal proximity, but residual confounding from unobserved traits remains possible in observational data. In the revision we will expand the methods section with a full description of the matching procedure and covariates, plus a limitations paragraph discussing the inability to fully block all selection effects. This will clarify the scope of the peer-influence claim without overstating isolation.","revision_made":"yes","referee_comment":"[meso-level analysis (matched triadic designs)] Meso-level matched triadic designs: the claim that these designs isolate peer influence on ASIs (and thereby support the complementarity result) is load-bearing for the central claim, yet the manuscript provides no details on how self-selection into hostile neighborhoods or algorithmic exposure is blocked; residual correlation between tie formation and unobserved user traits could produce the reported persistence differences without the claimed structural vs. toxic distinction."},{"response":"We agree that the macro-level results would be strengthened by explicit robustness checks. The current random and recommendation-based references follow common null-model practices, yet the manuscript does not report sensitivity to alternative matching criteria or recommendation parameters. In the revised version we will add these checks, including variation in recommendation-model parameters and alternative subgroup definitions, to verify that the reported differences in persistence and scope are not artifacts of the specific reference constructions.","revision_made":"yes","referee_comment":"[macro-level analysis] Macro-level reference models: the comparison of subgroup disruption against random and recommendation-based references is presented as evidence that structural negativity marks disruption more persistently, but without reported robustness checks (e.g., alternative matching criteria or sensitivity to recommendation algorithm parameters), it is unclear whether the scope differences survive plausible alternative reference constructions."}],"tokens_in":1332,"tokens_out":437,"duration_ms":21164,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core finding is that structural disconnection and toxic communication act as complementary signals across scales: the former tracks persistent subgroup breaks more reliably, while the latter picks up wider conflict. The work applies this at micro (friend/foe patterns), meso (matched triads for peer influence), and macro (subgroup disruption versus random and recommendation baselines) on two platform datasets, with public code.\n\nWhat stands out is the explicit three-scale design that brings both structural and content signals together rather than treating them separately. The matched triadic setup and the dual reference models at the macro level are reasonable attempts to tighten the comparison, and releasing the code lets others check the implementation.\n\nThe soft spot is the isolation of effects. The stress-test concern holds: platform recommendation systems and users sorting themselves into hostile neighborhoods can still correlate with the observed patterns even after the triadic matching and reference models. Without clearer reporting on how those controls were constructed, how sensitive the results are to alternative matching criteria, or any robustness checks against unobserved user traits, the claimed complementarity rests on an assumption that may not fully hold. The abstract gives directional results but no effect sizes, standard errors, or details on data cleaning, so the strength of the evidence is hard to gauge from what is shown.\n\nThis is useful for researchers already working on online community dynamics and platform moderation who want an empirical example of multi-scale adverse interaction patterns. It does not introduce new theory or first-principles derivations, but the framework is straightforward enough that a referee could evaluate the controls and ask for the missing checks.\n\nI would send it to peer review. The data and code are real assets, and the confounding issue is addressable with revisions rather than fatal.","headline":"The paper runs matched triadic and reference-model comparisons on X and Bluesky data to argue that structural negativity marks subgroup disruption more persistently than toxic content, but the controls leave room for algorithmic and self-selection confounds.","tokens_in":2293,"tokens_out":434,"would_cite":false,"duration_ms":15293,"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":"Structural disconnection marks subgroup disruption more persistently than toxic communication in online communities.","keywords":["adverse social interactions","online communities","structural disconnection","toxic communication","multi-level analysis","community disruption","social networks"],"falsifier":"A new dataset or controlled experiment in which toxic communication alone predicts the timing and persistence of subgroup disruption as strongly as structural disconnection does would falsify the claim that the two provide distinct complementary signals.","tokens_in":2574,"feed_emoji":"","tokens_out":601,"duration_ms":22928,"temperature":0.7,"pith_summary":"The paper develops a multi-level framework to track adverse social interactions across local patterns, neighborhood spread, and community-level disruption in large datasets from two platforms. It compares friend and foe structures at the smallest scale, uses matched triadic designs to isolate peer influence at the middle scale, and measures how subgroups fracture against random and recommendation baselines at the largest scale. The central finding is that broken connections and toxic messages act as complementary signals rather than interchangeable ones. Structural negativity stays tied to lasting subgroup breaks, while toxic content flags conflict that crosses community lines. This matters because it shows single-scale or single-type studies will miss how communities actually form and split over time.","feed_headline":"Structural disconnection tracks online subgroup breaks more persistently than toxic posts","feed_subtitle":"Multi-level analysis of X and Bluesky data shows the two signals mark distinct aspects of how communities fracture.","key_machinery":"The multi-level framework that examines local friend-foe patterns, peer influence through matched triadic designs at the meso level, and subgroup disruption against random and recommendation-based references at the macro level.","core_discovery":"Using data from X and Bluesky, the analysis shows that adverse social interactions operate as multi-scale processes: structural disconnection and toxic communication provide complementary signals where structural negativity more persistently marks subgroup disruption while toxic communication captures broader conflict both within and across communities.","pith_inferences":["Platform tools could combine structural monitoring with content flags to catch different kinds of disruption earlier.","The same complementarity might appear in other online platforms if the multi-level method is applied there.","Recommendation systems may need separate adjustments for structural versus content signals to limit community fracture."],"forward_implications":["Adverse social interactions influence how online communities form, fracture, and evolve as multi-scale processes.","Structural negativity serves as a more persistent marker for subgroup disruption than toxic communication.","Toxic communication signals conflict that spans both inside and across communities.","Single-scale studies of adverse interactions will miss key dynamics of community change."],"fun_headline_variants":["Structural disconnection persistently tracks online subgroup breaks","Toxic communication captures wider conflicts in online communities","Multi-scale analysis of X and Bluesky shows distinct fracture signals","Structural negativity marks subgroup disruption more persistently"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The matched triadic designs and random or recommendation-based reference groups at the middle and large scales remove confounding from platform algorithms and user self-selection in the two datasets.","fun_headline_variants_meta":{"raw":{"variants":["Structural disconnection persistently tracks online subgroup breaks","Toxic communication captures wider conflicts in online communities","Multi-scale analysis of X and Bluesky shows distinct fracture signals","Structural negativity marks subgroup disruption more persistently"]},"model":"grok-4.3","cost_usd":0.005029,"raw_usage":{"total_tokens":2417,"prompt_tokens":596,"num_sources_used":0,"completion_tokens":56,"cost_in_usd_ticks":50287000,"prompt_tokens_details":{"text_tokens":596,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1765,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":596,"tokens_out":56,"duration_ms":13425,"temperature":1.0,"reasoning_tokens":1765,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-26T14:46:52.408341+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A new dataset or controlled experiment in which toxic communication alone predicts the timing and persistence of subgroup disruption as strongly as structural disconnection does would falsify the claim that the two provide distinct complementary signals.","supporting_citations":[],"review_version":1}