{"id":"0e0b5a03-0387-46d9-90b6-63943b3a733f","arxiv_id":"2509.02052","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Multi-period line planning with asymmetric lines and transfers reduces modelled passenger GJT by up to 4.26% in a Dutch railway network case study.","lead":"This paper designs a mathematical model for railway line plans that can change during the day, including lines that stop differently in each direction. In a Dutch case study, the model cuts passengers' total travel time by up to 4.26% compared with a fixed-style baseline.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Asymmetric-line GJT gains depend on an unmodeled passenger cost for one-direction-only stations; Section 6 concedes this, so the 4.26% headline is not yet a robust real-world benefit.","rationale":"Reader's verdict is CONDITIONAL and I agree that the model is clearly specified and internally coherent. The strongest quantitative claim is the 4.26% reduction, and the most load-bearing condition for it is not solver quality—feasible incumbents already establish existence—but whether GJT as defined captures passenger disutility of asymmetric service. The authors are explicit about this limitation, so the correct verdict is unchanged: the paper should be read as a modeling contribution with a promising but not yet robust case-study quantification. I would not move to reject because the limitation is stated, the model is reproducible in principle, and the sensitivity check is straightforward.","tokens_in":26700,"tokens_out":14081,"duration_ms":171702,"concrete_test":"Re-run the MP-LPP case study with an added penalty term: for every passenger whose chosen path uses an origin, destination, or transfer station that is served in only one direction in that period, add α generalized minutes (e.g., α ∈ {2,5,10}) to their journey cost, or equivalently add α to the In/In-change/Out/Out-change arc costs at such stations. Compare the asymmetric-vs-symmetric GJT differences in Table 7 (same adjustment counts: 0,10,20,30,40). If the asymmetric advantage falls below ~1 percentage point or reverses for α ≥ 5 minutes, the headline benefit is not robust to the conceded behavioral limitation. This is a computational sensitivity check that directly tests the Section 6 caveat without needing new survey data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative claim is that allowing multi-period line plans with asymmetric lines reduces passenger generalized journey time by up to 4.26% (Abstract, Table 7). Internally, the comparison is between feasible solutions under the model's stated GJT definition, so the existence claim is not invalidated by the large optimality gaps (15.2% for the 70-adjustment asymmetric plan vs 4.6% for the 40-adjustment symmetric plan). The load-bearing threat is external: the model's GJT counts in-vehicle time, waiting time, and fixed boarding/transfer penalties only, and assigns zero cost to the defining feature of asymmetric plans—stations served in only one direction. Section 5.3.2 shows that in the asymmetric reference plan stations Gvm, Vst, and Dvnk are served in one direction only; Section 6 admits this 'might make the asymmetric line plans look better than they really are.' For a passenger whose desired direction is unserved, the modeled rerouting cost is just extra in-vehicle/wait time; the psychological or practical penalty of traveling away from one's destination first is absent. Since the asymmetric advantage is built on one-directional skip-stopping, a modest per-passenger penalty on affected trips could erase or reverse the 4.26% gap. This does not impugn the model's internal logic, but it makes the real-world claim conditional.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a mixed-integer linear programming model for multi-period railway line planning. The model chooses routes, stopping patterns, frequencies, transfers, and—unlike most existing work—allows asymmetric lines with direction-dependent stop patterns and frequencies. Passenger routing is modelled on a change-and-go network with origin-grouped flows, and the ε-constraint method is used to generate Pareto-optimal line plans with different numbers of daily adjustments. The approach is tested on a three-period case study of the Dutch network around Leiden, The Hague, Rotterdam, and Utrecht. The central quantitative claim is that allowing line-plan changes during the day can reduce total generalised journey time (GJT) by up to 4.26% relative to a symmetric 20-adjustment reference plan, with