{"id":"b0d0315b-68cf-4ebe-8a4f-437cfa183ee5","arxiv_id":"2608.05462","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A multiaxial dynamical taxonomy for insomnia, using a Landau-Ginzburg phenomenological model, organizes failure by operation, stage, and causal status and makes two falsifiable predictions.","lead":"This paper proposes a three-axis framework for classifying insomnia by which dynamical operation fails, where in the sleep cycle it fails, and whether the failure is causal. It uses a Landau-Ginzburg model as a common language and argues the taxonomy should be judged by whether it improves prediction and communication beyond existing labels.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Axis I labels may be non-identifiable from routine PSG: control-path, boundary-crossing, and stabilization errors trade off, so the central taxonomy's operational substrate is untested.","rationale":"The reader's weakest assumption is that the six operation classes are separable in patients, reliably assigned from PSG plus probes, and add predictive value. I agree that this is the load-bearing premise. My concern sharpens it: separability is not merely an empirical question to be settled after the taxonomy is applied; it has an identifiability prerequisite. Section 3.9 shows that passive data identify only effective parameter combinations, and Section 6.2 concedes that fitted parameters must outperform model-free measures. But Section 3.10 then asserts that routine recordings can classify the failed operation and location. The gap is that no analysis shows the coarsest Axis I categories are recoverable from the data the framework proposes to use. The paper itself lists simulation-based identifiability as a priority (Table 5), which supports the good-faith reading that this is a known open condition rather than an oversight. Still, the central claim as stated assumes a positive outcome. If simulation shows non-recoverability, the taxonomy's testable core collapses; if simulation succeeds, the remaining question is empirical utility. This does not change the verdict from CONDITIONAL: the paper is internally careful and falsifiable, but its central premise is unvalidated. I do not see a reason to reject or accept outright based on the current text.","tokens_in":18559,"tokens_out":4227,"duration_ms":44051,"concrete_test":"Run the 'Simulation-based identifiability' study listed in Table 5: generate 1,000 synthetic PSG-like datasets with known ground-truth Axis I operations by sampling from the proposed dynamics (Eqs. 2 and 8) with known coefficients, PSG sampling, noise, and missingness; apply the paper's proposed estimation pipeline; compute the confusion matrix for the six operation classes. If coarse Axis I labels are recovered with balanced accuracy >= 0.8 and no systematic operation confusions, the concern is resolved; if not, the Section 3.10 claim that routine PSG can classify operations must be downgraded, and the central taxonomy's utility tests are moot until a recoverable protocol is demonstrated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The framework's operational bridge to practice is the claim in Section 3.10 that routine PSG can classify the failed operation and its location. Yet Sections 3.9 and Table 3 concede that passive data identify only effective combinations (Gamma*mu, Delta V/Deff, Delta W/epsilon) and that pooled or driven trajectories can mimic bistability. At sleep onset, the coarsest Axis I split—control-path progression vs boundary-crossing failure vs post-transition stabilization—requires separating an unobserved homeostatic/circadian ramp from an unobserved boundary-loss process. These latent quantities can compensate: a model that mis-specifies the ramp can attribute a crossing failure to underdrive or a stabilization failure to delayed crossing. The paper says the driven-ramp null must be fitted explicitly, but it does not demonstrate that routine PSG contains sufficient information to do so reliably. If Axis I labels cannot be recovered even from synthetic data with known ground truth, the central three-axis claim has no reliable empirical substrate: utility tests would be validating labels that cannot be assigned. This is the load-bearing assumption behind the CONDITIONAL verdict.