{"id":"b2d4822e-6507-4b7d-841a-157f66fb7647","arxiv_id":"2608.03000","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A Landau-Ginzburg phenomenology maps each sleep-stage boundary to a distinct transition type, with a supported fold at wake-to-N1, and provides EEG testable predictions, validated only on synthetic data.","lead":"This paper proposes a Landau-Ginzburg framework for classifying sleep-stage transitions, assigning different dynamical mechanisms, such as fold, crossover, first-order-like, and tricritical-like, to different boundaries. It tests the framework only on synthetic data and explicitly calls for transition-centered EEG validation before clinical use.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The taxonomy's testability rests on the latent order parameter φ being identifiable from prespecified EEG features without using boundary-defining signals; this is specified but not fitted, so the conditional verdict is appropriate.","rationale":"The paper is a carefully scoped phenomenological framework; it repeatedly disclaims empirical validation and marks the within-N3 hypothesis speculative. My read agrees with the reader's CONDITIONAL verdict. The strongest claim is not that the taxonomy is true, but that it is testable and internally consistent. The internal-consistency simulations do support that, and the synthetic classification experiment (0.49 vs 0.17 baseline) is modest but non-trivial, with the scoring-artifact confound handled honestly. The load-bearing weak point is the same one the reader identified: every empirical prediction passes through the latent coordinate φ of Eq. (11), and the paper's own circularity control (Sec. IVE) makes the identifiable feature set smaller for exactly the boundaries where slow-wave activity is the defining signal. A linear factor model with only non-scoring indicators may not recover a coordinate with enough sensitivity to distinguish a fold from a crossover. This is an empirical question, not a formal defect; the paper says the factor model is not fitted. The concrete test above would settle whether the non-circularity constraint is compatible with the predicted discriminability. If it fails, the taxonomy loses its claimed evaluability, weakening the central claim; if it passes, the conditional path to ACCEPT is clear. Because the concern is already identified by the reader and the paper is honest about it, no verdict change is needed.","tokens_in":26428,"tokens_out":8509,"duration_ms":90967,"concrete_test":"Fit the factor model of Eq. (11) on Sleep-EDF Expanded with prespecified indicators (inverse Lempel-Ziv complexity, spatial synchrony, and, in a non-excluded variant, slow-oscillation dominance) on a training subset; test configural and metric invariance across subjects, cycles, and montages; then, on held-out subjects, estimate φ and test whether wake-to-N1 and NREM-to-REM boundaries show larger discontinuity or bimodality than N2-to-N3 after excluding slow-wave dominance from φ for the N3-related tests. Report the drop in model-comparison separation when boundary-defining features are excluded; if the separation vanishes, non-circularity is unattainable with current PSG and the taxonomy requires revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—that boundary-specific transition classes can be distinguished empirically—is conditional on the latent ordering field φ(t) in Eq. (11) being identifiable from prespecified EEG/PSG features and on that identification being non-circular. The authors specify the factor model but do not fit it, and Sec. IVE's own non-circularity rule requires excluding any feature used to define a boundary from the construction of φ for that test. For N2-to-N3 and the within-N3 subregime, the boundaries are defined by slow-wave activity (AASM/R&K), so φ must be built from residual features (inverse complexity, synchrony) that may carry too little signal to exhibit the predicted continuous versus discontinuous signatures. If φ is unidentified, mis-specified, or cannot be estimated independently of the scored labels, the proposed predictions are not evaluable in practice and the taxonomy loses its claimed empirical grounding. This is not an internal inconsistency—the paper is explicit that fitting is future work—but it is the load-bearing assumption on which the entire validation program rests.