{"id":"36fc44b4-472d-499c-b3ba-477cb636df92","arxiv_id":"2506.17280","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The paper introduces ROCOR, ROI, and TMR for continuous-time Markov systems and applies them to 18 wind farms, showing they expose transition dynamics hidden by static Weibull fits.","lead":"This paper defines three new time-varying reliability indices, the rate of repairs, a measure of internal transitions, and a total mobility rate, for Markov systems. It shows these indices, computed from wind-speed data, can tell wind farms apart even when their long-run wind distributions look the same.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The empirical section's matrix-logarithm embedding is the load-bearing step: Section 3.2 asserts a generator Q with P ≈ e^{QΔt} but gives no algorithm or existence conditions, and an invalid Q would invalidate every computed rate curve.","rationale":"The reader's weakest assumption is exactly the matrix-logarithm embedding, and I agree that it is the single most load-bearing concern. The entire empirical contribution—all four indicators in Figures 5–8, the mobility surfaces in Figures 9–10, and the claimed distinction between persistence-driven (Gansu) and transition-driven (Muppandal) regimes—is computed from a Q that is never shown to exist. Without a valid generator, the rates are not well-defined, so the wind-farm analysis cannot demonstrate the utility of the new measures. The theoretical portion stands on its own and is not challenged. The absence of error bars and the use of wind-speed states as proxies for failure are secondary: they weaken interpretation but are not as decisive as an invalid generator. The proposed concrete test directly checks whether Q exists and is a valid generator for every site, which would settle the concern.","tokens_in":18723,"tokens_out":6069,"duration_ms":76463,"concrete_test":"For each of the 18 farms, compute L = logm(P̂) from the published one-hour transition matrix, then verify three properties: (i) L is real; (ii) all off-diagonal entries of L are nonnegative and each row sums to zero within numerical tolerance; (iii) the relative residual ||P̂ - expm(L)||_∞ / ||P̂||_∞ is negligible (<1e-6). If any site fails these checks, the corresponding rate curves and the Gansu/Muppandal contrast in Section 3.6 are not valid CTMC indicators, and the embedding procedure must be disclosed and validated before the empirical claims are accepted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central mathematical development—Definitions 1, 3, 5; Theorems 2, 4, 6; Proposition 7—is internally consistent and correct. The load-bearing risk is in the empirical application. Section 3.2 estimates a discrete-time transition matrix P from hourly wind-speed counts and states: 'we embed the observed transition matrix into a continuous-time Markov process... This is achieved by solving a matrix logarithm problem: given the hourly transition probability matrix P, we compute a generator matrix Q ... such that P≈e^{QΔt}.' No algorithm, no existence conditions, and no handling of non-embeddable matrices are supplied. For equations (6), (8), (10), (24), and Proposition 7 to describe a CTMC, Q must satisfy q_ij ≥ 0 for i≠j and zero row sums. If the computed logarithm has negative off-diagonal entries or does not have zero row sums, then rof, ror, roi, and tmr are not rates of any continuous-time Markov process, and Figures 5–10 plus the persistence/transition classification are unsupported. A concrete red flag: the empirical P shows exact or near-zero probabilities for jumps of more than two wind-speed classes (Figure 3), while for an irreducible CTMC, e^{tQ} has strictly positive entries for every t>0; exact zeros in P are direct evidence that P=e^{Q} cannot hold exactly. The theory is not harmed by this, but the announced empirical added value is conditional on resolving it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper defines two new instantaneous reliability indicators for finite-state continuous-time Markov systems, the Rate of Occurrence of Repairs (ROCOR) and the Rate of Inoccurrence (ROI), and combines them with the classical ROCOF into a Total Mobility Rate (TMR). The theoretical part derives closed-form expressions for these rates, proves the identity roi(t)+rof(t)+ror(t)=tmr(t), and applies the framework to hourly