{"id":"d7a9b80e-9881-476e-972b-4e2e3d0dc05a","arxiv_id":"2501.13508","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Delays in IoT guidance messages, modeled as a per-node probability, monotonically increase simulated cruise-ship evacuation times, with a 50% penalty when every message is one step late.","lead":"A simulation of a cruise ship evacuation shows that when the guidance messages from the ship's internet-connected systems arrive late, passengers take longer to reach exits, up to 50% longer when all messages are one step behind. This matters because emergency planners usually design evacuation routes without counting the delays that communication networks introduce under stress.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Missing definition of the simulation step duration makes the headline 50% delay penalty unanchored to wall-clock IoT/ICT delay; the number may be an artifact of an arbitrary one-step lag.","rationale":"Reading in good faith, the paper's intended contribution is to show that delays in delivering ANT guidance instructions can impair cruise-ship evacuation, with a quantitative 50% penalty at full one-step staleness. For that claim to hold in any physically meaningful way, the abstract notion of 'one step' must correspond to a well-defined real-world delay, because the simulated outcome is measured in seconds. The reader's weakest_assumption correctly identifies the stylized nature of the one-step, node-independent, memoryless delay model; my concern sharpens this: even within the paper's own framing, no parameter connects the discrete informational lag to the continuous time axis, so the headline magnitude is underdetermined. This is not an external disagreement with the field but an internal underspecification: the model defines IL in terms of node arrival events but never states the event's duration or the direction-refresh interval. The qualitative direction of the result is plausible and supported by the reported curves, so rejection would be too strong; however, the quantitative headline should be interpreted conditionally pending a defined step-to-time mapping. Since the reader's verdict is already CONDITIONAL and my concern reinforces that condition rather than overturning it, the verdict should remain unchanged. No code or data are shipped, which makes the proposed rerun the natural way to establish whether the 50% number survives a defined step duration.","tokens_in":8838,"tokens_out":3514,"duration_ms":34815,"concrete_test":"Pin down the simulation step: specify the direction-update period Delta-t (in seconds) that corresponds to IL=1, or equivalently identify the event that separates 'current arrival' from 'just before arrival' and measure its mean duration. Then rerun the Section III-A experiment at PoD=1 with Delta-t = 1, 5, and 10 seconds (and at PoD=0 as baseline) using the same topology, passenger distribution, and ANT implementation. If the 50% performance ratio changes materially with Delta-t, the headline claim is not robust and must be restated as a function of actual delay; if the ratio stays near 1.5 for all Delta-t, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative result, Section III-A's claim that PoD=1 raises average evacuation time by 50%, depends on 'information lag' being exactly one discrete step, but the paper never defines the duration of a step or maps it to any real IoT/ICT delay such as an update interval, sensor sampling period, or packet transmission time. Evacuation times are reported in seconds, yet the one-step delay has no seconds attached. Consequently, the 50% figure could be made larger or smaller simply by redefining the step duration, and the result cannot be compared with any concrete communication delay budget. This is a missing model parameter rather than a claim contradicted by consensus, and it is load-bearing because the abstract and conclusions use exactly this penalty as evidence that IoT/ICT delays significantly hinder evacuation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper simulates passenger evacuation on a graph model of the Yangtze Gold 7 cruise ship using the ANT navigation algorithm embedded in AnyLogic. It introduces a per-node 'probability of delay' (PoD) that determines whether each node's guidance is based on current evacuee locations (IL=0) or on locations one simulation step older (IL=1). The central quantitative result, reported in Section III-A, is that the average evacuation time from all 346 nodes increases with PoD, and at PoD=1 the average evacuation time is 50% higher than at PoD=0. The paper also reports separate results for cabin passengers, restaurant passengers, and different total passenger counts, all showing that larger PoD increases evacuation time, and concludes that IoT/ICT delays can significantly hinder ship evacuation.","tokens_in":9006,"tokens_out":4922,"duration_ms":45771,"significance":"If the central quantitative claim is properly anchored, the paper makes a useful contribution: it applies an existing deadline-aware navigation algorithm to a real