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a Landau–Ginzburg phenomenology for sleep-stage transitions. It introduces a latent cortical-ordering coordinate φ, a prespecified measurement model (Eq. 11), and a boundary-specific taxonomy: wake-to-N1 is a supported fold with a conditional cusp embedding, N1-to-N2 and N2-to-N3 are continuous-like crossovers, NREM-to-REM is a candidate first-order-like desynchronizing switch, and a within-N3 tricritical-like subregime is a speculative hypothesis. The Ginzburg term yields spatial predictions for correlation-length growth and local-to-global recruitment. The paper includes illustrative time-dependent Ginzburg–Landau simulations, a synthetic classification experiment distinguishing six archetypes (cross-validated accuracy 0.49 against a 0.17 baseline), and a detailed validation protocol with circularity controls. It explicitly states that the measurement model is specified but not fitted, and that the taxonomy is not validated in human sleep.","tokens_in":26641,"tokens_out":10822,"duration_ms":105859,"significance":"If the framework holds, it offers a principled way to classify sleep-stage boundaries by dynamical mechanism rather than by descriptive label, with concrete, testable EEG/PSG signatures. The paper's strengths are its explicit separation of fold, cusp, crossover, and scoring-artifact mechanisms (Sec. IVD); the prespecified non-circularity rules (Sec. IVE); the detailed decision criteria and confound controls (Secs. IX–X); the reproducible synthetic experiments with overlapping noise ranges and cross-validation inside folds; and the honest labeling of the within-N3 hypothesis as speculative. The main risk is that the empirical testability of the framework rests on the unverified identifiability of φ under the exclusion rule, which is not demonstrated in the manuscript.","major_comments":[{"comment":"The paper's central testability claim (Sec. IVC) depends on the latent order parameter φ being identifiable from prespecified EEG/PSG features while respecting the non-circularity rule of Sec. IVE. For the N2-to-N3 boundary, which the paper identifies as the strongest continuous-ordering candidate, the AASM definition of N3 uses slow-wave activity; therefore slow-oscillation dominance cannot be used to construct φ for that test, and the paper must rely on residual indicators such as inverse complexity and spatial synchrony. The manuscript does not show, either empirically or in a synthetic identifiability analysis, that these residual features carry enough signal to distinguish the predicted continuous growth of correlation length from a scoring-induced discontinuity. Because Eq. (11) is specified but not fitted, and because Table V lists slow-oscillation dominance as a core indicator of NREM deepening, the falsifiable predictions for the NREM-deepening boundaries are conditional on an unverified premise. Please add a synthetic measurement-model analysis that estimates φ from a feature set that excludes the boundary-defining feature for each boundary test, and demonstrate that the predicted signatures remain detectable; alternatively, specify per-boundary feature sets that respect the exclusion rule and justify their sufficiency.","section":"Sec. IVE / Eq. (11), Sec. XC"},{"comment":"The synthetic classification experiment of Sec. VII does not incorporate the jittered or shuffled-boundary null control that Sec. XC identifies as necessary to separate a scoring-induced discontinuity from a genuine smooth transition. In the confusion matrix of Fig. 4, the smooth crossover and the scoring artifact are the least separable pair, with smooth-crossover recall at 0.41. Since scoring-induced discontinuity is the principal confound for the continuous-like N1-to-N2 and N2-to-N3 hypotheses, the demonstrated pipeline cannot by itself falsify the continuous hypothesis against that alternative. The paper acknowledges this confusion as 'honest and expected,' but then uses the same experiment to support the 'testability' of the framework. To make the support valid, the experiment should include the jittered-threshold null model as an additional feature or classifier layer, or the paper should explicitly state that the current pipeline does not separate these two archetypes and that the null-model controls of Sec. XC remain to be implemented.","section":"Sec. VII / Fig. 4, Sec. XC"},{"comment":"The statement in Sec. XF that the synthetic result 'establishes that the pipeline can make that discrimination when the ground truth is known' overstates the evidence: the cross-validated accuracy is 0.49 against a 0.17 chance level, and per-class recall ranges from 0.41 to 0.61. The abstract's wording ('partially distinguished') is accurate. Please revise the later claim to match the actual performance and to note that the discrimination is partial and archetype-dependent, with the smooth-crossover/scoring-artifact pair being a known overlap.","section":"Sec. XF vs. Sec. VII"}],"minor_comments":[{"comment":"Please