the largest gains attributed to asymmetric lines.","tokens_in":27175,"tokens_out":11553,"duration_ms":122934,"significance":"If the empirical claim were robust, the paper would make a useful contribution to network-level, time-dependent line planning. The change-and-go formulation with origin-grouped flows is technically solid and the explicit control of the number of line-plan adjustments is a pragmatic way to model regularity concerns. The paper also fills a clear gap in Table 1: no prior network-level railway model combines route selection, stop-pattern selection, frequencies, transfers, and asymmetric lines in a multi-period setting. However, two issues prevent me from endorsing the headline result as it stands: the 4.26% figure is a comparison of heuristic incumbents with large optimality gaps, and the asymmetric-line benefit depends on a passenger cost that is acknowledged in Section 6 but not modelled or quantified. These are load-bearing for the paper's practical claims, including the revenue-impact estimate in Section 5.1.","major_comments":[{"comment":"The headline 'up to 4.26%' reduction is the difference between two feasible incumbents: the symmetric 20-adjustment reference (GJT 5,632,353) and the asymmetric 70-adjustment solution (GJT 5,392,625). The reported optimality gaps are 6.7% for the reference and 15.2% for the asymmetric solution. Because the symmetric model's true optimum could be substantially lower than the incumbent, the comparison does not establish that 4.26% is the maximum achievable improvement, or even that asymmetric plans are better than the best symmetric plan at the same adjustment count. I recommend rephrasing the claim as 'we found feasible plans with GJT reductions of up to 4.26% under the model's GJT definition', and using the period-wise lower bounds to state what can and cannot be concluded about the true optimum. Ideally, solve at least the reference and the 40-adjustment symmetric cases to much smaller","section":"§5.1, Table 7"},{"comment":"The asymmetric plans achieve part of their advantage by serving stations such as Gvm, Vst, and Dvnk in only one direction. Section 6 concedes that 'How passengers feel about this is not taken into account in this study, which might make the asymmetric line plans look better than they really are.' This is not merely a cosmetic limitation: the same section uses the 4.26% GJT reduction to estimate a 3.45% revenue increase and €99.3 million for NS. A modest per-trip penalty for passengers whose desired direction is unserved could erase or reverse the asymmetric advantage. I ask for a sensitivity analysis: add a fixed GJT penalty (e.g., 0, 5, 10, 15 minutes) to affected trips and report at what penalty the 4.26% result disappears. This would turn a conceded limitation into a quantified boundary.","section":"§6 and §5.3.2"},{"comment":"There is an inconsistency in the definition and use of the frequency-change variable. Table 3 defines ef^{l,p} without a frequency index, but constraints (14) and (15) are written for each i ∈ F^l using the same ef^{l,p}, and constraint (16) sums ef^{l,p} over i. As written, each frequency change contributes |F^l| to the adjustment count rather than 1, which would make the 'number of adjustments' in Table 7 inconsistent with the ε bound. If ef is intended to be indexed by i, Table 3 and (21) should be corrected. If not, the sum over i in (16) should be replaced by a single term. This issue is load-bearing because the entire Pareto analysis is framed in terms of adjustment counts.","section":"§3.3.1, Table 3, Eqs. (14)-(16), (21)"}],"minor_comments":[{"comment":"The domain of the stopping variable x^{l,p}_s incorrectly includes 'i ∈ F^l'; x does not depend on the frequency index. Remove it.","section":"Eq. (17)"},{"comment":"The domain of ef^{l,pj} includes 'i ∈ F^l' although ef has no frequency index in Table 3. This should be aligned with the corrected definition from the major comment above.","section":"Eq. (21)"},{"comment":"The column 'Optimality gap' mixes Gurobi gaps (starred) with gaps computed from period-wise lower bounds. Please state this distinction in the table caption or in a footnote, and clarify which lower bound is used for each non-starred entry.","section":"Table 7"},{"comment":"The revenue-impact estimate of €99.3 million is presented without the caveats attached later in Section 6. It should be labelled as an illustrative upper bound that assumes the modelled GJT fully captures passenger welfare and that the incumbent gap does not affect the comparison.","section":"§5.1"},{"comment":"Minor typographical and style issues: 'AUGMECON2' is used inconsistently with spacing; the phrase 'approximate the Pareto optimal solutions' should acknowledge that the ε-constraint runs are terminated by a time limit, so the obtained frontiers are heuristic approximations.