cruise-ship layout, runs 100 independent simulations per setting, reports standard deviations and confidence intervals, and does not fit any parameters to produce the main result. The direction of the effect—stale guidance increases evacuation time—is intuitive and robust, and the paper is one of the few attempts to quantify IoT/ICT-induced information staleness in maritime evacuation. The main limitation is that the 'one-step' information lag is never mapped to a wall-clock duration, so the headline 50% penalty cannot currently be compared with concrete communication delay budgets or used as a system design target.","major_comments":[{"comment":"The definitions of IL and PoD in Section III do not state the wall-clock duration of the 'one step' that constitutes an information lag. Evacuation times are reported in seconds, but the lag is only defined as 'one step' in the simulation. Thus the headline result in Section III-A that PoD=1 raises average evacuation time by 50% is not anchored to any concrete IoT/ICT delay such as a sensor sampling interval, server computation time, or packet delay. The ratio may change if the step duration is redefined, and no comparison with a communication delay budget is possible. Please define the step duration in seconds, justify it from the system parameters in Section II-B, or report results for a range of step durations.","section":"Section III, Section III-A"},{"comment":"Section II-C states that 'the initial locations of the evacuees are randomized in each round,' whereas Section III says the 100 simulations are 'under the same initial conditions.' These statements are contradictory. If the first is true, the averages and confidence intervals in Figures 2-5 include variability from initial placement; if the second is true, the randomness is only in the PoD realizations. Please clarify which experimental design was used and report the statistics accordingly.","section":"Section II-C and Section III"},{"comment":"The model represents all IoT/ICT imperfections by a single per-node independent Bernoulli event that makes the advice one step stale. It does not include packet loss, correlated congestion, variable delay durations, or partial message content, although Section I mentions 'lost or delayed messages' and Section II-B mentions 'packet losses.' The resulting 50% figure is therefore a property of a stylized staleness model, not a system-level estimate. Please add a sensitivity/robustness discussion (for example, varying the step duration, the delay distribution, and the traversal-speed assumptions) and explicitly state which impairments are outside the model.","section":"Section III and Section II-B"}],"minor_comments":[{"comment":"There is a grammatical error: 'This paper presents explores the impact' should be 'This paper explores the impact.'","section":"Abstract"},{"comment":"The text contains 'trANTportation' in the paragraph on maritime transportation; this appears to be an unintended typo.","section":"Section I"},{"comment":"The simulator is referred to as 'AnLogic' in one place and 'AnyLogic' in others; please make the spelling consistent.","section":"Section II-A"},{"comment":"The paper would benefit from a small table with the numerical values behind the 50% claim, including means and confidence intervals for PoD=0 and PoD=1, since the figure alone makes verification difficult.","section":"Section III-A"},{"comment":"The description that the curves show 'a quasi-linear increase ... and a more gradual increase' is not supported by any quantitative fit; please provide the slope or another summary statistic if this characterization is meant to be precise.","section":"Section III"},{"comment":"The caption ends with 'as compared to the ideal case w P oD= 0'; the word 'with' is missing, and the formatting of P oD should be normalized throughout the paper.","section":"Figure 5 caption"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope as an IoT/ICT performance study for maritime safety. The novelty is incremental but acceptable if the authors address the step-duration anchoring and the contradictory description of the simulation design. The missing step-duration mapping is the main substantive barrier: it prevents the headline 50% figure from being compared with real communication delay budgets. I do not see grounds for rejection, provided the requested clarifications and sensitivity analysis are added."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper runs an established evacuation simulator (ANT + AnyLogic) on a real cruise ship layout and perturbs it with a per-node probability that the guidance each evacuee sees is one step out of date. The result is a clean monotone curve: as the probability of delay rises, average evacuation time rises, reaching about 50% above the ideal case at full staleness. That is a useful qualitative message for anyone designing IoT/ICT for ship evacuation.