display the lead-field generalization of Eq. (11) as a numbered equation, since it is used as the basis for the spatial tests in Sec. XE.","section":"Sec. IVE"},{"comment":"The caption of Fig. 3 reports onset exponents 0.58 and 0.26 while the text later gives seed-averaged values 0.56±0.05 and 0.25±0.01; please make the caption and the corresponding text consistent.","section":"Sec. VII / Fig. 3"},{"comment":"The statement that the sign of b 'may be unrecoverable from short noisy trajectories' is an important caveat for distinguishing first-order from continuous transitions; please specify what trajectory length or ensemble size is needed, or indicate how Sec. X's model-comparison framework addresses this.","section":"Appendix A, Table VII"},{"comment":"Please mark the related work by Passaro and Poltorak as a preprint in the reference list and state explicitly that it has not undergone peer review, since the text describes it as 'currently a preprint.'","section":"Sec. VIII, Ref. [49]"},{"comment":"The terms 'self-organized criticality' and 'self-organized bistability' are used without definition; a sentence clarifying the distinction would help readers outside the active-matter community.","section":"Sec. IVC / Sec. XII"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is unusually careful about circularity, confounds, and the distinction between internal consistency and empirical validation. The primary gap is the lack of any demonstration that the non-circular measurement model can be instantiated; the identifiability of φ under the exclusion rule is load-bearing for the empirical program. I would be willing to review a revision that adds a synthetic identifiability analysis or clearly delimits the testability claim to the case where φ can be estimated from non-boundary-defining features. The conflict-of-interest disclosure is exemplary. The self-citation to Ref. [49] is acceptable but should be clearly marked as a preprint."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a framework paper, not a result paper, and it is honest about that. What's actually new is the boundary-specific taxonomy: wake-to-N1 as a fold with a conditional cusp embedding, N1-to-N2 and N2-to-N3 as continuous-like crossovers, NREM-to-REM as a first-order-like switch, and a speculative within-N3 tricritical-like subregime. The Ginzburg spatial term adds predictions—correlation-length growth, local-to-global recruitment—that scalar sleep-onset models do not make. The paper also does the field a service by keeping the five candidate mechanisms (fold, crossover, coexistence, noise-driven escape, scoring artifact) distinct and specifying what evidence would separate them. The synthetic classification experiment is honestly reported: 0.49 accuracy versus 0.17 baseline, with the expected confusion between smooth crossover and scoring-induced jump acknowledged as the honest outcome. The limitations section is unusually candid; the author explicitly says this establishes internal consistency, not the taxonomy in human sleep.\n\nThe soft spot is exactly where the reader put it. The latent ordering field phi, estimated from prespecified EEG/PSG features via Eq. (11), is the load-bearing construct, and it is specified but not fitted. No measurement invariance testing, no demonstration that boundary-defining features can be cleanly excluded from phi for the N2-to-N3 and within-N3 tests, where slow-wave activity defines the boundary. That is not an internal contradiction, but it means the central empirical claims are untested. The REM-to-wake section is a second, smaller soft spot: the paper admits a multidimensional state vector is needed there, which is fine, but it narrows the abstract's promise more than the headline suggests. The self-cited preprint is disclosed and used only as motivation; that is not a problem.\n\nI agree with the conditional verdict. The math is standard Landau-Ginzburg, the scoping is careful, and the claims are properly hedged. This deserves a serious referee—ideally a statistical physicist and a sleep EEG researcher working together. I would not desk-reject it. My own verdict: engage with it as a useful conceptual framework and a detailed validation template, but do not mistake the taxonomy for established fact. The paper should be published with the measurement model and the planned empirical work front and center, and the speculative status of the within-N3 regime should stay clearly marked.","headline":"A carefully scoped Landau-Ginzburg taxonomy of sleep-stage boundaries; the math is standard and the honesty is refreshing, but the promised non-circular measurement model is the