wind-speed data from 18 wind farms, claiming that the indicators distinguish 'persistence-driven' from 'transition-driven' reliability behaviors even when Weibull parameters are similar. The mathematical derivations in Sections 2 and 3 are internally consistent, but the empirical application rests on an under-specified matrix-logarithm embedding of an estimated discrete-time transition matrix into a continuous-time generator.","tokens_in":19027,"tokens_out":7620,"duration_ms":95773,"significance":"If the empirical embedding is made rigorous, the paper offers a genuinely useful complement to ROCOF: the decomposition into failure, repair, and within-subset transition rates gives a time-dependent, direction-sensitive picture of Markov reliability systems, and the TMR identity (25) is a clean, correct summary. The proof of Theorem 6 is carefully developed, Proposition 7 is correct, and Remark 3 properly connects TMR to the known Geweke-Marshall-Zarkin mobility index. The wind-farm case study addresses a real operational problem and could be valuable for comparing sites, but the current empirical claims are conditional on an unvalidated numerical step and on the absence of any uncertainty quantification.","major_comments":[{"comment":"The embedding of the estimated hourly transition matrix P into a continuous-time generator Q via a matrix logarithm is the load-bearing step for every empirical result in the paper, but Section 3.2 gives no algorithm, no existence conditions, and no verification that the computed Q satisfies q_ij >= 0 for i != j and zero row sums. Without a valid generator, equations (6), (8), (10), (24), and (25), together with Figures 5–10, are not rates of any continuous-time Markov process, and the persistence/transition classification in Sections 3.5 and 3.6 is unsupported. The concern is concrete: Figure 3 shows that hourly jumps of more than two wind-speed classes are 'virtually absent', so if the estimated P contains exact zero entries, P = e^Q cannot hold for any irreducible Q because e^{tQ} has strictly positive entries for every t > 0. The approximation 'P ≈ e^{QΔt}' therefore needs an explicit error criterion, and the non-embeddable case must be handled explicitly, for example by a projection method, by direct estimation of Q from holding times, or by reporting discrete-time analogues of the indicators.","section":"Section 3.2"},{"comment":"The headline empirical claim that the indicators 'distinguish between sites with similar long-term wind profiles' is presented through point estimates only. In particular, Section 3.6 contrasts Gansu (TMR_infinity about 0.30 h^-1) with Muppandal (about 0.11 h^-1) as evidence of fundamentally different reliability profiles, but the paper provides no confidence bands, bootstrap intervals, or sensitivity analysis for any of the estimated rate curves. Because Q is itself estimated from a finite sample, the displayed differences could in principle be driven by estimation error, so the classification needs at least a bootstrap or simulation-based assessment before the applied conclusion can be accepted.","section":"Sections 3.5–3.6"}],"minor_comments":[{"comment":"The asymptotic formulas contain notation errors: 'rof(∞)=⟨α L, Q_W⟩' should be '⟨L_W, Q_W⟩', and 'L=[L_W, L_W]' should be '[L_W, L_F]'; the formulas for ror(∞) and roi(∞) should use L_F and L_W consistently.","section":"Section 2.1, Remark 1"},{"comment":"The text says the process starts in the calmest wind class 0–2 m/s, which is state 0, while Figure 5's caption says 'state 1'; state 1 corresponds to 2–4 m/s. Please reconcile this inconsistency.","section":"Section 3.3 and Figure 5"},{"comment":"The statement that the final incomplete segment of each day is 'ignored' appears to discard the legitimate one-hour transition from hour 23 to hour 0 of the next day. If the goal is to keep Δt = 1 h, those transitions should be included, or the exclusion should be justified.","section":"Section 3.2"},{"comment":"The claim that Gansu and Muppandal 'both have shape factors near k≈2.5' is inaccurate: Table 1 lists k = 2.09 for Gansu and k = 2.85 for Muppandal. Either revise the wording or choose sites with closer shape parameters.","section":"Section 3.6"},{"comment":"The term 'rate of inoccurrence' may mislead readers, since ROI counts transitions within the working and failure subsets rather than the absence of transitions; consider a more descriptive term or an explicit remark clarifying that ROI measures mobility inside the two subsets.","section":"Section 2, Definitions 3 and 5"},{"comment":"The paper does not state how the matrix logarithm was computed or whether code and data are available; a reproducibility statement and a description of the numerical method (for example, a Schur–Parlett algorithm) would be needed for the empirical figures to be auditable.