{"id":"293ee49a-4dcf-4252-bfab-7a3a0a5f335c","arxiv_id":"2501.00219","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"A semi-on-demand autonomous minibus service can reduce generalized costs versus fixed-route buses in low-density suburb-to-downtown corridors when demand is sparse and spread is moderate.","lead":"This paper analyzes an autonomous minibus service that drives on-demand through suburban grids to pick up passengers, then follows a fixed route downtown. It derives formulas and selection indicators for when this hybrid service lowers passenger and operator costs compared with fixed-route buses, and tests them in simulations and Chicago bus cases.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Assumption A2 (same fleet and headway) is physically inconsistent and the omitted fleet/waiting cost could flip the sign of ΔTC in Table 4; the simulation inherits this bias.","rationale":"The reader's weakest_assumption is correct and is the single most load-bearing issue: the analytical comparison and the simulation both maintain headway H while AMSoD cycle times are longer, so the waiting and fleet consequences of the longer cycle are not charged. This is an internal accounting inconsistency, not a disagreement with prior work; the paper even lists the fleet-size issue as Limitation 7, which supports the concern. I do not think the verdict should move because the paper could be repaired by re-running with a fleet constraint and per-bus-hour costs, and low-demand, low-MD corridors might still show savings. A secondary observation is that the empirical support is weaker than the abstract suggests: Table 4 confidence intervals for Model 1, Model 2, and CTA #84 include zero, so a corrected dispatch-constrained test is necessary before claiming reductions. The rule-based simulation shares the derivation's assumptions and therefore does not independently validate the SI<1 threshold.","tokens_in":17308,"tokens_out":12092,"duration_ms":130799,"concrete_test":"Add a dispatch constraint to the Python simulation: fix the AMSoD fleet at ceil(T_cycle/H) buses (and also at the fixed-route fleet size), dispatch a minibus from the terminal only when a bus is available, record realized headways, and add a per-bus-hour fixed cost (e.g., $20/h, sensitivity $0–$40/h). Recompute ΔTC and SI for Model 1, CTA #126, and CTA #84. If the median ΔTC crosses zero or SI reaches 1 in any scenario, Assumption A2 is the decisive bias.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Assumption A2 states that the number of minibuses assigned and the departure headways are the same as those of fixed routes. Under the natural reading, this pair is infeasible whenever the AMSoD cycle is longer: Eq. (19) adds per-passenger pick-up stops, and Eq. (23) adds (K/2)MD(Y)/v of y-directional travel. With headway H held fixed, the required fleet grows from about (Lx/v + t_d NS/2)/H to about (Lx/v + K MD/v + K t_d')/H; with the fleet held fixed, the realized headway grows. Neither consequence appears in the cost model: Eq. (27) charges the extra distance to the operator as γ_d λ MD(Y), but no per-bus-hour or capital cost is included, and Eq. (21) keeps the deterministic waiting term H/2 unchanged. The Monte Carlo simulation dispatches trips at the same H without a return-availability constraint, so it inherits the same bias; the Selection Indicator SI in Eq. (34) is built from Eq. (33) and inherits it as well. The margins are small: in Table 4, three of four median ΔTC values have 95% confidence intervals that include zero (Model 1, Model 2, CTA #84), and the CTA #126 saving of $197 per 3 h could be erased by one extra bus at a modest hourly cost or by a few minutes of headway increase. Limitation 7 acknowledges that more or fewer minibuses may be needed but does not quantify this or re-run the comparisons.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces an Autonomous Minibus Semi-on-Demand (AMSoD) service that combines fixed-route downtown segments with on-demand, dynamically routed suburbs in a grid network. The authors derive closed-form generalized cost expressions for fixed-route and AMSoD services, including access, waiting, riding, and operator costs, and propose a Selection Indicator (SI) to identify existing bus routes suitable for conversion. They extend the formulation to zonal express and parallel-route configurations. The analytical results are supported by Monte Carlo simulation and two Chicago Transit Authority bus route case studies. The paper is clearly structured and provides a transparent cost framework, but the central comparison rests on an assumption of equal headway and fleet size that is not internally consistent.","tokens_in":17673,"tokens_out":9001,"duration_ms":79195,"significance":"The paper addresses an important and timely problem: whether and when autonomous on-demand minibuses can reduce generalized cost in low-density suburban corridors. The cost model is derived from first principles, and the SI indicator offers a practical, parameterized screening tool for transit planners. The authors are transparent about their assumptions and list nine limitations. However, the