\n\nWhat's genuinely new is the specific quantification on this ship layout: the 50% penalty at PoD=1 and the sensitivity across cabin and restaurant origins are not in the cited prior work. The simulations are run 100 times per setting with confidence intervals, and the findings are internally consistent.\n\nThe main soft spot is the one that also drives the headline number: the 'information lag' is exactly one simulation step, but the paper never says how long a step is in seconds. Evacuation times are reported in seconds, yet the delay imposed on the advice has no wall-clock duration. Without that mapping, the 50% figure is an artifact of the chosen step granularity; redefining the step would change the magnitude. The qualitative direction is robust, but the quantitative claim is unanchored. The authors should either define the step duration relative to realistic update intervals, sensor sampling, or packet delays, or present the result as a function of step duration.\n\nA second issue is that the delay model is extremely stylized: one-step, memoryless, node-independent lag, with no message loss or variable congestion. That is a defensible first cut, but it should be labeled as such, not taken as a real network model. Also, the paper's claim that this is the 'first examination' of ICT/IoT performance in ship evacuation is overreach; there is related work on communication delays in emergency navigation. And no code or data are shipped, which limits reproducibility.\n\nNone of these break the central message. The paper is worth a serious referee: it addresses a real gap, uses a realistic layout, and produces a plausible sensitivity curve. The referee should push for a defined step duration, a softer novelty claim, and a clearer statement of the model's limitations.","headline":"A plausible sensitivity study showing stale guidance hurts evacuation, but the headline 50% penalty is not tied to any real-world delay duration.","tokens_in":9500,"tokens_out":2217,"would_cite":false,"duration_ms":19640,"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":"Delayed IoT guidance can raise cruise-ship evacuation time by 50 percent.","keywords":["maritime evacuation","Internet of Things","information lag","probability of delay","cruise ship simulation","adaptive navigation strategy","communication delay","pedestrian evacuation"],"falsifier":"Re-run the same ANT/AnyLogic simulation with the one-step per-node delay replaced by delays drawn from a measured congested shipboard network (or a queueing model with the same mean delay), and compare the full-delay average evacuation time to the PoD=0 baseline; reproducing the 50% penalty is required for the claim, and a materially different ratio would show the one-step model is the source of the number.","tokens_in":8639,"feed_emoji":"🚢","tokens_out":5871,"duration_ms":52557,"temperature":0.7,"pith_summary":"This paper tries to establish that the time it takes to evacuate a cruise ship depends not only on the layout and passenger behavior, but also on the speed and reliability of the IoT and ICT system that sends navigation instructions. It simulates a real ship, the Yangtze Gold 7, using the deadline-aware Adaptive Navigation Strategy (ANT) routing algorithm, and introduces a per-node probability that the instruction shown to evacuees is one navigation step out of date. The central quantitative result is that when every node delivers one-step-old guidance, the average evacuation time over all starting points rises by about 50 percent relative to the ideal case with no lag. Because the available evacuation time on passenger ships is tight, the paper argues, network congestion inside the ship is a safety issue, not just a performance nuisance.","feed_headline":"Stale IoT directions add 50% to ship evacuation time","feed_subtitle":"Simulating a real cruise ship, one-step communication delays raise average evacuation time by half.","key_machinery":"The load-bearing machinery is the pair of parameters 'information lag' (IL) and 'probability of delay' (PoD). IL=1 means a node's displayed directions are computed from evacuee positions one navigation step in the past, an idealization of any communication or computation delay in the guidance loop; PoD is the per-node probability that the node is in that stale state. The rest of the framework—the ANT path planner, the AnyLogic pedestrian simulator, and the Yangtze Gold 7 graph with 346 nodes, 600 passageway segments, and 5 staircases—is the vehicle that turns this parameter into a quantitative reading of evacuation time.","core_discovery":"The paper's central claim is that a one-step information lag in the delivery of evacuee guidance is enough to measurably worsen evacuation, and that the degradation grows with the probability of delay. In the simulator, each node is marked as having information lag IL=1 with a probability PoD, meaning the directions it displays are computed from evacuee positions one time step earlier; otherwise IL=0, meaning directions are based on current positions. Averaging over 100 independent runs and all 346 nodes, the authors find a quasi-linear increase in mean evacuation time as PoD rises to 0.5, then a more gradual increase, reaching a 