load-bearing piece and it is not yet fitted.","tokens_in":27209,"tokens_out":2068,"would_cite":true,"duration_ms":23519,"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":"Each sleep-stage boundary may be its own kind of dynamical transition: fold, crossover, or switch, all generated by one noisy neural-field potential.","keywords":["sleep architecture","sleep stages","sleep-stage transition","Landau-Ginzburg","phase transitions","state transitions","cortical dynamics","nonlinear dynamics"],"falsifier":"Run the paper's own proposed test on public polysomnography: estimate $\\varphi$ from features that exclude the boundary-defining ones, estimate transition times independently of the scored labels, and compare smooth, switching, and mixed models at each boundary with subjects as the held-out unit. The continuous-crossover reading of N2-to-N3 is falsified if the order parameter shows a large jump and within-subject bimodality that survives jittered scoring boundaries and cycle-matched controls; the fold reading of wake-to-N1 is falsified if no reproducible tipping point with rising variance and autocorrelation precedes loss of wake stability; and the first-order reading of NREM-to-REM is falsified if forward and reverse paths overlap after matching sleep pressure, circadian phase, and prior sleep history.","tokens_in":26174,"feed_emoji":"🌙","tokens_out":14737,"duration_ms":133678,"temperature":0.7,"pith_summary":"This paper proposes that the boundaries between sleep stages are not one phenomenon but several: wake-to-N1 may be a fold-like loss of stability, N1-to-N2 and N2-to-N3 gradual ordering crossovers, NREM-to-REM a first-order-like desynchronizing switch, and a deep within-N3 regime a speculative mixed or tricritical-like transition. All four are presented as different parameter regimes of a single effective potential — a Landau-Ginzburg functional, an energy-like surface whose shape sets the stability and transition character of a noisy, spatially extended neural field — ordered by a latent cortical-ordering coordinate to be estimated from prespecified EEG/PSG features through a measurement model designed to prevent circularity. The paper's contribution is a testable organizational scheme: each boundary carries its own EEG/PSG signature (critical slowing, hysteresis and bimodality, correlation-length growth, or scoring artifact), and a synthetic classification experiment shows the six archetypes are separable above chance (cross-validated accuracy 0.49 versus a 0.17 baseline). The author is explicit that this establishes internal consistency and testability, not the taxonomy in human sleep; only the sleep-onset fold has prior external support, and the measurement model is specified but not yet fitted.","feed_headline":"Sleep-stage shifts may be fold, crossover, or switch events","feed_subtitle":"One noisy neural-field potential generates all four signature classes, each with its own EEG test.","key_machinery":"The carrying object is the effective Landau-Ginzburg functional, $$\\mathcal{F}[\\psi] = \\int_\\$\\Omega$ \\left(\\frac{a(\\$\\lambda$)}{2}\\$psi^{2}$ + \\frac{b(\\$\\lambda$)}{4}\\$psi^{4}$ + \\frac{c}{6}\\$psi^{6}$ + \\frac{\\kappa}{2}|\\nabla\\psi|^2 - h(\\$\\lambda$)\\psi\\right)d^d r,$$ with $\\psi = \\varphi - \\varphi_0$ the centered latent cortical-ordering coordinate, together with its stochastic relaxational dynamics, the time-dependent Ginzburg-Landau equation $\\partial\\psi/\\partial t = -\\Gamma(\\delta\\mathcal{F}/\\delta\\psi) + \\eta$. The coefficients carry the operational meaning: $a$ sets the stiffness of the current state (recovery time, variance, lag-1 autocorrelation), $b$ the transition character (smooth ordering versus bistability), $c$ high-amplitude saturation, $\\kappa$ the spatial coupling whose observable is the correlation length $\\xi = \\sqrt{\\kappa/V''(\\psi_{\\rm eq})}$, and $h$ the biasing field that rounds sharp transitions into crossovers. The latent coordinate is estimated, not assumed, through the prespecified measurement model $\\mathbf{y}(t) = \\Lambda\\varphi(t) + \\varepsilon(t)$, with the requirement that features defining a boundary be excluded from $\\varphi$ for that boundary's test. One Ginzburg-Landau family generates the fold, cusp-with-hysteresis, continuous-order, and tricritical signatures, which is what lets the paper assign different dynamical classes to different boundaries while keeping the description unified.","core_discovery":"The central claim is that different sleep-stage boundaries may instantiate different forms of dynamical reorganization, and that these forms can be told apart empirically with tools adapted from nonequilibrium