{"id":"f825c0b4-de0e-476f-b9f8-be2fc329fa11","arxiv_id":"2607.25214","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Node degree in a visibility graph ranks data points so that global and local extreme values occupy the top positions, usable even when the time series is nonstationary.","lead":"This paper maps time series into visibility graphs and uses each point's number of connections (node degree) to rank data values, arguing that large spikes and locally prominent peaks sit at the top of this ranking. Because the degree–value relation is monotonic and parameter-free, the authors propose it as a way to spot extreme values in both stationary and nonstationary records, including climate data.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The nonstationary half of the central claim rests on an untested link between VG degree and local prominence; for random walks the paper itself concedes k and x are unrelated, so top-degree nodes are not shown to be extreme values.","rationale":"The reader's weakest assumption identifies exactly the most load-bearing vulnerability: for nonstationary processes, the paper provides no ground-truth validation that degree ranking marks local extremes. My independent read of the manuscript confirms this. The theoretical anchor (Eq. 2) is explicitly restricted to stationary processes, and the Appendix's Fig. 9E removes any ambiguity about the random-walk case: k and x are unrelated. The VG-stationarity of the degree distribution is real (Refs [11,15]) but only ensures that the degree sequence has stable statistics; it does not connect degree to data-value prominence. The GMT and random-walk figures are therefore illustrative, not evidential. I considered whether the M free parameter or the stationary rank mismatches in Table 1 might be more fundamental, but those are secondary: the stationary mechanism is grounded in a monotone average relation and the mismatches can be interpreted as detecting local rather than global extremes. The nonstationary claim, by contrast, has no independent criterion against which to check. Since the reader already assigned CONDITIONAL and the concern does not demand outright rejection — the stationary contribution and peak-subsampling remain useful — the appropriate verdict is unchanged.","tokens_in":13430,"tokens_out":4884,"duration_ms":55419,"concrete_test":"Simulate an ensemble of nonstationary processes with known ground-truth local extremes: e.g., Gaussian random walks plus randomly placed transient pulses of varying amplitude (known times), and fBm with H ∈ {0.3, 0.5, 0.7}. For each realization, build the VG (full and peak-subsampled), and rank nodes by degree. Define ground-truth events as pulse windows (or, for fBm, exceedances of a time-local detrended residual). Compute precision/recall/rank correlation of top-M degree nodes versus (i) top-M height, (ii) rolling-window z-score, (iii) random selection, sweeping M. If degree ranking does not significantly exceed the rolling-window baseline or random selection, the nonstationary claim is falsified. Also compute E[degree | local-extremeness] and E[x | k] for these processes; if neither is monotone, there is no theoretical bridge from degree to local prominence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The nonstationary half of the central claim is not supported by any validated link between VG degree and an independent notion of local extremeness. §3 states that Eq. (2) 'completely breaks down for nonstationary processes,' and the Appendix explicitly concedes that for an unbiased random walk 'there is no relation really between k and x' (Fig. 9E). The paper then pivots to VG-stationarity of the degree distribution (Refs [11,15]) to argue that hubs are meaningful, but stationarity of P(k) only says the graph's marginal degree distribution is time-invariant; it says nothing about whether high-degree nodes are locally prominent in the signal. The GMT and random-walk demonstrations (Figs. 7–8) select top-six nodes by degree and visually label them as 'locally important events' or local records, with no independent event list, no null model, and no comparison to a baseline detector. Because for nonstationary data there is no well-defined marginal distribution f(x) against which to call a point an outlier, the degree-ranking output cannot be validated as 'extreme values'; the claim reduces to 'hubs are hubs.' Since the abstract's 'effective alternative' for nonstationary data rests entirely on this step, it is the load-bearing assumption. The stationary half of the paper is substantially better supported (Eq. 2 + Appendix A–D), so the concern is not about the whole method but specifically about extending the conclusion to nonstationary processes.