physical inconsistency of Assumption A2 (same fleet size and headway despite longer cycle time) and the wide confidence intervals in the case studies currently prevent acceptance of the empirical claims. If the authors can remedy the A2 issue, the framework could be a valuable contribution to the flexible transit literature.","major_comments":[{"comment":"Assumption A2 fixes the number of minibuses and departure headway to those of the fixed route. This pair is not jointly feasible when the AMSoD cycle time is longer: Eq. (23) adds y-directional travel (K/2 MD(Y)/v_b) and pick-up dwell time (t_d' K/2) to each trip. With a fixed headway H, the required fleet size grows; with a fixed fleet, the realized headway grows. Neither consequence appears in the cost model. Eq. (27) charges the operator only for distance (γ_d [Lx/H + λ MD(Y)]) and Eq. (21) keeps the deterministic waiting term H/2 unchanged. The Monte Carlo simulation dispatches vehicles at the same H without enforcing a return-availability constraint, so it inherits the same bias; the Selection Indicator in Eq. (34) is constructed from Eq. (33) and inherits it as well. Notably, Limitation 7 acknowledges that 'more or fewer minibuses may be needed' but does not quantify the effect. Because the margins in Table 4 are small (three of the four ΔTC confidence intervals include zero), this omitted cost could plausibly change the sign of the result. The authors should re-run the analytical and simulation comparisons under a consistent fleet-size constraint, either by including the capital/operating cost of the additional vehicles needed to maintain headway or by recomputing the effective headway for a fixed fleet.","section":"Service Description, Assumption A2; Eqs. (21), (23), (26)-(27); Table 4; Limitation 7"},{"comment":"The abstract and conclusions state that simulations and case studies 'show reductions' in generalized costs, but the 95% confidence intervals in Table 4 include zero for Model 1 (-134 to 67), Model 2 (-312 to 43), and CTA #84 (-193 to 38); only CTA #126 (-299 to -67) is significant at the 5% level. The empirical support is therefore weaker than claimed. The authors should either report the proportion of simulated repetitions with negative ΔTC, or qualify the statements by explicitly noting that the case-study differences are not statistically significant for three of the four configurations. The analytical derivations may still support the concept, but the case studies should be presented as illustrative rather than confirmatory.","section":"Table 4; Abstract; Conclusion"}],"minor_comments":[{"comment":"Equation (34), the Selection Indicator, is hard to parse in the typeset manuscript; please display it with a clear fraction bar and explicit multiplication signs, since it is a central result.","section":"Eq. (34)"},{"comment":"Some notation in Table 2 (e.g., the bar over s_w in the waiting-time row) is garbled; please ensure all symbols and accents render correctly in the final version.","section":"Table 2"},{"comment":"In the derivation of the waiting-cost difference leading to Eq. (32), the paper uses Eq. (21) with a headway-variance term but then effectively sets the fixed-route variance to zero. Please clarify that the fixed-route headway variance is assumed to be zero (or that it cancels because both services share the same H and σ_H²).","section":"Expected Waiting Time, Riding Time, and Total Costs"},{"comment":"The Monte Carlo simulation parameters are given, but the code and data are not. Providing a public repository or, at minimum, the synthetic demand generator would improve reproducibility and allow readers to test the fleet-size constraint directly.","section":"Simulation"},{"comment":"Figure 2 is referenced without an accompanying explanation in the text; a sentence describing how the proposed service is positioned relative to fixed-route bus and shared autonomous vehicles would aid readers.","section":"Introduction, Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the journal's scope. The primary concern is the internal consistency of Assumption A2; this is fixable and should be addressed before acceptance. The wide confidence intervals in the case studies also need more careful reporting. No concerns about novelty or attribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a competent, readable modelling paper that gives transit planners a closed-form screening tool for converting fixed suburban routes to semi-on-demand minibus service. The MD(Y) detour model and the SI/SI* indicators are genuinely new, and the extensions to zonal express and parallel routes are useful. But the paper's headline claim, that AMSoD lowers generalized cost, rests on Assumption A2, which is internally infeasible, and the simulation inherits the same bias. The sign of delta TC could flip once fleet size or headway is treated consistently.\n\nWhat the paper does well: the cost decomposition is transparent, parameter tables are complete, and the sensitivity analyses on capacity and demand are honest. The two Chicago cases are a real attempt at external grounding, and the limitations section is unusually candid; Limitation 7 admits the fleet-size question but does not quantify it, which is exactly where the problem sits.