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a multiaxial research ontology for insomnia, in which a patient profile is specified by (Axis I) the dynamical operation that fails—control-path progression, boundary crossing, post-transition stabilization, spatial recruitment, architectural sequencing, or observation mapping—(Axis II) the sleep stage or boundary at which it fails, and (Axis III) the causal status of the abnormality, with objective sleep duration and other clinical variables treated as modifiers. A local Landau-Ginzburg quasipotential model is introduced for boundary-local transitions, with careful distinctions between gradient and non-gradient regimes, curvature vs. barrier vs. escape action, and effective parameter combinations vs. individual coefficients. The paper makes two affirmative predictions (local recovery/escape dissociation; rounded-bifurcation or bistable models outperforming a driven Process-S/C smooth sigmoid for sleep-onset trajectories) and proposes a near-term crux reanalysis plus simulation-based identifiability and taxonomy-utility studies. The authors explicitly state that the framework is phenomenological, not a validated diagnostic or treatment-selection system.","tokens_in":18856,"tokens_out":5143,"duration_ms":48411,"significance":"If the framework survives operationalization, it would provide a common language for separating mechanism-level hypotheses in insomnia heterogeneity and generate tests that conventional onset/maintenance/terminal-awakening subtyping does not. The paper's strengths are its discipline: it guards against circularity (§3.7), identifies identifiability limits and effective combinations (§3.9, Table 3), preregisters falsifying conditions for the model and the ontology (§6.4–6.6), and gives a concrete near-term experiment. Its significance is conditional on the yet-unmet empirical assumption that the six operation classes can be reliably assigned and add predictive value; the paper itself acknowledges this, but the referee's assessment should reflect that the central utility claim is not yet demonstrated.","major_comments":[{"comment":"The manuscript's central three-axis claim requires that the failed dynamical operation and its location be assignable, yet Section 3.9 and Table 3 show that passive PSG identifies only effective combinations (Γμ, ΔV/Deff, ΔW/ε) and that the coarsest sleep-onset split (control-path progression vs. boundary crossing vs. post-transition stabilization) requires separating an unobserved homeostatic/circadian ramp from an unobserved boundary-loss process. These latent quantities can compensate for each other, so the paper should demonstrate—at minimum on synthetic data with known ground truth, as listed in Table 5—that the coarse Axis I labels are recoverable from the proposed PSG-plus-covariate protocol. Without such an identifiability check, the utility tests proposed in Section 6.4 would validate labels that may already be non-identifiable from the planned measurements.","section":"§3.9, §3.10, Table 3"},{"comment":"The six Axis I operation classes and the candidate profiles in Table 4 are not accompanied by an operational coding rubric specifying which measurements, thresholds, or hierarchical decision rules assign a given recording to a given operation. Section 6.4 makes inter-rater reliability a utility criterion, but the manuscript does not provide the coding manual that would allow trained raters to apply the labels; for example, the rule for assigning primary N2 organization deficit versus failure of early-NREM consolidation is stated qualitatively. A minimal scoring/classification protocol should be specified before the taxonomy can be tested.","section":"§4.1, §4.5, Table 4"},{"comment":"The near-term crux experiment compares trajectory-prediction performance of driven sigmoid vs. rounded/bistable/escape models, but the central three-axis claim is about taxonomy-level labels and their incremental predictive utility. The paper should state precisely how a positive crux result advances the three-axis ontology (rather than only the sleep-onset mechanical model), and which prespecified outcome would lead to simplification or retirement of the ontology despite a positive model-comparison result; otherwise the mapping between model falsification and taxonomy validation remains ambiguous.","section":"§6.6"}],"minor_comments":[{"comment":"Equations (1)–(8) are referenced throughout the text but are not displayed in the manuscript; the formulas should be included in the published version or in an appendix.","section":"§3.1–§3.8"},{"comment":"Figure 1 and Figure 2 are described in the text but are not present in the manuscript; they should be provided or explicitly marked as schematic placeholders in the submission.","section":"Figures 1 and 2"},{"comment":"Reference 59 is a preprint from the author's own group; Section 1.2 states that preprints carry no essential evidentiary burden, but the declarations should explicitly flag this as a self-citation for transparency.","section":"Declarations"},{"comment":"The evidence-selection paragraph is thorough, but it would be easier to verify if the references classified as peer-reviewed, accepted, or preprint were listed in a table or footnote rather than only described in prose.","section":"§1.2"}],"recommendation":"major_revision","confidential_remarks":"This is a conceptually coherent, carefully hedged proposal rather than an empirical study. The main risk is not circularity or overclaiming—the authors are exemplary on those fronts—but that the central taxonomy cannot yet be grounded because Axis I label identifiability is untested. I would advise the editor that the paper is publishable after the identifiability and operationalization issues are addressed; whether that requires new simulation results is an editorial judgment. The author's financial affiliation is disclosed, and the self-citation of Reference 59 should be made explicit to reviewers."