{"id":"105309f0-a837-4f9f-955f-627eefac66ab","arxiv_id":"1908.10862","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Gibbs-sampled wind and demand scenarios are fed into a Stackelberg capacity-investment game to estimate equilibrium renewable capacities for a UK private network upgrade.","lead":"Private investors can build power lines and charge local wind farms for access, but choosing how much generation capacity to install is a strategic game under uncertain wind and demand. This paper combines Gibbs sampling with a leader-follower game to simulate many scenarios and estimate equilibrium capacities for a real UK network upgrade.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm 2 as written returns negative curtailment: RD=PD-(PG1+PG2) and PCi=PGi*RD/(PG1+PG2) give PCi<0 whenever generation exceeds demand, inverting the sign of the curtailment that drives the Stackelberg equilibrium.","rationale":"The reader's CONDITIONAL verdict is appropriate, and the sign error in Algorithm 2 reinforces it. The reader identified the demand-conditional distribution as the weakest assumption, but also noted a 'sign error in the curtailment pseudocode' in the rationale. I agree that the demand model is an unvalidated load-bearing choice, but the sign error is more fundamental because it is an internal contradiction between the theoretical definition of curtailment (positive) and the implemented algorithm (negative). A corrected pseudocode and an explicit statement of the sign convention would make the methodology reproducible, which is exactly the kind of condition the reader's verdict anticipates. The reported convergence of the Gibbs sampler (Tables I and II) is internally consistent and supports the sampling part of the claim, but it cannot rescue the downstream game computation if the curtailment values are negative. Therefore the verdict remains CONDITIONAL: the article should be accepted only after the algorithm is corrected, the sign convention is stated, and the numerical results are regenerated and shown to be robust to this correction. No change to the reader's verdict is needed; the concern deepens the justification for the conditional rather than altering the decision.","tokens_in":10042,"tokens_out":5309,"duration_ms":49477,"concrete_test":"Re-implement Algorithm 2 exactly as published (RD <- PD - (PG1 + PG2); if RD > 0 set RD <- 0; PCi <- PGi * RD / (PG1 + PG2)) on the same 40,000 Gibbs-sampled data points and capacity grid used in the paper. Compute EC1 and EC2 for any grid point where PG1 + PG2 > PD. If these sums are negative, the published algorithm cannot generate the positive curtailment values required by Eqs. (1)-(3). Then rerun the full Stackelberg equilibrium computation with the corrected sign (e.g., RD <- PG1 + PG2 - PD, or PCi <- PGi * (PG1 + PG2 - PD) / (PG1 + PG2)). If the resulting (P_N1*, P_N2*) differ from Fig. 2 by more than the reported 10 MW range, the headline results are an artifact of the missing sign correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the simulated curtailment events, which determine the equilibrium capacities and profits, are computed correctly. Algorithm 2 defines RD (residual demand) as PD - (PG1 + PG2). When generation exceeds demand (the curtailment event), RD < 0, which is exactly the case the algorithm must handle. Lines 11-12 then compute PC1 = PG1*RD/(PG1+PG2) and PC2 = PG2*RD/(PG1+PG2). For RD < 0, these values are negative. The theory in Eq. (3) and the allocation rule require positive curtailment, e.g., PGi/(PG1+PG2) * (PG1+PG2 - PD) or equivalently -PGi*RD/(PG1+PG2). As written, EC1 and EC2 in lines 16-17 are sums of negative numbers, so in the profit equations (1)-(2) the term (EGi - ECi) exceeds EGi, inflating revenues and distorting the best-response calculations. This is not a cosmetic typo: the pseudocode implements the opposite sign of curtailment in precisely the region of state space that defines the game. If the reported results were produced by code matching the pseudocode, the results in Figs. 2-3 are not trustworthy. If the actual code used a corrected sign, then the published algorithm omits a sign convention essential for reproducibility. The paper provides no clarification or erratum, so the central claim that the generated scenarios can be used to estimate optimal capacities is not reproducible from the published algorithm. This internal inconsistency is more load-bearing than the reader's identified demand-conditional assumption, because the demand model is at least a coherent modeling choice, whereas the sign error makes the computational method self-contradictory with its own theoretical equations.