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a mathematical optimization / transport methods journal, and I see no evidence of misconduct or inappropriate citation practices. The model formulation is a genuine contribution, and the authors are transparent about the main limitation. My major-revision recommendation is driven by (a) the need to correct the adjustment-count inconsistency, and (b) the need to either strengthen the empirical claims with tighter bounds/penalty sensitivity or substantially temper the headline and revenue statements. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"It's a real modeling contribution, but the headline number is softer than the abstract suggests. The model is the first, as far as Table 1 shows, to combine route selection, stop-pattern choice, frequencies, transfers, and asymmetric lines in a multi-period network line planning problem. The change-and-go network extension to handle one-directional stopping patterns is a clean piece of work, and the ε-constraint Pareto frontier is a useful practical output. Credit also goes to the authors for flagging their own blind spots, especially around one-directional stations.\n\nThe soft spots are three.\n\nFirst, the 4.26% GJT reduction (Table 7) is a comparison between two heuristic incumbent solutions, not proven optima. Gurobi gaps range from 4.6% to 16.5% for symmetric scenarios and 15.2% to 26.5% for asymmetric ones. That doesn't invalidate the existence claim—both are feasible plans with a common objective—but it means the number is not a precise estimate, and the gap structure makes the asymmetric advantage less certain.\n\nSecond, the baseline is internally generated: a symmetric plan with 20 adjustments, chosen so each period uses at least 98.5% of budget. It's a defensible choice, but it is not the actually operated 2023 line plan, so the 4.26% is relative to a counterfactual, not to the status quo.\n\nThird, and most load-bearing, the model assigns zero cost to the defining feature of asymmetric plans. Stations Gvm, Vst, and Dvnk are served in only one direction in the reference asymmetric plan (Section 5.3.2), and the authors concede in Section 6 that this might make asymmetric plans 'look better than they really are.' Passengers heading the other way must first ride away from their destination; the model only counts the extra in-vehicle and waiting time, not the psychological or practical penalty. A modest per-trip surcharge on those detours could erode or erase the 4.26% difference. That is a missing ingredient in the GJT definition, not a math error.\n\nThe demand data are proprietary NS smart-card data and the GJT arc costs come from the authors' own stated-preference studies. That is not circular—those are externally collected inputs, not fitted to this result—but it limits independent replication.\n\nWho gets value: people working on line planning with time-dependent demand, and practitioners thinking about asymmetric service patterns. The modeling is serious and the limitations are honestly stated. I would send it to peer review, but I'd push for two things: better bounds on optimality gaps, at least on smaller instances, and a sensitivity test where one-directional service carries a nonzero passenger cost. If the 4.26% survives that test, it's a useful, citable result. If not, the modeling contribution still stands, but the headline should be revised.","headline":"A genuine modeling contribution—first to combine route selection, stop patterns, frequencies, transfers, and asymmetric lines in a network—but the headline 4.26% benefit is an incumbent-vs-incumbent comparison under large optimality gaps and an unmodeled one-directional service cost; the modeling deserves peer review, the headline needs tempering.","tokens_in":27611,"tokens_out":5507,"would_cite":false,"duration_ms":54015,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C11","90B06","90C29"],"pacs":[],"model":"deepseek-v4-flash","headline":"Allowing line plans to change during the day can cut total passenger journey time by up to 4.26%, with asymmetric lines doing the heavy lifting.","keywords":["multi-period line planning","asymmetric lines","generalised journey