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First, this is a serious conceptual paper, not a data paper. Second, its biggest risk is exactly where the authors are most cautious: whether the six Axis I operations can be assigned from routine PSG at all.\n\nWhat's actually new is the three-axis operation-location-causal taxonomy and two predictions: recovery and escape dissociate reproducibly, and rounded/bistable models beat a driven sigmoid on subject-held-out sleep-onset trajectories. The paper is honest about what it is: the Landau-Ginzburg machinery is adopted, the evidence map is explicitly tiered, and the limitations section tells you which results would kill the framework (Section 6.4: retire if poor inter-rater reliability or no incremental prediction; Section 6.6: the driven-sigmoid crux). Most taxonomy papers don't build in their own falsification; this one does, and it deserves credit for that.\n\nThe soft spot is the operational bridge. Section 3.10 claims routine PSG can classify the failed operation and its location, but Section 3.9 shows that passive data identify only effective combinations—Γμ, ΔV/Deff, ΔW/ε—and that h(t) is confounded with asymmetric noise and pooled trajectories can mimic bistability. At sleep onset, the coarsest Axis I split (control-path underdrive vs. boundary-crossing failure vs. early-NREM consolidation) requires separating an unobserved homeostatic/circadian ramp from an unobserved boundary-loss process. These latent quantities can compensate: mis-specify the ramp and you can attribute a crossing failure to underdrive or a stabilization failure to delayed crossing. The paper says the driven-ramp null must be fitted explicitly, but it does not demonstrate that routine PSG contains enough information to do so reliably, even from synthetic data. That gap is load-bearing for the taxonomy's practical value, not for the math, which is internally consistent. The authors' own utility criteria concede that if the labels cannot be assigned reliably, the ontology should be retired—so this is a fair criticism, not a fatal one.\n\nThe citation pattern looks fine: preprints are flagged as such, and the self-citations are to the author's own speculative extension, not to inflate the core claims. The paper is over-long and sometimes repetitive, but the structure is clear.\n\nWho is this for: sleep researchers, clinical trial designers, and computational modelers who want a shared language for dynamical failure modes. It is not for anyone wanting a validated biomarker or a treatment algorithm. It deserves a serious referee—the framework is original, testable, and honestly bounded. My own verdict would be positive with major revision: the authors should either show a synthetic-data identifiability simulation for the coarse Axis I labels or explicitly downgrade the claim that routine PSG can assign them.","headline":"A carefully hedged taxonomy paper with real organizational value and an honest limitations section; its central Axis I assignments are untested and may not be identifiable from routine PSG, but it deserves a serious referee.","tokens_in":19335,"tokens_out":3202,"would_cite":true,"duration_ms":27677,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Insomnia is not one syndrome but a family of separable dynamical failures, classified along three axes.","keywords":["insomnia","sleep architecture","sleep-stage transitions","dynamical systems","Landau-Ginzburg","bistability","sleep onset","two-process model"],"falsifier":"Run the paper's near-term crux analysis on a large sleep-onset cohort with continuous presleep EEG: fit each subject's onset trajectory with a driven Process-S/Process-C sigmoid, a rounded bifurcation model, stochastic bistable/escape models, and hidden-state baselines, holding out subjects rather than epochs. If the driven sigmoid predicts held-out trajectories as well as or better than the dynamical alternatives, the sleep-onset mechanistic claims must be downgraded to descriptive state change.","tokens_in":18338,"feed_emoji":"😴","tokens_out":6435,"duration_ms":55034,"temperature":0.7,"pith_summary":"Insomnia disorder is defined by symptoms, but the same complaint—trouble falling or staying asleep—can arise through different physiological routes. This paper proposes that the field would do better to classify insomnia profiles by three separate axes: the dynamical operation that fails, the sleep stage or boundary at which it fails, and whether that failure is primary, secondary, maintaining, or just a biomarker. The claim is organizational rather than a new diagnosis: a Landau-Ginzburg field model supplies a common vocabulary for nested mechanistic hypotheses, and the taxonomy should be judged by whether it improves communication, stratification, and