50% penalty at PoD=1. Passengers starting in cabins see a stronger effect, passengers in the restaurant a milder one, and the penalty persists whatever the number of evacuees.","pith_inferences":["If the one-step, independent per-node delay model is replaced by correlated congestion—for example, a failed switch that freezes many signs at once—the average penalty could be larger than 50% because stale answers would cluster instead of being isolated.","The nearly linear region at low PoD suggests the largest return on investment comes from improving already-fast parts of the network before congestion becomes widespread, a prioritization the paper does not explicitly draw.","The same 50% benchmark could be used as a rough planning number for any centrally guided pedestrian evacuation in a building with a comparable guidance loop, though its exact value would need recalibration."],"forward_implications":["If the one-step lag is representative, then evacuation-time computations that assume instant instructions are systematically optimistic, and the 50% penalty is a worst-case bound of that optimism.","The result gives network designers a concrete target: cutting the probability of delay has a large effect in the low-to-mid range, so even partial network improvements yield meaningful evacuation-time reductions.","Because the penalty is largest for cabin-originating passengers, the communication system should be provisioned to keep the most remote or interior compartments from falling into a stale state.","The persistence of the penalty across all passenger counts means congestion in the corridor and congestion in the network both matter, so the two must be co-designed."],"supporting_citations":[{"why":"Sets the evacuation time budget (TS=60 min, TA=5 min, TEL=25 min, TD=30 min) that turns the 50% delay penalty into a safety-relevant result.","marker":"[19]"},{"why":"Provides the AnyLogic pedestrian simulator with social-force movement dynamics in which the evacuation is run.","marker":"[23]"},{"why":"Supplies the deadline-aware ANT navigation algorithm and the ship graph that the information-lag study perturbs.","marker":"[30]"},{"why":"ANT is built on this rapid-routing-with-guaranteed-delay-bounds algorithm, establishing the path-selection baseline.","marker":"[31]"}],"fun_headline_variants":["Stale IoT directions add 50% to ship evacuation time","One-step IoT lag boosts ship evacuation time by half","IoT delay spikes ship evacuation time by 50%","Even one-step IoT lag slows ship evacuation by half"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The numbers rest on treating every communication delay as exactly one navigation step, independent from node to node, with the same probability everywhere; if real delays are longer, correlated, or sometimes lost messages rather than late ones, the 50% penalty will not transfer.","fun_headline_variants_meta":{"raw":{"variants":["Stale IoT directions add 50% to ship evacuation time","One-step IoT lag boosts ship evacuation time by half","IoT delay spikes ship evacuation time by 50%","Even one-step IoT lag slows ship evacuation by half"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00095,"raw_usage":{"total_tokens":4011,"prompt_tokens":858,"completion_tokens":3153,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":474,"completion_tokens_details":{"reasoning_tokens":3088}},"tokens_in":474,"tokens_out":3153,"duration_ms":22116,"temperature":1.0,"reasoning_tokens":3088,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:52:19.148847+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same ANT/AnyLogic simulation with the one-step per-node delay replaced by delays drawn from a measured congested shipboard network (or a queueing model with the same mean delay), and compare the full-delay average evacuation time to the PoD=0 baseline; reproducing the 50% penalty is required for the claim, and a materially different ratio would show the one-step model is the source of the number.","supporting_citations":[{"cited_title":"Guidelines for evacuation analysis for new and existing passenger ships,","cited_arxiv_id":null,"evidence_quote":"Sets the evacuation time budget (TS=60 min, TA=5 min, TEL=25 min, TD=30 min) that turns the 50% delay penalty into a safety-relevant result."},{"cited_title":"Emergency evacuation simulation study based on improved yolov5s and anylogic,","cited_arxiv_id":null,"evidence_quote":"Provides the AnyLogic pedestrian simulator with social-force movement dynamics in which the evacuation is run."},{"cited_title":"Ant: Deadline-aware adaptive emergency navigation strategy for dynamic hazardous ship evacuation with wireless sensor networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the deadline-aware ANT navigation algorithm and the ship graph that the information-lag study perturbs."},{"cited_title":"Rapid routing with guaranteed delay bounds,","cited_arxiv_id":null,"evidence_quote":"ANT is built on this rapid-routing-with-guaranteed-delay-bounds algorithm, establishing the path-selection baseline."}],"review_version":1}