statistical mechanics. Concretely: wake-to-N1 is a local fold (saddle-node) whose loss of wake stability has external empirical support, with the open question of whether that fold is a spinodal of a globally bistable cusp showing hysteresis; N1-to-N2 and N2-to-N3 are continuous-like ordering crossovers, with N2-to-N3 the strongest candidate and the spatial Ginzburg term predicting correlation-length growth and local-to-global slow-wave recruitment; NREM-to-REM is a candidate first-order-like switch between a synchronized high-ordering NREM basin and a desynchronized low-ordering REM basin; and a consolidated within-N3 subregime is a clearly marked speculative hypothesis of mixed or tricritical-like character. These are not separate models: one time-dependent Ginzburg-Landau equation family, $\\mathcal{F}[\\psi] = \\int_\\Omega \\left(\\frac{a(\\lambda)}{2}\\psi^2 + \\frac{b(\\lambda)}{4}\\psi^4 + \\frac{c}{6}\\psi^6 + \\frac{\\kappa}{2}|\\nabla\\psi|^2 - h(\\lambda)\\psi\\right)d^d r$, with $\\psi$ the centered latent cortical-ordering coordinate, produces all four signature classes as its coefficients move through fold, cusp, and crossover regimes. The paper claims the framework is internally consistent and testable, not that the taxonomy has been found in human sleep.","pith_inferences":["If the taxonomy is validated, sleep-disorder phenotyping could be re-expressed as changes in effective potential coefficients — basin depth at sleep onset, barrier height between NREM and REM, spatial coupling $\\kappa$ for slow-wave recruitment — rather than as changes in stage durations; the paper gestures at this direction but does not develop it.","A natural next experiment the paper does not run is to fit the time-dependent Ginzburg-Landau equation directly to EEG-derived $\\varphi$ trajectories per subject and boundary, converting the qualitative taxonomy into estimated parameter sets (sign and scale of $b$, field $\\tilde{h}$) that could be compared across ages or clinical groups.","The framework's 'one family, many regimes' structure suggests a sharper version of the tricritical test: high-density EEG nucleation maps — local patches of high $\\varphi$ expanding as traveling slow-wave fronts — would distinguish a mixed continuous-plus-switch subregime from a smooth crossover in a way that scalp-spectral averages cannot.","Because REM frequently emerges from N2, pooling N2-to-REM and N3-to-REM transitions could manufacture the bimodality the first-order hypothesis predicts; the paper flags stratification by source stage, and a reader might go further and treat source-stage mixing as a built-in control in any cohort test."],"forward_implications":["Sleep-onset analysis shifts from scalar models to spatial ones: the Ginzburg term predicts that slow waves recruit larger cortical territories as N2-to-N3 is approached, testable as correlation-length growth in high-density or source-reconstructed EEG.","Each boundary gets a distinguishing battery: critical slowing and a reproducible tipping point license the fold reading of wake-to-N1; hysteresis and path dependence that survive matched-control comparison license the first-order reading of NREM-to-REM; absence of bimodality and jump argues against first-order behavior at N2-to-N3.","Scoring is itself an archetype: a scoring-induced discontinuity and a smooth crossover share the same latent dynamics and are the least separable pair in the synthetic experiment, so the validation pipeline must jitter or re-estimate scored boundaries and exclude boundary-defining features from the order parameter.","A within-N3 consolidated-SWS substate, if present, should show a continuous precursor plus a sharper derivative change, local-to-global nucleation, or late bimodality; if high-density EEG shows none of these, the tricritical-like hypothesis is to be weakened or rejected.","Hypnogram labels are demoted to temporal priors: transition times and transition classes are inferred from continuous EEG/PSG features, with subjects rather than epochs as the held-out unit in model comparison."],"supporting_citations":[{"why":"Supplies the two-cohort empirical support for fold-like sleep onset with critical slowing; the only boundary in the taxonomy with prior external evidence.","marker":"[15]"},{"why":"Mean-field model treating SWS-to-REM as a first-order transition; the main theoretical precedent for the NREM-to-REM assignment.","marker":"[13]"},{"why":"Flip-flop model of sleep-wake bistability; provides the biological substrate for interpreting mutual inhibition as a bistable effective potential.","marker":"[23]"},{"why":"Landau-Ginzburg theory of the cortex producing synchronized, down-state, and