{"id":"d333aa4c-6365-494a-8dfc-40c5772d4fd8","arxiv_id":"2411.13227","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"A trust-score-based rule for splitting traffic across AOMDV paths is derived, but no experiments show that it reduces delay.","lead":"This paper sketches a way to make the AOMDV wireless routing protocol prefer reliable, well-charged, un-congested nodes by giving each link a trust score and splitting data across paths in proportion to that score. It derives a simple split formula, but does not test the idea in a simulator or a real network, so the promised delay reduction is not demonstrated.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The delay-minimization claim in Eq. (10) treats p1 and p2 as constants, but the paper's own availability index in Eq. (4) makes the per-path success probability depend on the amount of data assigned to that path, so the independent-path objective max(f1/p1, (1-f1)/p2) is not the correct one.","rationale":"Reading in good faith, the paper's core algebra in Section III-C is correct under the stated assumption that per-path delay is proportional to allocated data and inversely proportional to a fixed trust value p. The two-path minimizer of max(f1/p1, (1-f1)/p2) is indeed f1 = p1/(p1+p2). However, the manuscript is not merely missing a simulation; it contains an internal tension. The trust value p in Eq. (6) includes the Availability Index Ai(t) defined in Eq. (4), which decreases with transmission activity. Once data is split into f and 1-f, the nodes on each path transmit more or less, changing Ai and therefore p. The optimization in Section III-C holds p fixed, so it cannot be the minimizer of the actual delay model the paper has proposed. This is a stronger and more specific version of the independence concern the reader identified: it is grounded in the paper's own coupling equation, not just general wireless interference. A concrete analytical check using the steady-state version of Eq. (4) would settle whether Eq. (10) survives the coupling; if not, the central optimality claim must be revised or heavily qualified. I agree with the reader's weakest assumption to a large extent; my concern is a sharper form. Because the reader already conditioned the verdict on justification of the trust metric and validation, this concern reinforces that conditionality rather than changing it. Hence verdict remains UNCHANGED.","tokens_in":3111,"tokens_out":13686,"duration_ms":148030,"concrete_test":"Analytically substitute Eq. (4) into Eq. (6): model a node on path i as transmitting for a fraction f_i of time during the transfer, so its steady-state availability is approximately A_i(f_i) = min(1, r/d * (1 - f_i)) (or the exact solution of the min-plus recursion). Then p_i(f_i) = (alpha/N) f(B) A_i(f_i) / hops. Re-solve min max(f1/p1(f1), (1-f1)/p2(1-f1)) and check whether the minimizer equals f1 = p1/(p1+p2) evaluated at the optimum. If the fixed point differs, Eq. (10) fails under the paper's own coupling model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III-C derives fi = pi / sum pj by minimizing max(f1/p1, (1-f1)/p2), which assumes each path's expected delay is f_i/p_i with p_i fixed. But p_i is not fixed: Eq. (6) multiplies the Beta trust value by the Availability Index Ai(t) from Eq. (4), and Ai(t) degrades by d whenever the node transmits. Increasing the fraction f_i of data sent over path i increases the transmission activity of nodes on that path, thereby reducing Ai(t) and hence p_i. The optimization treats p1 and p2 as exogenous constants, but they are endogenous functions of the allocation f. Including this feedback turns the minimization into a fixed-point problem, and the closed-form f1 = p1/(p1+p2) is generally not the solution. The paper's own admission in Fig. 3 that local optimization can be globally suboptimal, and the future-work item on queue lengths, acknowledge that load affects path quality, contradicting the constant-p premise. Thus the central 'optimal split' claim is not established by the derivation as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a trust-based enhancement to the AOMDV routing protocol. It models link trust with a Beta distribution (Eqs. (1)-(2) and Algorithm 1), combines that with a predicted battery level, a per-node Availability Index Ai(t) that degrades during transmission, and a hop-count penalty into a composite link probability (Eq. (6)). It then derives the expected delay of a stop-and-wait transmission as 2t/p (Eqs. (7)-(8)) and argues that, to minimize the worst-case delay when splitting traffic across paths, the split should be proportional to the path trust values (Eqs. (10)-(11)). The paper is purely analytical; the conclusion explicitly defers simulation and experimental validation to future work.","tokens_in":3348,"tokens_out":7797,"duration_ms":80919,"significance":"If the central derivation were valid, the paper would provide a simple closed-form, decentralized load-splitting rule for multipath routing: each path receives a fraction equal to its trust value divided by the sum of trust values. The Beta expectation proof in Eq. (2) is correct, and