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a methodology that combines Gibbs sampling with a Stackelberg game model to determine optimal generation capacities for a private network upgrade. Wind speed and demand scenarios are generated from historical data via a Gibbs sampler, and these scenarios are used to compute expected generation, curtailment, and profits for a line investor (leader) and local generators (follower). The approach is applied to the Kintyre-Hunterston link in the UK, and the paper reports convergence diagnostics for the sampler and equilibrium capacities and profits across multiple realisations and cost scenarios.","tokens_in":10393,"tokens_out":2644,"duration_ms":27111,"significance":"If the methodology is sound, the paper offers a useful template for using MCMC scenario generation in game-theoretic investment models with stochastic renewable resources, which is a relevant problem for network planning. The authors provide a reasonably detailed description of the Gibbs sampler and report internal convergence checks (Tables I and II) that show the expected central-limit behavior for increasing sample size. However, the central algorithm contains a sign error in the curtailment calculation that directly affects the equilibrium results, and the conditional demand model is not validated. These issues must be resolved before the claims can be trusted.","major_comments":[{"comment":"The curtailment calculation has a sign error. With RD = PD - (PG1 + PG2), a curtailment event corresponds to RD < 0, but lines 11–12 compute PC1 = PG1 * RD / (PG1 + PG2) and PC2 = PG2 * RD / (PG1 + PG2), which are negative when generation exceeds demand. This reverses the sign of curtailment, so the terms (EGi - ECi) in Eqs. (1)–(2) exceed EGi, inflating revenues and distorting the best-response calculations. This is not a cosmetic typo: the published algorithm implements the opposite sign in exactly the state-space region that defines the game. If the reported results were produced by code matching the pseudocode, the equilibrium capacities and profits in Figs. 2–3 are not trustworthy; if the code used a corrected sign, the pseudocode omits a sign convention essential for reproducibility. The paper provides no clarification or erratum, so the central claim is not reproducible from the published algorithm.","section":"Algorithm 2, lines 7–12"}],"minor_comments":[{"comment":"The sentence 'curtailment expressions for each player under a “common access” regime can be reasonably can be approximated by ECi = EGi/(EGi + EG−i) EC' contains a duplicated phrase 'can be reasonably can be'; please revise to 'can be reasonably approximated by'.","section":"Section II"},{"comment":"Reference [3] lists 'IEEE 2012 PES ISGT Europe' but is dated 2016; either the conference year or the publication year is inconsistent, and the venue should be checked.","section":"References"},{"comment":"In Table II, the maximum error (ME) for wind speed does not decrease as the number of realisations N increases (e.g., ME for w1 goes from 6.64% at N=100 to 9.87% at N=50,000). The text explains that WCI narrows, but the ME behavior is not discussed; please clarify whether ME is the deviation of the overall sample mean from the historic mean and why it does not improve with N.","section":"Tables I and II"},{"comment":"The text states 'µw1 = 12.1029, µw1 = 12.1950' for the historic means; the second symbol should likely be µw2. Please correct the notation.","section":"Section IV.B, text after Table I"}],"recommendation":"major_revision","confidential_remarks":"The sign error in Algorithm 2 is severe enough that, if it reflects the actual implementation, the reported numerical results are invalid. Even if the code is correct, the published algorithm is not reproducible as written. The authors must correct the pseudocode, rerun the simulations, and report whether the equilibrium results change. In addition, the demand model (Algorithm 1, line 11) is a non-standard assumption that needs empirical justification; without it, the simulated curtailment events may be unrealistic. If the authors cannot provide the corrected implementation, the paper may need to be rejected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper is a plausible integration of Gibbs sampling with a Stackelberg game for private network upgrades, and it is applied to a real UK project. But there is a sign error in the curtailment pseudocode that, if it reflects the actual code, flips the sign of curtailment exactly when curtailment matters. That makes the headline results untrustworthy until fixed or explained.