avalanche regimes; licenses the nonequilibrium field language for sleep.","marker":"[7]"},{"why":"Defines critical-slowing early-warning signatures (rising variance, autocorrelation, recovery time) that the fold diagnostics are built on.","marker":"[24]"},{"why":"Fitted stochastic double-well model of sleep-onset EEG with noise-driven switching; evidence for the plausibility of the bistable cusp at wake-to-N1.","marker":"[29]"},{"why":"Originals of the Landau-Ginzburg effective-potential construction that the entire framework adapts to neural fields.","marker":"[10, 11]"},{"why":"Transition-level spectral analysis of a large PSG cohort; motivating feasibility evidence for transition-centered EEG analysis.","marker":"[49]"},{"why":"Public PSG dataset preserving R&K Stage 3 and Stage 4 separately; the proposed corpus for validating the within-N3 hypothesis.","marker":"[50]"}],"fun_headline_variants":["Sleep-stage boundaries: fold, crossover, or switch?","One Landau-Ginzburg potential, four transition classes","Four types of sleep-stage transitions, one theory","Landau-Ginzburg model: how sleep stages switch, fold, and cross over","Fold, crossover, or switch: each sleep boundary has an EEG test"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that one latent cortical-ordering coordinate, estimated from a prespecified linear measurement model, can capture the dynamics at each NREM boundary without using the features that define the boundary under test; the paper states plainly that this model is specified but not fitted and that fitting and invariance testing are deferred, so if that coordinate cannot be identified non-circularly, the taxonomy is unevaluable.","fun_headline_variants_meta":{"raw":{"variants":["Sleep-stage boundaries: fold, crossover, or switch?","One Landau-Ginzburg potential, four transition classes","Four types of sleep-stage transitions, one theory","Landau-Ginzburg model: how sleep stages switch, fold, and cross over","Fold, crossover, or switch: each sleep boundary has an EEG test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.003089,"raw_usage":{"total_tokens":11814,"prompt_tokens":1184,"completion_tokens":10630,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":800,"completion_tokens_details":{"reasoning_tokens":10542}},"tokens_in":800,"tokens_out":10630,"duration_ms":73540,"temperature":1.0,"reasoning_tokens":10542,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T04:15:43.352461+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's own proposed test on public polysomnography: estimate $\\varphi$ from features that exclude the boundary-defining ones, estimate transition times independently of the scored labels, and compare smooth, switching, and mixed models at each boundary with subjects as the held-out unit. The continuous-crossover reading of N2-to-N3 is falsified if the order parameter shows a large jump and within-subject bimodality that survives jittered scoring boundaries and cycle-matched controls; the fold reading of wake-to-N1 is falsified if no reproducible tipping point with rising variance and autocorrelation precedes loss of wake stability; and the first-order reading of NREM-to-REM is falsified if forward and reverse paths overlap after matching sleep pressure, circadian phase, and prior sleep history.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the two-cohort empirical support for fold-like sleep onset with critical slowing; the only boundary in the taxonomy with prior external evidence."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Mean-field model treating SWS-to-REM as a first-order transition; the main theoretical precedent for the NREM-to-REM assignment."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Flip-flop model of sleep-wake bistability; provides the biological substrate for interpreting mutual inhibition as a bistable effective potential."},{"cited_title":"Scheffer, J","cited_arxiv_id":null,"evidence_quote":"Defines critical-slowing early-warning signatures (rising variance, autocorrelation, recovery time) that the fold diagnostics are built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Fitted stochastic double-well model of sleep-onset EEG with noise-driven switching; evidence for the plausibility of the bistable cusp at wake-to-N1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Transition-level spectral analysis of a large PSG cohort; motivating feasibility evidence for transition-centered EEG analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Public PSG dataset preserving R&K Stage 3 and Stage 4 separately; the proposed corpus for validating the within-N3 hypothesis."}],"review_version":1}