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes using node degree in the visibility graph (VG) as a parameter-free detector of extreme values in time series. For stationary series, the method builds on an established monotonic, nonlinear relation between data value and ensemble-averaged degree (Eq. 2), arguing that this relation, together with its logarithmic amplification, justifies ranking data points by degree. For nonstationary series, the paper asserts that degree ranking remains a 'robust indicator of relative importance' even though amplitude-based extremes are ill-defined. The method is illustrated on a simulated laser rogue-wave series, the Niño3.4 SST index, a global temperature anomaly index, and an unbiased random walk. The authors also propose subsampling the series to local maxima to reduce computational cost while preserving detection.","tokens_in":13788,"tokens_out":5667,"duration_ms":54484,"significance":"The stationary half of the paper is grounded in a concrete external result (Eq. 2 is an exact prior result for HVGs of uncorrelated processes) and the nonlinear-amplification mechanism is a plausible and potentially useful idea. If the stationary claims were quantitatively validated, the method would be a simple, parameter-free complement to threshold-based EV detection. The nonstationary extension, which is central to the abstract and conclusions, is currently an unsupported assertion: for random walks the paper concedes that no relation between k and x exists, and the claimed robustness of degree ranking for nonstationary data is not tested against any ground truth. The paper would be significantly improved by either restricting its claims to stationary processes or by validating the nonstationary claim with a synthetic model with known local extremes and a baseline comparison.","major_comments":[{"comment":"The central stationary inference leaps from an ensemble-averaged relation to a per-node statement. Eq. (2) is an average over realizations for HVG and is admitted to be 'only an approximation' for VG (Fig. 9B). The Appendix states that the logarithmic divergence 'suggests' that rank(k)=rank(x) holds for finite-sample extremes, but no proof or numerical test of this is provided. The abstract's claim that extreme values 'systematically occupy the top positions' requires a quantitative statement, such as the expected overlap between the top-m by height and top-m by degree under a given process, against a null model. The visual selection of the first six positions in Figs. 4–6 is illustrative, not a validation.","section":"§3 and Appendix, Fig. 9"},{"comment":"The nonstationary half of the central claim is unsupported. The paper states that Eq. (2) 'completely breaks down for nonstationary processes' and the Appendix explicitly concedes for an unbiased random walk 'there is no relation really between k and x' (Fig. 9E). The subsequent argument that VG-stationarity of P(k) (Refs. [11,15]) makes hubs a meaningful extreme-event indicator does not follow: stationarity of the degree distribution says nothing about whether high-degree nodes are locally prominent in the signal. For the random walk there is no independent notion of an extreme to check against, and for the GMT record (Fig. 7) the top-degree nodes are labeled as 'locally important events' with no external event list or baseline detector. The claim that degree ranking provides 'a robust indicator of relative importance' for nonstationary data is therefore an assertion. The manuscript sho","section":"§4.2 and Appendix Fig. 9E"},{"comment":"The 'joint ranking' methodology is not quantified. In Table 1, the top-6 by height and the top-6 (or top-7, due to a tie) by degree only partially overlap; for example, the sixth-ranked height point (index 297) is absent from the top degree ranks, while indices 553 and 930 appear in the top-degree list despite being ranked 27th and 10th by height. The paper neither reports a rank-correlation measure nor evaluates whether the degree-selected points are actually 'locally extreme' according to an independent criterion. Because the method's purpose is to identify EVs that height-based ranking misses, a quantitative evaluation (e.g., precision/recall against a synthetic ground truth or a threshold-based reference) is essential to support the claim that 'extreme values systematically occupy the top positions in these rankings.'","section":"§4.1, Table 1 and §5"}],"minor_comments":[{"comment":"The green threshold line in Fig. 3(a) is not defined; if it is a free parameter, this should be stated and its choice justified since the paper emphasizes parameter-free detection.","section":"Fig. 3"},{"comment":"The authors note that the number of positions kept in the rankings is a free parameter, but no guidance is given for choosing it. This is relevant because the method's output depends on this choice.","section":"§4.1"},{"comment":"Eq. (2) is derived for HVGs, yet the paper exclusively uses VGs and notes the relation is only approximate there. This should be stated prominently in the main text, not only in the Appendix.","section":"§3"},{"comment":"The caption says for a random walk 'there is no relation really between k and x,' while the text immediately concludes that 'k or rank(k) could be used here' for extreme-event detection. This apparent contradiction needs to be resolved explicitly: either the meaning of 'extreme' is being redefined as a hub, or the relation is not as absent as suggested.","section":"Appendix Fig. 9E"},{"comment":"The phrase 'extreme events ... definition ... in terms of degree rankings is still perfectly possible' redefines extremes as graph hubs. This conceptual shift should be stated explicitly and discussed, since it is not the standard notion of an extreme value.","section":"§4.2"}],"recommendation":"major_revision","confidential_remarks":"The paper cites earlier VG-based extreme-event works [36–38] but does not position itself against them; a comparison with these methods would strengthen the contribution. I also note that the nonstationary claim is likely to be controversial because 'extreme value' for nonstationary data is redefined through graph degree without a formal definition; the authors should be encouraged to either narrow the claims or provide a rigorous validation of the redefined notion."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: the genuinely useful piece here is the stationary-case protocol. Ranking points by VG node degree, taking the union with height ranking, and subsampling peaks to cut cost — that combination is new, and the ORW and Niño examples show it catching local extremes that a global threshold misses. The theoretical anchor (Eq. 2, from Luque-Lacasa) is prior work, but the paper uses it honestly, noting that it is exact for HVG and approximate for VG.\n\nThe soft spot is the nonstationary claim. The abstract says degree ranking 'provides a robust indicator of relative importance' for nonstationary processes, but the Appendix says for an unbiased random walk 'there is no relation really between k and x' (Fig. 9E). The paper never tests the nonstationary half against any ground truth: no independent event list for the temperature record, no null model, no baseline comparison. That is load-bearing, not cosmetic. There are also smaller issues: 'no external parameters' sits awkwardly with the admitted M free parameter, and the one threshold shown in Fig. 3(a) is not given a value, so no quantitative benchmark exists.\n\nThe paper is transparent about most of these gaps, which earns credit. But as written, the conclusion overreaches. I would send it to peer review — the stationary protocol deserves a serious referee — with a request for major revision: benchmark against a threshold-based method, report sensitivity to M and tie-handling, and either provide independent evidence for the nonstationary claim or cut it from the abstract. If the nonstationary half is fixed, this becomes a much stronger paper.\n\nRead it if you work on extreme-event detection in climate or optics; the stationary part is worth the time. I wouldn't cite it yet.","headline":"Stationary-case degree ranking is a useful, cheap extreme-value detector; the nonstationary half is asserted, not shown, and the paper's own appendix concedes the degree–value link dies for random walks.","tokens_in":14321,"tokens_out":3111,"would_cite":false,"duration_ms":31898,"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":"Ranking a time series by the degree of its visibility-graph nodes identifies extreme values—global and local—without thresholds or detrending.","keywords":["extreme events","extreme values","time series analysis","visibility graph","node degree","nonstationary processes","peak subsampling","climate data"],"falsifier":"Take a long random-walk time series, independently define 'locally extreme' events as record-high excursions above a local window (or via a separate event catalog for temperature data), and check whether the top-ranked visibility-graph degree nodes coincide with those events significantly better than random. If they do not, the nonstationary half of the claim collapses.","tokens_in":13232,"feed_emoji":"📈","tokens_out":6990,"duration_ms":58605,"temperature":0.7,"pith_summary":"The paper tries to establish that the degree of a node in a visibility graph—a network built from a time series by connecting mutually visible points—carries enough information to rank data values by extremeness. For stationary series, the node degree is monotonically linked to the data value (through a logarithmic amplification), so high-degree nodes are the largest or locally most prominent values. For nonstationary series, where threshold-based methods lose meaning because the marginal distribution drifts, the paper claims degree ranking still highlights relatively important events. The authors validate the approach on simulated laser rogue waves, an El Niño index, global temperature anomalies, and a random walk, and show that restricting to peak values preserves detection while cutting computational cost. If correct, the method offers a parameter-free, detrend-free extreme-event detector.","feed_headline":"Rank extreme values by visibility-graph degree, no thresholds","feed_subtitle":"Network structure highlights global and local spikes in stationary and nonstationary time series alike.","key_machinery":"The central object is the visibility