\n\nWhat is actually new: the authors combine a two-player Stackelberg investment game from their prior work [11] with an empirical Gibbs sampler that generates wind speed and demand scenarios from historic data. The convergence checks in Tables I and II are sound as internal consistency checks: as n increases, the sample means converge to the historic means and the standard deviations behave as expected. The case study is real and the scenario analysis around costs is informative. So the building blocks are solid and the integration is a legitimate step beyond [11].\n\nThe soft spots, in order of severity. First, Algorithm 2 as printed computes RD = PD - (PG1 + PG2); when generation exceeds demand—the curtailment event—RD is negative, but the code then computes PCi = PGi * RD / (PG1 + PG2). That gives negative curtailment. The sign is exactly backwards. The effect is to add curtailed energy back into revenues, which will distort the best-response calculation and the equilibrium. This is not a cosmetic typo; it sits in the heart of the profit calculation. If the actual code used the corrected sign, then the published pseudocode is misleading and the paper should say so.\n\nSecond, the demand model in Algorithm 1 assumes demand depends only on average wind speed at the two sites. That is a strong assumption and the paper gives no justification or sensitivity check. It could change the curtailment distribution materially.\n\nThird, several parameters are missing: the sigmoid coefficients alpha and beta, the line cost CT, and the exact capacity grid are not specified. That hurts reproducibility. Fourth, the equilibrium is not validated against any observed investment or curtailment outcome; the only validation is internal consistency of the sampler.\n\nWho this is for: readers working on network investment under uncertainty and on the use of MCMC in energy systems. It deserves a serious referee, but the sign error and missing parameters should force a major revision. I would want to see corrected pseudocode, a justification or replacement for the demand assumption, and a clear statement of what code produced the numbers.","headline":"A plausible integration of Gibbs sampling with a Stackelberg investment game, undermined by a sign error in the curtailment pseudocode that, if present in the code, inverts the central cost in the profit equations.","tokens_in":10971,"tokens_out":2546,"would_cite":false,"duration_ms":23459,"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":"The paper shows that Gibbs sampling can supply the stochastic scenarios needed to compute the Stackelberg equilibrium of a private network upgrade game, and that the resulting optimal capacities are stable to within about 10 MW on a real…","keywords":["Gibbs sampling","Markov chain Monte Carlo","Stackelberg game","network reinforcement","curtailment","distributed generation","wind power","transmission investment"],"falsifier":"Hold out a period of historical demand not used in building the conditional table, sample demand from $G(P_D \\mid (w_1+w_2)/2)$ for the same period's wind speeds, and compare the sampled distribution with the actual held-out demand. If the two differ substantially in seasonal shape, autocorrelation, or tail behaviour, the equilibrium capacities are not trustworthy. A second check is to rerun the Stackelberg search with an alternative demand model, such as a temperature-driven or autoregressive model, and see whether the equilibrium capacities move outside the reported 10 MW range.","tokens_in":9813,"feed_emoji":"⚡","tokens_out":7172,"duration_ms":67511,"temperature":0.7,"pith_summary":"The paper is trying to establish that a private transmission-line investment, shared by a line investor and local renewable generators under common-access rules, can be solved as a two-level Stackelberg game even though the joint distribution of wind speeds and demand is unknown and has no closed form. Its proposed solution is to learn the conditional distributions from historical records—wind at two nearby sites plus national demand—and to draw synthetic scenarios with a Gibbs sampler, then to search over capacity pairs to find the subgame-perfect equilibrium. The payoff is practical: a regulator or investor can estimate the optimal generation capacities and a feasible range of transmission fees before committing to a costly upgrade. On the Kintyre–Hunterston case in the UK, the equilibrium capacities are stable across 170 Monte Carlo realisations within about a 10 MW range, and sensitivity runs show how the equilibrium shifts with generation costs and the