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The theoretical part of the paper is sound and publishable in principle, and I do not see any issue with the attribution of Ding-hua's ROCOF formula or with the authors' use of the Geweke-Marshall-Zarkin mobility index. The main obstacle is the empirical section: the matrix-logarithm embedding must be validated and the uncertainty of the estimated indicator curves addressed. These are fixable within the scope of the manuscript, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The theoretical part is correct: ROCOR, ROI, and TMR are well-defined, Proposition 7 holds, and the proofs (especially Theorem 6) are sound. But the measures are not new in any deep sense—ROCOR is ROCOF with working and failure states swapped, ROI is the internal-transition component of the total jump rate, and TMR is the expected jump rate whose asymptotic version the paper itself attributes to Geweke, Marshall, and Zarkin (1986). The time-dependent formulation is a reasonable packaging, not a breakthrough.\n\nThe wind-farm application is where the paper has a real problem. Section 3.2 says the hourly transition matrix P is embedded into a continuous-time generator Q by solving a matrix log problem, P≈e^{QΔt}, with no algorithm, no existence conditions, and no verification that Q is a valid generator. If Q has negative off-diagonal entries or row sums that are not zero, then rof, ror, roi, and tmr are not rates of any continuous-time Markov chain, and every figure in the empirical section is unsupported. This is not a small detail; it is load-bearing. The exact-zero rows in Figure 3 are a concrete red flag: for an irreducible CTMC, e^{tQ} is strictly positive for every t>0, so the observed zeros cannot be exactly matched. The authors need to explain how they computed Q and show that it is a valid generator.\n\nTwo smaller issues. The reliability interpretation uses wind-speed bins as proxies for failure, which is fine but should be acknowledged as a modeling choice rather than actual failure data. And the site comparisons come without error bars or any uncertainty quantification, so the 'persistence-driven versus transition-driven' contrasts may not be statistically real.\n\nBottom line: the mathematics is solid, the empirical section is not convincing as written, and the claims about distinguishing sites need validation. Still, this deserves a serious referee. The core idea is defensible and the application domain is relevant; a good referee could push the authors to fix the embedding, add validation, and report uncertainty. I would not cite it in its current form, but I would send it to review.","headline":"The theory is correct but elementary, and the wind-farm application rests on an unjustified matrix-logarithm embedding that invalidates the empirical claims as written.","tokens_in":19563,"tokens_out":3907,"would_cite":false,"duration_ms":45664,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J27","60K10","90B25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper introduces two new instantaneous rates, ROCOR and ROI, that complement ROCOF for Markov systems, and proves an identity that accounts for every transition as failure, repair, or internal movement.","keywords":["Markov processes","reliability","ROCOF","ROCOR","rate of inoccurrence","total mobility rate","wind farm","wind-speed states"],"falsifier":"Re-estimate the rates without the embedding step by treating the hourly transition counts directly as discrete-time estimators of mobility, or by computing $Q$ from an explicit embedding algorithm and checking whether $e^{Q}$ reproduces the estimated transition matrix within sampling error. If the Gansu-versus-Muppandal difference in long-run total mobility ($\\sim0.30$ versus $\\sim0.11\\,\\mathrm{h}^{-1}$) collapses or reverses under this re-estimation, the claim that the indicators distinguish the two sites