prediction. The authors are explicit that only sleep onset currently has strong boundary-level evidence, and that even there a driven homeostatic-circadian ramp remains an alternative to genuine bifurcation or bistability.","feed_headline":"New taxonomy sorts insomnia by which sleep mechanism fails","feed_subtitle":"Which sleep operation fails, at which boundary, and why—a three-axis framework for subtyping insomnia.","key_machinery":"The machinery is a three-axis taxonomic grammar—operation, stage or boundary, causal status—layered on an effective field theory. The central object is a dimensionless latent NREM-ordering coordinate $\\phi(r,t)$ governed by a stochastic time-dependent Ginzburg-Landau equation, a statistical-physics description of a scalar field relaxing in a potential landscape, with an effective quasipotential functional whose local curvature, barrier height, bias, noise, and relaxation coefficient are kept distinct. A gradient approximation with additive noise permits Kramers-type escape language, while non-gradient regimes require a Freidlin-Wentzell quasipotential or minimum action; whole-night REM-NREM sequencing requires reactive or oscillatory dynamics beyond the relaxational form. The taxonomy is the organizing contribution; the field formalism supplies a falsifiable way to turn each operation-location cell into a candidate mechanistic hypothesis.","core_discovery":"The central claim is that a candidate insomnia profile is specified by the dynamical operation that fails, the sleep stage or boundary at which it fails, and its causal status. Six operation classes are proposed: control-path progression, boundary crossing, post-transition stabilization, spatial recruitment, architectural sequencing, and observation mapping. Objective sleep duration, age, circadian phase, comorbidity, medication exposure, and night-to-night variability are treated as modifier or covariate dimensions rather than additional mechanistic classes. Within this organization, the paper adapts a stochastic Landau-Ginzburg description of a latent NREM-ordering coordinate, and insists that local curvature, gradient barrier or non-gradient escape action, bias, noise, and relaxation are distinct quantities that routine polysomnography usually identifies only in combination. It makes two affirmative predictions: local recovery and escape will dissociate reproducibly across patients, and rounded-bifurcation or bistable/escape models will predict subject-held-out sleep-onset trajectories better than a Process-S/Process-C driven smooth sigmoid.","pith_inferences":["A natural extension the author leaves implicit is that the same three-axis grammar could be applied beyond insomnia, to other sleep-wake disorders or to daytime vigilance transitions, where the same operations may fail in different locations.","The near-term crux experiment can be run on already-collected polysomnography cohorts, so the strongest claim can be tested without new data collection; if the driven sigmoid wins, the dynamical mechanism claims lose their strongest empirical anchor.","The recovery-escape dissociation could be tested in animal models or with closed-loop stimulation protocols, where graded perturbation amplitudes are easier to deliver than in human PSG.","One could also fit the competing models to individual subjects rather than group trajectories; if bistable parameters are not stable within a person across nights, the endotype claim weakens even if group-level fits look similar."],"forward_implications":["If recovery and escape dissociate reproducibly, patients with similar symptom labels can be separated into profiles that predict how they respond to graded perturbations.","If rounded-bifurcation or bistable models beat a driven smooth sigmoid on held-out sleep-onset data, sleep onset should be described as a transition near a loss of stability rather than a smooth consequence of homeostatic and circadian drive.","The framework implies that two treatments producing the same total sleep time can alter different dynamical domains, so mechanism studies should measure more than sleep duration.","The paper's own utility criterion means the taxonomy stands or falls on inter-rater reliability and incremental held-out prediction; otherwise it should be simplified or retired.","Cross-night reproducibility of fitted dynamical parameters, if it exceeds that of conventional stage percentages, would be evidence that these profiles are genuine endotypes."],"supporting_citations":[{"why":"Identifies a low-dimensional bifurcation-like trajectory and passive early-warning signals for sleep onset in two large cohorts; this is the paper's strongest boundary-level evidence.","marker":"[40]"},{"why":"Fits a stochastic bistable model of wake-sleep flickering; supplies the bistable/escape model class that is contrasted with a driven smooth sigmoid.","marker":"[41]"},{"why":"Defines the two-process model whose homeostatic-circadian ramp is the key driven-smooth null for sleep-onset transition