transmission fee.","feed_headline":"Gibbs sampling sizes private grid upgrades from wind and demand data","feed_subtitle":"A Stackelberg game plus Monte Carlo scenarios finds equilibrium capacities stable to within 10 MW on a UK link.","key_machinery":"The load-bearing object is the Gibbs sampler in Algorithm 1, which cycles through the three variables $(w_1, w_2, P_D)$ by sampling each wind speed from the empirical conditional distribution of the other site's wind speed and then sampling demand from the empirical conditional distribution given the average wind speed, with a burn-in of 20% of samples to erase dependence on the starting state. This generates synthetic wind and demand series that are passed to Algorithm 2, which uses a sigmoid power curve to convert wind speed to per-unit power and computes total generation and curtailment for every pair of rated capacities. Algorithm 3 then evaluates the two profit functions for those generation and curtailment totals and finds the Stackelberg equilibrium by backward induction. The machinery works because the Gibbs chain's sample mean converges to the historical mean as the sample size grows, which the paper verifies by increasing $n$ and $N$ and tracking the maximum error from the historical mean.","core_discovery":"The central claim is that Markov chain Monte Carlo, specifically Gibbs sampling, makes the equilibrium computation tractable by replacing an unknown analytic joint distribution with empirical conditional distributions built from data. For each player's wind speed, the sampler draws a new value conditional on the other player's current value, then draws demand conditional on the average of the two wind speeds; after discarding a burn-in period, the resulting chain approximates the historic joint distribution. These synthetic hourly observations feed a full enumeration of capacity pairs, from which expected generation, curtailment, and profits are computed, and backward induction yields the leader's capacity $P^*_{N1}$ and the follower's best response $P^*_{N2}$. The paper reports that these equilibrium capacities show a 10 MW range across realisations and that, as generation costs or the transmission fee rise, total installed capacity falls, with the follower's cost having the larger leverage on both players' capacities.","pith_inferences":["A natural extension, not pursued in the paper, is multi-location sampling: pair each site's wind with a conditional distribution over all other sites, but the chain's mixing and convergence would need separate testing because sparse conditional bins multiply with dimension.","The demand model $P_D \\mid (w_1+w_2)/2$ is the fragile link. Replacing it with a demand model that includes seasonality, temperature, or time-of-day would be a direct stress test; if the equilibrium capacities move outside the reported 10 MW band, the current demand assumption is the cause.","The reported 10 MW stability likely understates total uncertainty because the paper holds cost parameters, the power curve, and the demand conditional fixed across realisations; a full propagation of these uncertainties would widen the credible range.","The comparative-statics result—total capacity falls as costs or fees rise—suggests a regulator can use this pipeline to set $p_T$ before construction, searching for the fee that keeps both players profitable while maximising installed renewable capacity."],"forward_implications":["With a sufficiently large Gibbs sample (here $n = 50{,}000$ after burn-in), the simulated wind speeds and demand reproduce the historical means, so the method can generate large synthetic datasets even when historical records have gaps.","The equilibrium capacities $(P^*_{N1}, P^*_{N2})$ can be recomputed for any set of cost parameters, so the model maps how generation costs and transmission fees shift who builds what.","There is a minimum transmission fee, around £12/MWh or $p_T \\approx 0.16 p_G$, below which the line investor cannot break even, and an upper bound above which local generators stop investing, giving a feasible regulatory range.","Across 170 Monte Carlo realisations of the Kintyre–Hunterston case, equilibrium optimal capacities vary within a 10 MW band, suggesting that scenario-sampling uncertainty does not dominate the investment decision.","Because the game is set up for any two-node system where demand and renewable generation are not co-located, the same pipeline applies to other constrained network upgrade problems with wind and demand data."],"supporting_citations":[{"why":"Supplies the profit equations, the previous one-shot empirical algorithm, and the