would be put in doubt.","tokens_in":18516,"feed_emoji":"🌬️","tokens_out":12955,"duration_ms":137158,"temperature":0.7,"pith_summary":"This paper argues that the Rate of Occurrence of Failures (ROCOF), alone, gives an incomplete instantaneous picture of a Markov reliability system. It introduces two companion rates, the Rate of Occurrence of Repairs (ROCOR), which counts transitions from failure states back to working states, and the Rate of Inoccurrence (ROI), which counts transitions that stay inside the current working or failure subset, and it proves explicit formulas for both in terms of the generator matrix. It then packs the three rates into a Total Mobility Rate (TMR) and proves the identity $\\operatorname{roi}(t)+\\operatorname{rof}(t)+\\operatorname{ror}(t)=\\operatorname{tmr}(t)$, so every infinitesimal transition is accounted as failure, repair, or internal movement. On hourly wind-speed data from 18 wind farms, the rates separate sites that have similar Weibull fits into persistence-driven and transition-driven reliability profiles, which would matter for scheduling maintenance and sizing fast reserves. If the claim is right, reliability analysis gains a direction-sensitive, time-resolved toolkit that static summaries cannot provide.","feed_headline":"New rates show how wind farms recover, stall, and persist","feed_subtitle":"They separate persistence-driven from transition-driven sites even when static wind fits look identical.","key_machinery":"The load-bearing object is the generator matrix $Q=(q_{h,j})$ of a continuous-time Markov chain, whose off-diagonal entries are instantaneous transition rates and whose row sums are zero. Partitioning $Q$ according to the working/failure split gives the block structure that separates failure transitions ($W\\to F$), repair transitions ($F\\to W$), and internal transitions (within $W$ or within $F$); each new rate is then an inner product of the unconditional occupancy probabilities $p_h(t)=\\sum_i\\alpha_i(e^{tQ})_{i,h}$ with the corresponding exit-intensity vector. The connective mechanism is the counting identity $N(t)=N_f(t)+N_r(t)+N_i(t)$, which turns the bookkeeping of every transition into three directionally distinct rates and yields the total mobility identity. In the empirical part, the same generator is recovered by embedding the estimated hourly transition matrix into continuous time through a matrix logarithm.","core_discovery":"For a finite-state continuous-time Markov chain with state space $E$ partitioned into working states $W$ and failure states $F$, the paper defines ROCOR as the instantaneous expected rate of transitions $F\\to W$, and ROI as the instantaneous expected rate of transitions within $W$ or within $F$. It derives the closed forms $\\operatorname{ror}(t)=\\sum_{i\\in E}\\sum_{f\\in F}\\sum_{w\\in W}\\alpha_i(e^{tQ})_{i,f}q_{f,w}$ and an analogous sum for ROI, using only the generator $Q$ and the initial distribution $\\alpha$. Proposition 7 states the decomposition $\\operatorname{roi}(t)+\\operatorname{rof}(t)+\\operatorname{ror}(t)=\\operatorname{tmr}(t)=\\sum_{i}\\sum_{h}\\alpha_i(e^{tQ})_{i,h}\\sum_{j\\neq h}q_{h,j}$, and the paper notes TMR is independent of the partition into working and failed states, recovering the long-run mobility index of the literature as its limit. Applied to hourly wind-speed records discretized into eleven states, with working states chosen as the operational wind classes, the indicators are computed for eighteen wind farms and claimed to expose 'reliability logics' that static Weibull parameters and ROCOF alone do not reveal.","pith_inferences":["Inference: the same counting decomposition should carry over to semi-Markov reliability models, replacing the exponential holding times and $e^{tQ}$ occupancy with the transition probabilities of a semi-Markov kernel, although the paper proves the formulas only for the Markov case.","Inference: because the empirical 'failures' are defined by wind-speed thresholds rather than actual component failures, the rates are a proxy for operational regime dynamics; comparing them with turbine downtime logs would test whether the inferred reliability logics match real outage behavior.","Inference: the strong dependence on the initial state shown in the transient curves suggests a practical forecasting use not developed in the paper: set the initial distribution