Eqs. (7)-(11) are internally coherent under the stated stop-and-wait and delay-proportional-to-load assumptions; Eq. (10) is a genuine algebraic consequence of the minimax objective. The main value of the manuscript is therefore as a clearly stated design proposal with a falsifiable prediction. However, the connection between the heuristic trust score and a true per-transmission success probability, and the treatment of path quality as exogenous to the allocation, are not established; these gaps currently prevent the optimality claim from being accepted.","major_comments":[{"comment":"The quantity p_{A→B,t} in Eq. (6) is a product of heuristic submetrics, not a well-defined per-transmission success probability. In particular, α/N is an empirical success fraction over the last N observations and differs from the Beta expected value α/(α+β) used in Eq. (2) and from the continuation of the Beta model; multiplying by f(Bt), Ai(t), and 1/num hops produces an unnormalized composite score. The recurrence in Eq. (7) requires p to be the constant probability that a single transmission attempt succeeds, so E[delay]=2t/p and the allocation result are not justified for the score defined in Eq. (6).","section":"Section III-A5 and Section III-B, Eqs. (6)-(8)"},{"comment":"The optimization treats p1 and p2 as fixed, but Eq. (4) makes Ai(t) decrease by d when a node transmits, and Eq. (6) includes Ai(t) multiplicatively. Increasing the fraction f_i of data assigned to path i increases transmission activity on that path, reducing Ai(t) and hence reducing p_i. Thus p_i is endogenous to the allocation f_i, and the true minimization is a fixed-point problem; the closed-form f_i = p_i / Σ_j p_j is generally not the minimizer of the actual expected delay. The paper's own Fig. 3 and the future-work item on queue lengths acknowledge that load affects path quality, which contradicts the constant-p premise.","section":"Section III-C, Eqs. (10)-(11), and Section III-A3, Eq. (4)"},{"comment":"The assertion that 'Delay ∝ amount of data being transmitted' and that each path's delay depends only on its own trust value and load ignores shared-medium interference and possible overlap of AOMDV paths. Even if AOMDV returns node-disjoint or link-disjoint paths (Section II-B), simultaneous transmissions on different paths can still interfere at the physical layer or queue at shared intermediate or neighboring nodes, and these interactions are not captured by the per-path linear objective. Hence the minimax step establishes at best a conditional result for isolated, independent paths, not the 'optimal' distribution claimed in the abstract.","section":"Section III-C, assumption before Eq. (10)"},{"comment":"The paper contains no simulation or protocol-level evaluation; the conclusion explicitly states that simulating the method is 'the next step'. For a routing-protocol enhancement with several free parameters (γ, r, d, N, α0, β0) and no calibration procedure, analytical derivations under idealized assumptions are insufficient to demonstrate that the proposed scheme improves AOMDV in delay, throughput, or reliability. At minimum, a simulation study on packet delivery fraction, end-to-end delay, and overhead is needed to make the central claims testable.","section":"Section IV, Conclusions"}],"minor_comments":[{"comment":"Eq. (11) states that n is the number of neighbors of the transmitting node, but the data-distribution problem is over multiple paths to a given destination; not every neighboring node necessarily has a valid path, so the universal quantification over neighbors is imprecise.","section":"Section III-C, Eq. (11)"},{"comment":"Eq. (4) mixes the time-scaled quantity rΔt with the per-transmission decrement d in the same updating expression; please clarify the intended time interval for one degradation decrement.","section":"Section III-A3, Eq. (4)"},{"comment":"The derivation of Eq. (10) is omitted; adding the one-line equalization argument (set f1/p1 = (1-f1)/p2 and solve) would improve reproducibility.","section":"Section III-C, Eq. (10)"},{"comment":"Algorithm 1 increments α and β indefinitely, but the text says α and β are determined from only the latest N observations; the algorithm does not describe the sliding-window deletion of older observations.","section":"Section III-A1, Algorithm 1"},{"comment":"Eq. (3) uses two closest battery observations Bt0 and Bt1 but does not specify how these are stored, synchronized, or refreshed in a distributed MANET.","section":"Section III-A2, Eq. (3)"},{"comment":"The phrase 'simultaneously transmitting through multiple paths' is in tension with the stop-and-wait model in Eq. (7), which sends a new packet only after the previous ACK; please clarify whether the scheme is concurrent-flow or interleaved-serial.","section":"Abstract and Section III-B, Eq. (7)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript explicitly acknowledges being developed in a course, and it reads as a proposal rather than a completed evaluation: the mathematics is coherent in the narrow conditional sense, but the mapping from heuristic scores to probabilities and the exogeneity of path trust are load-bearing and need reworking. I would not reject outright because the proportional-split theorem can be salvaged, but I would expect a substantial revision, ideally with a simulation study, before publication in a networking venue. The citation list is very short and does not engage with the broader trust-routing and multipath-routing literature; the editor may wish to request a more complete related-work survey."