\n\nThe soft spot is not obscure. A2 says the number of minibuses and the headways are the same as for fixed routes. But the AMSoD cycle is longer, since Eq. (23) adds y-detour and per-passenger dwell, so the same headway forces more buses, or the same fleet forces a longer headway. The cost model keeps H fixed, charges only distance-based operator cost, and keeps the deterministic H/2 waiting term. The Monte Carlo simulation dispatches at the same H with no return-availability constraint, so it validates the model rather than testing it. The confidence intervals in Table 4 include zero for three of the four comparisons, so the claimed savings are not robust even before adding the omitted fleet cost.\n\nI think the analytical framework is sound enough to be rescued. A corrected version with either a bus-hour cost term or an endogenous headway would be a solid screening tool. As it stands, the quantitative conclusion is conditional on an assumption that cannot hold in practice.\n\nThis paper deserves peer review, not desk rejection, but the referee should demand a re-derivation with consistent fleet/headway and a re-run of the simulations. I would bring it to a reading group as a good example of a transparent continuous-approximation model with a fixable flaw.","headline":"Useful screening indicators for semi-on-demand minibus routes, but Assumption A2 (same fleet and headway) is infeasible and the claimed savings are not robust.","tokens_in":18198,"tokens_out":2184,"would_cite":true,"duration_ms":22587,"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":"Replacing the suburban leg of a fixed bus route with an on-demand minibus lowers generalized cost whenever a Selection Indicator is below 1; simulations and Chicago cases give median hourly savings of $51–$197.","keywords":["demand responsive transit","autonomous bus on-demand","shared autonomous vehicle","minibus","semi-on-demand routes","generalized cost","grid network","bus route conversion"],"falsifier":"A direct test is to recompute the cost comparison without the equal-headway assumption: for one of the modeled corridors, set the minibus departure interval equal to the fixed-route interval plus the extra round-trip time caused by sideways detours and pick-up stops, then re-evaluate the hourly generalized-cost difference. If the difference becomes positive for the Chicago #126 or #84 parameter settings, the paper's central claim fails.","tokens_in":1676,"feed_emoji":"🚐","tokens_out":9405,"duration_ms":179155,"temperature":0.7,"pith_summary":"The paper tries to establish that in low-density suburban corridors feeding a downtown, an autonomous minibus that detours to pick up passengers on demand rather than serving only fixed stops can replace the suburban portion of a fixed bus route at lower total generalized cost. The cost comparison is built from access, waiting, riding, and operator costs on a grid network, and the deciding condition is a single ratio, the Selection Indicator: when $SI<1$ (Eq. 34), the access-time savings outweigh the extra detour, waiting, and operator costs. The authors extend the service to zonal express and parallel routes to widen the scenarios where it wins. If the claim holds, transit agencies would have a simple, data-ready rule for choosing which existing routes to convert, with reported Chicago case studies suggesting median savings on the order of $1–$2.5 per passenger.","feed_headline":"Semi-on-demand minibuses beat fixed buses in low-density suburbs","feed_subtitle":"A simple indicator tells transit agencies when on-demand minibuses beat fixed buses on suburban legs.","key_machinery":"The central object is the mean absolute difference of passenger side-to-side positions, $MD(Y)=\\mathbb{E}|Y_i-Y_j|$, which in Eq. (3) evaluates to $(b-a)/3$ for uniform demand on an interval of width $b-a$ and to $2\\sigma/\\sqrt{\\pi}$ for normal demand. This single number converts the spatial spread of demand into the expected perpendicular distance the minibus must travel to connect consecutive request points, so it controls the extra riding time, operator distance, and waiting-time variance in the cost equations. Around it, the paper builds a standard generalized-cost decomposition (access plus waiting plus riding plus operator cost) and derives the hourly cost difference $\\Delta TC$ (Eq. 33), the Selection Indicator $SI$ (Eq. 34), the parallel-route indicator $SI^*$ (Eq. 37), and the optimal number of zones $n^*$ for zonal express (Eq. 30).","core_discovery":"On the paper's own terms, the discovery is a condition under which the semi-on-demand autonomous minibus service (AMSoD) dominates a fixed-route bus on a directional suburb-to-CBD corridor. Passengers in the suburb are picked up at request points, so access cost drops to zero; the bus pays for this with side-to-side detours, extra dwell time, and the randomness those add to waiting time. The total hourly cost difference $\\Delta TC$ in Eq. (33) is negative, meaning AMSoD wins, exactly when the access-cost saving $\\gamma_a VOT \\lambda \\overline{s_a}$ exceeds the sum of the added waiting, riding, and operator costs. Eq. (34) packages that trade-off as $SI<1$, where $SI$ is the ratio of added costs per unit of access-time saving. Simulations and two Chicago bus-route case studies give median hourly differences of $-51$, $-154$, $-197$, and $-99$ in favor of AMSoD, with $SI$ values $0.80$, $0.97$, $0.75$, and $0.91$.","pith_inferences":["A natural extension the authors do not pursue is to re-derive the Selection Indicator under an endogenous headway: if the fleet is fixed, the extra cycle time from detours should feed back into the departure interval, changing $SI$ and likely shrinking the set of routes where AMSoD wins.","The same machinery could be used as a monitoring rule once AMSoD is running: using realized request locations instead of an assumed uniform distribution, an operator could recompute $MD(Y)$ and $SI$ continuously and switch routes back to fixed service when $SI$ crosses 1.","Because the model assumes demand patterns stay the same, induced demand from shorter access is a blind spot; a testable extension is to couple $SI$ with a ridership-elasticity model and see whether conversion creates enough new riders to push a route past the favorable threshold."],"forward_implications":["Agencies can screen their fixed-route network by computing the Selection Indicator for each corridor; routes with $SI<1$ are candidates for AMSoD conversion without running a full simulation.","The service is a low-density, off-peak tool: the paper derives an upper bound on the demand-headway product $\\lambda H$ in Eq. (35), so fixed routes retain the advantage once that product exceeds the bound.","For corridors with wide sideways demand spread, splitting service into two parallel AMSoD routes restores the advantage, with its own condition $SI^*<1$ in Eq. (37).","For busier corridors, zonal express—assigning each minibus to one of $n$ zones and letting it skip other zones—can be optimal; the optimal zone count $n^*$ in Eq. (30) grows with operator cost and the highway-speed advantage.","The reported Chicago cases put median savings at about $0.9–$2.5 per passenger, but the 95% confidence intervals include positive differences in several scenarios, so the benefit is not uniform across demand draws."],"supporting_citations":[{"why":"Supplies the generalized-cost decomposition (access, waiting, riding, operator) that the paper's cost equations are built on.","marker":"(26)"},{"why":"Provides the grid-network flexible-transit modeling approach and the analytical cost-function style adapted for semi-on-demand routes.","marker":"(5)"},{"why":"Provides automated demand-responsive transit cost and capacity parameters, plus the positioning of minibus services against shared autonomous vehicles.","marker":"(13)"},{"why":"Underlies the penalty multipliers for waiting and access time relative to riding time used in the generalized-cost model.","marker":"(27)"},{"why":"Gives the value-of-travel-time savings used to convert access-time savings into dollars in the simulations and case studies.","marker":"(29)"},{"why":"Supplies the bus-stop boarding data that define demand patterns for the two Chicago case studies.","marker":"(31)"}],"fun_headline_variants":["When on-demand minibuses beat fixed-route buses","Semi-on-demand minibuses slash costs in suburbs","Grid network minibuses: on-demand routing wins","Chicago case: on-demand minibuses cut bus costs","New metric picks routes for on-demand minibuses"],"cache_read_input_tokens":20224,"weakest_assumption_plain":"The load-bearing premise is the assumption that the on-demand minibuses keep the same number of vehicles and the same time gap between departures as the fixed route, even though their detours and extra pick-up stops make each round trip longer; if the gap has to widen or extra vehicles have to be added, the extra waiting or fleet cost could erase the access-time savings.","fun_headline_variants_meta":{"raw":{"variants":["When on-demand minibuses beat fixed-route buses","Semi-on-demand minibuses slash costs in suburbs","Grid network minibuses: on-demand routing wins","Chicago case: on-demand minibuses cut bus costs","New metric picks routes for on-demand minibuses"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000149,"raw_usage":{"total_tokens":1144,"prompt_tokens":850,"completion_tokens":294,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":466,"completion_tokens_details":{"reasoning_tokens":215}},"tokens_in":466,"tokens_out":294,"duration_ms":3191,"temperature":1.0,"reasoning_tokens":215,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:57:22.434580+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct test is to recompute the cost comparison without the equal-headway assumption: for one of the modeled corridors, set the minibus departure interval equal to the fixed-route interval plus the extra round-trip time caused by sideways detours and pick-up stops, then re-evaluate the hourly generalized-cost difference. If the difference becomes positive for the Chicago #126 or #84 parameter settings, the paper's central claim fails.","supporting_citations":[],"review_version":1}