claims.","marker":"[2-4]"},{"why":"Provides the flip-flop circuit mechanism for rapid state switching, which the paper treats as a plausible implementation of bistability rather than a classification.","marker":"[5-7]"},{"why":"Supplies the Landau-Ginzburg theory of cortical dynamics from which the paper adopts its field formalism.","marker":"[45]"},{"why":"Supplies the dynamic-critical-phenomena classification, including Model A relaxational dynamics, that sets the scope of the gradient approximation.","marker":"[61]"},{"why":"Reports dissociable changes in NREM delta/beta ratio and transition-matrix stability after CBT-I, supporting the separation of cortical activation from macrostate stability.","marker":"[30]"},{"why":"Reports greater evoked complexity and altered coupling after auditory perturbation, motivating the perturbation-resistance and recovery-escape distinction.","marker":"[31]"},{"why":"Provides EEG early-warning-signal evidence for sleep onset, relevant to the question of whether passive slowing reflects a true fold or a driven ramp.","marker":"[47]"}],"fun_headline_variants":["Insomnia subtyped by which sleep operation fails","Six failure modes define a dynamical taxonomy of insomnia","New framework: locate the failing sleep stage and operation","Multiaxial insomnia taxonomy: operation, boundary, causal status","Architectural failure: organizing insomnia by mechanism"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the six operation classes are separable in real patients, so trained raters or algorithms can assign operation-location-causal labels reliably and those labels add predictive value beyond conventional insomnia subtypes.","fun_headline_variants_meta":{"raw":{"variants":["Insomnia subtyped by which sleep operation fails","Six failure modes define a dynamical taxonomy of insomnia","New framework: locate the failing sleep stage and operation","Multiaxial insomnia taxonomy: operation, boundary, causal status","Architectural failure: organizing insomnia by mechanism"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000257,"raw_usage":{"total_tokens":1620,"prompt_tokens":1026,"completion_tokens":594,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":642,"completion_tokens_details":{"reasoning_tokens":520}},"tokens_in":642,"tokens_out":594,"duration_ms":6036,"temperature":1.0,"reasoning_tokens":520,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T12:36:04.191917+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's near-term crux analysis on a large sleep-onset cohort with continuous presleep EEG: fit each subject's onset trajectory with a driven Process-S/Process-C sigmoid, a rounded bifurcation model, stochastic bistable/escape models, and hidden-state baselines, holding out subjects rather than epochs. If the driven sigmoid predicts held-out trajectories as well as or better than the dynamical alternatives, the sleep-onset mechanistic claims must be downgraded to descriptive state change.","supporting_citations":[{"cited_title":"Falling asleep follows a predictable bifurcation dynamic","cited_arxiv_id":null,"evidence_quote":"Identifies a low-dimensional bifurcation-like trajectory and passive early-warning signals for sleep onset in two large cohorts; this is the paper's strongest boundary-level evidence."},{"cited_title":"Learning the bistable cortical dynamics of the sleep-onset period","cited_arxiv_id":null,"evidence_quote":"Fits a stochastic bistable model of wake-sleep flickering; supplies the bistable/escape model class that is contrasted with a driven smooth sigmoid."},{"cited_title":"Landau-Ginzburg theory of cortex dynamics: scale-free avalanches emerge at the edge of synchronization","cited_arxiv_id":null,"evidence_quote":"Supplies the Landau-Ginzburg theory of cortical dynamics from which the paper adopts its field formalism."},{"cited_title":"Theory of dynamic critical phenomena","cited_arxiv_id":null,"evidence_quote":"Supplies the dynamic-critical-phenomena classification, including Model A relaxational dynamics, that sets the scope of the gradient approximation."},{"cited_title":"The effectiveness of cognitive behavioral therapy for insomnia on sleep EEG hyperarousal: a multicentric polysomnographic study","cited_arxiv_id":null,"evidence_quote":"Reports dissociable changes in NREM delta/beta ratio and transition-matrix stability after CBT-I, supporting the separation of cortical activation from macrostate stability."},{"cited_title":"Impaired sleep resilience underlies transient neural instability in insomnia disorder","cited_arxiv_id":null,"evidence_quote":"Reports greater evoked complexity and altered coupling after auditory perturbation, motivating the perturbation-resistance and recovery-escape distinction."},{"cited_title":"Dynamics of sleep onset: early-warning signals in EEG","cited_arxiv_id":null,"evidence_quote":"Provides EEG early-warning-signal evidence for sleep onset, relevant to the question of whether passive slowing reflects a true fold or a driven ramp."}],"review_version":1}