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nShort version: this is a cleanly written design sketch for a trust-aware load split in AOMDV. The Beta reputation update and the stop-and-wait delay algebra are correct, and the paper is honest that it contains no simulation. But the main optimization result does not follow from the paper's own model, because the per-path success probability p_i is treated as a constant when it actually depends on the data fraction f_i through the availability index in Eq. (4). That is a load-bearing gap, not a nit.\n\nWhat's new: relative to the two cited references (AODV RFC 3561 and Marina-Das AOMDV), the composite trust score in Eq. (6) and the split rule f_i = p_i / sum p_j are not present. The derivation from the delay model to that rule is algebraically correct under a fixed-p independence assumption. The paper also states its limitations clearly in the future-work list, including local-vs-global optimality and queue length.\n\nSoft spots, in proportion: (1) The endogeneity problem is real and central. Eq. (4) degrades availability whenever a node transmits, so increasing f_i lowers A_i(t) and hence p_i. The minimax objective max(f1/p1, (1-f1)/p2) ignores this feedback, so the closed-form split is not generally the optimum. (2) Eq. (6) is an ad hoc product of four factors with unspecified parameters and no calibration or sensitivity analysis. (3) No simulation or baseline comparison appears anywhere; the conclusion explicitly says simulation is the next step, which contradicts the abstract's claim that the scheme reduces delay. (4) Related work is thin: only two references, with no engagement of the substantial trust-based MANET routing literature.\n\nIf you read the paper as a class-project report, it is internally readable and the math that is done is done correctly. As a research contribution, the central claim is unproven, and the one clear derivation rests on an assumption the paper's own model contradicts. I would not cite this in my own work, and I would not send it to a full journal as is. The fix is straightforward: redo the optimization with load-dependent p, or justify the constant-p approximation, and add at least a small simulation. Until then, it is a work-in-progress, not a published result.","headline":"A readable course-project-style design sketch whose split rule doesn't survive the paper's own load-dependent trust model; no simulation, so the delay-reduction claim is unsupported.","tokens_in":3941,"tokens_out":2361,"would_cite":false,"duration_ms":26023,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Splitting AOMDV traffic in proportion to path trust minimizes expected delay.","keywords":["AOMDV","trust-based routing","Beta distribution","optimal data distribution","delay minimization","mobile ad hoc networks","multipath routing"],"falsifier":"A testbed or simulation with two AOMDV paths that share a common intermediate node, where path 1 has much higher trust than path 2, would refute the rule if the trust-proportional split produces higher end-to-end delay than a deliberately equal split under the same traffic load.","tokens_in":2815,"feed_emoji":"📡","tokens_out":3848,"duration_ms":38839,"temperature":0.7,"pith_summary":"This paper tries to show that a mobile ad hoc network using the AOMDV multipath routing protocol can reduce the expected delay of a data transfer by splitting traffic across paths in proportion to each path's computed trust value. The proposal adds a probabilistic trust model, built from a Beta-distributed estimate of acknowledgment success, battery level, node availability, and hop count, to a protocol that currently chooses routes without regard to node reliability. If the derivation is right, the optimal split for two paths is $f_1 = p_1/(p_1+p_2)$, and for $n$ paths it is $f_i = p_i / \\sum_j p_j$, so a node can choose fractions purely from local reliability estimates. A sympathetic reader would care because this is a parameter-free allocation rule that promises lower delay in multi-path ad hoc routing without changing the underlying route-discovery machinery.","feed_headline":"Trust-proportional split minimizes AOMDV delay","feed_subtitle":"Sending data over each path in proportion to its reliability gives a closed-form split that minimizes expected delivery time.","key_machinery":"The load-bearing object is the minimax allocation rule. With the stop-and-wait expected delay $E[\\text{delay}] = 2t/p$ for each path, and the assumption that a path's delay scales linearly with its assigned data fraction, the delay of path $i$ is proportional to $f_i/p_i$. Minimizing the largest of these, $\\max_i f_i/p_i$, gives $f_i = p_i / \\sum_j p_j$. This split is driven by the