graph (VG): a network whose nodes are time-series points and whose edges connect two points if every intermediate point lies below the straight line between them (Eq. 1). The load-bearing identity is the ensemble-averaged relation k(x)=2−2ln[1−F(x)], proven for horizontal visibility graphs and empirically approximated for VGs across stationary processes. This monotone, logarithmically amplifying relation lets the paper replace amplitude ranking with degree ranking; the same degree concept is carried to nonstationary signals, where the formula no longer holds but degree is argued to remain stable and informative. A peak-subsampling step (keeping only local","core_discovery":"The central claim is that jointly ranking data points by their amplitude and by the degree of their associated visibility-graph node separates extremes from the bulk: extreme values systematically occupy the top positions in both rankings. The mechanism is a known relation (exact for horizontal visibility graphs, approximate for visibility graphs) between a node's degree k and the data value x: k(x) = 2 − 2 ln[1 − F(x)], with F the cumulative distribution. Monotonicity makes degree rank a proxy for value rank, while the logarithmic divergence at the upper tail amplifies large values, so the correspondence is strongest exactly at the extremes. For nonstationary processes the formula is void,","pith_inferences":["If the degree-ranking claim generalizes, a natural extension is to turn the top-degree nodes into a candidate list that a downstream extreme-event catalog (e.g., heatwave definitions) can validate, giving precision-recall numbers.","The degree–value monotonicity is exact only for uncorrelated series; for multi-scale or long-memory processes, the rank correspondence may degrade, so a useful benchmark is measuring how often the top-degree node is also the top-value node across correlation lengths.","Peak subsampling removes the redundancy of multiple points belonging to one extreme event, which suggests a fast pre-filtering step: compute degrees only at local maxima and use that ranking as an inexpensive extreme-event screener for very long series.","The random-walk case (Appendix Fig. 9E) shows no direct k–x relation, so a decisive test is to compare degree-hub detections against independent local-extreme definitions on surrogate nonstationary data; if hubs are generic structural artifacts of the graph, the claim would lose its evidential basis."],"forward_implications":["For stationary time series, high-degree nodes provide a parameter-free identification of both global and local extreme values, complementing or replacing threshold-based detection.","The logarithmic amplification in the degree–value relation means the degree ranking is more reliable for true extremes than for mid-range values, suppressing noise.","Subsampling to peaks yields a nearly identical ranking while reducing computational cost, allowing the method to scale to long series.","For nonstationary series—where mean or variance drift makes threshold outlier definitions ill-posed—degree ranking still yields a ranked list of locally important events without detrending or window choices.","The method is demonstrated on real climate data (Niño3.4, global temperature anomaly), indicating it can be directly applied to observational records."],"fun_headline_variants":["Graph degree ranks extreme events without thresholds","Visibility graphs spot extremes in any time series","No-threshold extreme detection via network topology","Node degree reveals extreme values in nonstationary data","Ranking spikes by visibility-graph connectivity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that, for nonstationary processes, a node's degree remains a meaningful marker of local importance even though the theoretical degree–value link is broken; the paper asserts this without testing it against any independent ground-truth event list.","fun_headline_variants_meta":{"raw":{"variants":["Graph degree ranks extreme events without thresholds","Visibility graphs spot extremes in any time series","No-threshold extreme detection via network topology","Node degree reveals extreme values in nonstationary data","Ranking spikes by visibility-graph connectivity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000152,"raw_usage":{"total_tokens":1070,"prompt_tokens":801,"completion_tokens":269,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":202}},"tokens_in":545,"tokens_out":269,"duration_ms":3266,"temperature":1.0,"reasoning_tokens":202,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:07:25.849174+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a long random-walk time series, independently define 'locally extreme' events as record-high excursions above a local window (or via a separate event catalog for temperature data), and check whether the top-ranked visibility-graph degree nodes coincide with those events significantly better than random. If they do not, the nonstationary half of the claim collapses.","supporting_citations":[],"review_version":1}