Kintyre–Hunterston case-study data that this work extends to a Gibbs-sampled multi-scenario setting.","marker":"[11]"},{"why":"Introduces the Stackelberg formulation of curtailment rules and common-access line investment on which the present game model is built.","marker":"[3]"},{"why":"Provides the practical MCMC convergence and burn-in guidance used to set the 20% burn-in and to justify ergodicity checks.","marker":"[15]"},{"why":"Establishes MCMC as a wind-power simulation method, the methodological precedent for using Gibbs sampling on wind data.","marker":"[12]"},{"why":"Demonstrates Markov chain Monte Carlo for wind power time series, supporting the choice of MCMC for scenario generation.","marker":"[13]"},{"why":"Compares interruptible connections with firm reinforcements, framing the private-upgrade decision that the game models.","marker":"[5]"}],"fun_headline_variants":["Game theory plus Gibbs sampling finds stable grid investment equilibrium","MCMC scenarios determine equilibrium capacity for private network upgrades","Stackelberg game with Gibbs sampling sizes renewable line investments","Gibbs draws from wind data to solve grid investment game"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that demand at the consumer node is fully captured by the empirical conditional distribution of historical demand given the average wind speed at the two wind-farm sites; if real demand responds mainly to temperature, season, or time of day rather than wind, the simulated curtailment and the resulting equilibrium capacities and profits will be off.","fun_headline_variants_meta":{"raw":{"variants":["Game theory plus Gibbs sampling finds stable grid investment equilibrium","MCMC scenarios determine equilibrium capacity for private network upgrades","Stackelberg game with Gibbs sampling sizes renewable line investments","Gibbs draws from wind data to solve grid investment game"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00052,"raw_usage":{"total_tokens":2502,"prompt_tokens":914,"completion_tokens":1588,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":530,"completion_tokens_details":{"reasoning_tokens":1523}},"tokens_in":530,"tokens_out":1588,"duration_ms":12161,"temperature":1.0,"reasoning_tokens":1523,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:40:24.768649+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Hold out a period of historical demand not used in building the conditional table, sample demand from $G(P_D \\mid (w_1+w_2)/2)$ for the same period's wind speeds, and compare the sampled distribution with the actual held-out demand. If the two differ substantially in seasonal shape, autocorrelation, or tail behaviour, the equilibrium capacities are not trustworthy. A second check is to rerun the Stackelberg search with an alternative demand model, such as a temperature-driven or autoregressive model, and see whether the equilibrium capacities move outside the reported 10 MW range.","supporting_citations":[{"cited_title":"Game-theoretic modeling of curtailment rules and network investments with distributed generation,","cited_arxiv_id":null,"evidence_quote":"Supplies the profit equations, the previous one-shot empirical algorithm, and the Kintyre–Hunterston case-study data that this work extends to a Gibbs-sampled multi-scenario setting."},{"cited_title":"Game-theoretic modeling of curtailment rules and their effect on transmission line investments,","cited_arxiv_id":null,"evidence_quote":"Introduces the Stackelberg formulation of curtailment rules and common-access line investment on which the present game model is built."},{"cited_title":"Practical Markov chain Monte Carlo,","cited_arxiv_id":null,"evidence_quote":"Provides the practical MCMC convergence and burn-in guidance used to set the 20% burn-in and to justify ergodicity checks."},{"cited_title":"MCMC for wind power simulation,","cited_arxiv_id":null,"evidence_quote":"Establishes MCMC as a wind-power simulation method, the methodological precedent for using Gibbs sampling on wind data."},{"cited_title":"Markov chain Monte Carlo method for the modeling of wind power time series,","cited_arxiv_id":null,"evidence_quote":"Demonstrates Markov chain Monte Carlo for wind power time series, supporting the choice of MCMC for scenario generation."},{"cited_title":"Options for allocating and releasing distribution system capacity: Deciding between interruptible connections and firm DG connections,","cited_arxiv_id":null,"evidence_quote":"Compares interruptible connections with firm reinforcements, framing the private-upgrade decision that the game models."}],"review_version":1}