to the current wind state of a site and read the resulting ROCOR and ROCOF peaks as short-term transition risk warnings.","Inference: the same partition-independent TMR could be applied outside reliability, for instance to health-state, credit-rating, or labor-market Markov chains, wherever one wants a time-varying summary of how often agents change category."],"forward_implications":["Operators can classify a wind farm by its reliability logic: a flat high TMR plateau with strong ROI signals persistence and allows longer maintenance windows, while short-lived ROCOR and ROCOF bursts signal transition-driven volatility that needs fast reserves.","The identity $\\operatorname{roi}+\\operatorname{rof}+\\operatorname{ror}=\\operatorname{tmr}$ gives a monitoring decomposition: any change in total mobility can be attributed to more failures, more repairs, or more internal churn, so each movement direction is separately tracked over time.","Because TMR does not depend on the working/failure partition, the same function can be used as a generic time-dependent mobility measure for any finite-state continuous-time Markov process.","Sites with nearly equal Weibull scale and shape parameters can differ by a factor of roughly three in long-run total mobility (about $0.30\\,\\mathrm{h}^{-1}$ versus $0.11\\,\\mathrm{h}^{-1}$ in the contrasted pair), so static wind fits are not a reliable proxy for operational dynamism."],"supporting_citations":[{"why":"Supplies the ROCOF formula for Markov processes that ROCOR and ROI extend by symmetric counting arguments.","marker":"Ding-hua (1985)"},{"why":"Provides the bounding lemma on the number of transitions in a small time interval used in the proof of the ROI formula, and analyzes ROCOF for denumerable Markov chains.","marker":"Yeh (1997)"},{"why":"Defines the long-run expected rate of state changes in continuous-time Markov chains that TMR generalizes to a time-dependent function.","marker":"Geweke, Marshall, and Zarkin (1986)"},{"why":"Cited as the source for the availability and reliability formulas and their nonparametric estimation that the block generator representation relies on.","marker":"Sadek and Limnios (2005)"}],"fun_headline_variants":["New rates expose wind farms' persistence vs transition logic","Markov reliability rates now include repair and mobility","Wind farm dynamics: beyond ROCOF with three new rates","Rates for wind farm recovery, stall and persistence unveiled","TMR, ROCOR, ROI: new lens on wind farm reliability"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The wind-farm analysis assumes that the one-hour transition matrix estimated from the data has a matrix logarithm giving a valid continuous-time generator $Q$ with nonnegative off-diagonal rates and zero row sums; the paper states no algorithm, existence condition, or validity check for this embedding, and if the computed $Q$ is not a valid generator then every continuous-time indicator in the empirical study is undefined.","fun_headline_variants_meta":{"raw":{"variants":["New rates expose wind farms' persistence vs transition logic","Markov reliability rates now include repair and mobility","Wind farm dynamics: beyond ROCOF with three new rates","Rates for wind farm recovery, stall and persistence unveiled","TMR, ROCOR, ROI: new lens on wind farm reliability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001014,"raw_usage":{"total_tokens":4354,"prompt_tokens":1088,"completion_tokens":3266,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":704,"completion_tokens_details":{"reasoning_tokens":3184}},"tokens_in":704,"tokens_out":3266,"duration_ms":24577,"temperature":1.0,"reasoning_tokens":3184,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:51:48.170892+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-estimate the rates without the embedding step by treating the hourly transition counts directly as discrete-time estimators of mobility, or by computing $Q$ from an explicit embedding algorithm and checking whether $e^{Q}$ reproduces the estimated transition matrix within sampling error. If the Gansu-versus-Muppandal difference in long-run total mobility ($\\sim0.30$ versus $\\sim0.11\\,\\mathrm{h}^{-1}$) collapses or reverses under this re-estimation, the claim that the indicators distinguish the two sites would be put in doubt.","supporting_citations":[],"review_version":1}