time","change-and-go network","time-dependent demand","railway networks","epsilon-constraint method","stop planning"],"falsifier":"Re-solve the Dutch case study with a calibrated 'wrong-direction penalty' added whenever a passenger must ride away from their destination to board a train, using values from a stated-preference survey of travellers at one-direction-only stations. If the 4.26% gain survives realistic penalty values, the result is robust; if it collapses, the headline improvement is an artifact of system-optimal assignment.","tokens_in":26643,"feed_emoji":"🚆","tokens_out":6279,"duration_ms":62229,"temperature":0.7,"pith_summary":"Most railway networks run one line plan all day even though demand shifts in volume and direction between morning peak, midday, and afternoon peak. This paper argues the fixed plan is wasteful, and builds a mixed-integer optimization model that chooses routes, stopping patterns, and frequencies separately for each period — including lines that stop or run more often in one direction than the other. Tested on a real Dutch network, the model finds that letting the line plan adapt through the day cuts total generalised journey time by up to 4.26% compared with a near-fixed reference plan, and that asymmetric lines drive much of the gain. The authors translate the saving into a revenue increase of roughly 3.45% using standard demand elasticities, which is why the result would matter to any railway operator deciding whether daily plan changes are worth the operational hassle.","feed_headline":"Daily line-plan switches cut rail journey time up to 4.26%","feed_subtitle":"A model that adapts stops and frequencies to peak and off-peak demand beats the fixed all-day timetable.","key_machinery":"The change-and-go network, which represents each line by departure and arrival nodes and each station by In, Change, and Out nodes so that passenger paths, transfers, and waits become linear flow constraints; frequency-indexed boarding arcs whose costs come from a stated-preference entry-resistance curve, which keeps the model linear; a symmetry parameter σ that doubles capacities, costs, and change counts when symmetric (two-direction) lines are enforced; and ε-constraint search (AUGMECON2) that repeatedly solves the model with different caps on line-plan changes to trace the Pareto frontier between journey time and timetable stability.","core_discovery":"The paper claims that optimizing a railway line plan separately for each demand period — instead of operating one fixed plan all day — measurably improves passenger service under the same operating budget. The model minimises total generalised journey time, defined as in-vehicle time plus frequency-dependent waiting and transfer penalties, across three periods: morning hyper peak, midday off-peak, and afternoon hyper peak, while capping how much the plan may change between periods. On a real Dutch network, symmetric lines with 40 allowed adjustments cut GJT by 1.94% versus the 20-adjustment reference plan, while asymmetric lines with 70 adjustments cut it by 4.26%. Asymmetric plans serve low","pith_inferences":["The 4.26% gain is computed under system-optimal passenger routing; a penalty for the psychological cost of one-direction-only stations, which the authors flag as missing, would likely shrink it.","The pattern that frequency changes dominate early savings suggests a design heuristic: spend the first budget of daily adjustments on frequencies and use stop-pattern changes only when frequency headroom is exhausted.","Including rolling-stock circulation and transition logistics between period plans — both outside this model — would raise the real cost of adjustments and may move the practical optimum to fewer changes."],"forward_implications":["Operators get a quantified trade-off frontier: the first adjustments, mostly to frequencies, yield the largest journey-time savings, while later stopping-pattern changes give diminishing returns.","At equal budget and equal number of daily adjustments, asymmetric lines beat symmetric lines by 0.2% to 2.5% in the case study.","Skipping low-demand stops in one direction only keeps those stations served while speeding the dominant passenger flow, so directional service can replace full service at lower cost.","A 4.26% GJT reduction translates to roughly 3.45% revenue growth under the cited elasticity, about €99 million at the operator's 2023 revenue level."],"supporting_citations":[{"why":"Supplies the origin-grouped passenger-flow variables and frequency-indexed