trust model: $p$ is a Beta-distributed link reliability updated by ACK outcomes, then adjusted by battery level $f(B_t)$, an availability index $A_i(t)$ that accounts for recent transmissions, and the number of hops to the destination.","core_discovery":"The paper claims that when a node transmits simultaneously over multiple paths, the expected delay of the whole transfer is minimized by splitting data in proportion to each path's trust value. For two paths the optimal fraction is $f_1 = p_1/(p_1+p_2)$; for $n$ paths it generalizes to $f_i = p_i / \\sum_j p_j$. This rule follows from modeling each path's delay as inversely proportional to its trust value and proportional to the amount of data sent on it, then minimizing the maximum path delay. The trust values themselves are computed from a Beta-distribution model of ACK success, battery level, node availability, and hop count, with a windowed update rule that forgets very old observations.","pith_inferences":["Editorial extension: The independence assumption is most plausible when paths are node-disjoint and spatially separated; if AOMDV paths share intermediate nodes or spectrum, the optimal split under contention would likely differ from $p_i / \\sum_j p_j$, sending less traffic to a path that shares a bottleneck.","Editorial extension: A natural simulation test would compare trust-proportional splitting against equal splitting and against a predictive equal-delay operating point, measuring both mean and tail end-to-end delay under node mobility and packet loss.","Editorial extension: The proposed local rule could be ported to multipath transport protocols like MPTCP by mapping each subflow's loss or throughput to a trust value $p$, giving a congestion-independent splitting rule.","Editorial extension: The paper's own admission that local optimization can be globally suboptimal suggests that aggregating trust over full paths, rather than first-hop trust only, would likely improve the split at the cost of route-discovery overhead."],"forward_implications":["If correct, AOMDV nodes can choose split ratios locally from trust estimates without extra control messages, reducing expected transfer delay.","The Beta-prior update rule gives a memory-limited trust estimate that can react to node misbehavior or battery drain over time.","The minimax derivation yields a closed-form $n$-path split, so the rule can be applied at any node with multiple neighbors to the destination.","Because the derivation works for any path reliability $p$, the same proportion rule could incorporate loss rate, delay, or energy metrics into one allocation.","The scheme preserves the underlying AOMDV route discovery, so it can be layered onto existing multipath routing without redesigning the network layer."],"supporting_citations":[{"why":"Defines AODV, the base reactive routing protocol whose route discovery and message types the trust enhancement modifies.","marker":"[1]"},{"why":"Defines AOMDV, the multipath extension that supplies the multiple loop-free paths over which the trust-based data split is applied.","marker":"[2]"}],"fun_headline_variants":["Split data by trust to cut AOMDV delay","Proportional trust routing reduces AOMDV latency","AOMDV: send more on trusted paths","Trust-weighted multipath cuts worst-case delay","Optimal split: data proportional to trust in AOMDV"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire allocation rule rests on treating each path as an independent channel whose delay depends only on its own assigned load, so transmissions on different paths do not interfere and paths do not share bottleneck nodes.","fun_headline_variants_meta":{"raw":{"variants":["Split data by trust to cut AOMDV delay","Proportional trust routing reduces AOMDV latency","AOMDV: send more on trusted paths","Trust-weighted multipath cuts worst-case delay","Optimal split: data proportional to trust in AOMDV"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000225,"raw_usage":{"total_tokens":1368,"prompt_tokens":756,"completion_tokens":612,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":372,"completion_tokens_details":{"reasoning_tokens":538}},"tokens_in":372,"tokens_out":612,"duration_ms":5462,"temperature":1.0,"reasoning_tokens":538,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:41:45.908797+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A testbed or simulation with two AOMDV paths that share a common intermediate node, where path 1 has much higher trust than path 2, would refute the rule if the trust-proportional split produces higher end-to-end delay than a deliberately equal split under the same traffic load.","supporting_citations":[{"cited_title":"Ad hoc on-demand distance vector (AODV) routing,","cited_arxiv_id":null,"evidence_quote":"Defines AODV, the base reactive routing protocol whose route discovery and message types the trust enhancement modifies."},{"cited_title":"Ad hoc on-demand multipath distance vector routing,","cited_arxiv_id":null,"evidence_quote":"Defines AOMDV, the multipath extension that supplies the multiple loop-free paths over which the trust-based data split is applied."}],"review_version":1}