boarding arcs that keep the model linear.","marker":"Bull et al. (2019)"},{"why":"Introduces the change-and-go network on which all passenger routing, transfers, and waiting costs are built.","marker":"Schöbel and Scholl (2005)"},{"why":"Provides the AUGMECON2 ε-constraint algorithm used to generate the Pareto-optimal line plans.","marker":"Mavrotas and Florios (2013)"},{"why":"Clusters real smart-card demand into the three homogeneous periods and supplies the case-study OD data.","marker":"Van der Knaap et al. (2024)"},{"why":"Stated-preference entry-resistance curve that sets the frequency-dependent boarding wait costs.","marker":"Guis et al. (2023)"},{"why":"Transfer-resistance function that sets the costs of changing lines.","marker":"de Bruyn et al. (2023)"},{"why":"Meta-analytic generalized-journey-time elasticity used to convert the time saving into a revenue gain.","marker":"Wardman (2012)"},{"why":"Shows flexible stopping rules outperform fixed ones under diverse demand, motivating stop-pattern optimisation.","marker":"Zhang et al. (2020)"},{"why":"Bounds dissimilarity between period line plans; the paper adapts the idea but counts stop and frequency changes instead of shared edges.","marker":"Schiewe et al. (2023)"}],"fun_headline_variants":["Asymmetric line plans trim rail travel time by 4.26%","Switch rail lines per period to cut journey time 4.26%","Adaptive train plans shave 4.26% off travel time","Daily line changes beat fixed timetable by 4.26%","Scheduling per demand period cuts rail commute 4.26%"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that passenger experience is fully captured by rerouting time: serving a station in only one direction is assumed to cost passengers nothing beyond the extra travel time, and the authors concede this may make asymmetric plans look better than they are.","fun_headline_variants_meta":{"raw":{"variants":["Asymmetric line plans trim rail travel time by 4.26%","Switch rail lines per period to cut journey time 4.26%","Adaptive train plans shave 4.26% off travel time","Daily line changes beat fixed timetable by 4.26%","Scheduling per demand period cuts rail commute 4.26%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001041,"raw_usage":{"total_tokens":4181,"prompt_tokens":673,"completion_tokens":3508,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":417,"completion_tokens_details":{"reasoning_tokens":3415}},"tokens_in":417,"tokens_out":3508,"duration_ms":25426,"temperature":1.0,"reasoning_tokens":3415,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T11:55:03.663632+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-solve the Dutch case study with a calibrated 'wrong-direction penalty' added whenever a passenger must ride away from their destination to board a train, using values from a stated-preference survey of travellers at one-direction-only stations. If the 4.26% gain survives realistic penalty values, the result is robust; if it collapses, the headline improvement is an artifact of system-optimal assignment.","supporting_citations":[{"cited_title":", author Larsen, J","cited_arxiv_id":null,"evidence_quote":"Supplies the origin-grouped passenger-flow variables and frequency-indexed boarding arcs that keep the model linear."},{"cited_title":", author Florios, K","cited_arxiv_id":null,"evidence_quote":"Provides the AUGMECON2 ε-constraint algorithm used to generate the Pareto-optimal line plans."},{"cited_title":", author De Bruyn, M","cited_arxiv_id":null,"evidence_quote":"Clusters real smart-card demand into the three homogeneous periods and supplies the case-study OD data."},{"cited_title":", author de Bruyn , M","cited_arxiv_id":null,"evidence_quote":"Stated-preference entry-resistance curve that sets the frequency-dependent boarding wait costs."},{"cited_title":", author Vaatstra, I","cited_arxiv_id":null,"evidence_quote":"Transfer-resistance function that sets the costs of changing lines."},{"cited_title":", year 2012","cited_arxiv_id":null,"evidence_quote":"Meta-analytic generalized-journey-time elasticity used to convert the time saving into a revenue gain."},{"cited_title":", author Nie, L","cited_arxiv_id":null,"evidence_quote":"Shows flexible stopping rules outperform fixed ones under diverse demand, motivating stop-pattern optimisation."},{"cited_title":", author Sch \\\"o bel, A","cited_arxiv_id":null,"evidence_quote":"Bounds dissimilarity between period line plans; the paper adapts the idea but counts stop and frequency changes instead of shared edges."}],"review_version":1}