{"id":"3be59f23-8f7e-4c45-a88e-6921b642dc5c","arxiv_id":"2507.22740","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"This paper provides a survey and conceptual framework for energy-aware protocol design in zero-energy IoT devices, organized around energy information acquisition, task energy profiles, storage dynamics, and harvesting-aware adaptation.","lead":"This paper gives a structured overview of how energy-harvesting IoT devices can measure, predict, and spend their tiny energy budgets through energy-aware protocols. It catalogs energy information acquisition methods, task-level costs, storage dynamics, and adaptive protocol strategies, with illustrative examples drawn from the authors' own prototypes.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper's central claim that small modeling inaccuracies make ZED protocols unworkable is asserted, not demonstrated; Table VIII itself lists sensitivity analysis as an open challenge.","rationale":"After reading the full text, I find the paper's value lies in its structured taxonomy and illustrative examples, but its foundational claim—that realistic protocol design must explicitly model EI overhead, task variability, and storage dynamics because small inaccuracies cause unworkable protocols—is the key assertion that would justify reorienting the field. The paper supports this with selected examples and design insights, but no systematic quantitative evidence. In fact, the authors acknowledge in Table VIII that sensitivity analysis of modeling assumptions is an open challenge, which is a self-flagged limitation. The reader's weakest_assumption about taxonomy completeness is plausible, but a missing category would be a scope limitation; an unsupported necessity claim cuts at the paper's central thesis. My recommended verdict remains CONDITIONAL because the paper can be revised to include a sensitivity analysis or to soften the claim; the concern does not warrant rejection but does require action.","tokens_in":45108,"tokens_out":7640,"duration_ms":84202,"concrete_test":"Implement the task-deferring model of Example 3 (Poisson energy arrivals with mean 0.75 per slot, Bernoulli task arrivals with mean 0.35, task energy 2 units, storage capacity E_M and buffer B as in Fig. 11) and sweep EI acquisition cost E_c from 0.01 to 1.0 units and measurement interval Q from 1 to 50, computing task completion rate over 10^5 slots. Determine whether any (E_c, Q, E_M, B) combination drives the energy-aware scheme below a hard feasibility threshold, e.g., below the energy-blind scheme's best completion rate or below 50%. If no such feasibility cliff exists across a realistic range, the 'small inaccuracies make protocols unworkable' thesis is not supported by the paper's own flagship example. Ideally, the authors also release the simulation code to verify the exact Fig. 11 curves.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section I-B and Remarks 1 and 8 assert that EI-acquisition overhead, task-level variability, and storage dynamics must be explicitly modeled because 'even small modeling inaccuracies may result in unworkable protocols.' This necessity claim is the load-bearing pillar of the paper's proposed research agenda. The supporting evidence, however, consists of qualitative examples (Example 3, Example 5, Example 6) showing performance degradation in selected parameter regimes, not a demonstration that small errors cause a feasibility collapse. Example 6, for instance, shows dynamic RF combining is merely 'prejudicial' when the exploration phase is long, and Example 3 displays a concave completion-rate curve rather than a threshold below which protocols fail. Moreover, Table VIII itself lists 'sensitivity of modeling assumptions' and 'lack of systematic analysis on which physical factors ... dominate system dynamics' as an open challenge, conceding that the impact of these factors has not been quantified. Without such a sensitivity analysis, the paper's prescriptive conclusion that the field must shift to overhead-aware, storage-aware models is an unsupported leap; it remains possible that for typical ZED parameter regimes existing idealized models are adequate, making the proposed shift unnecessary. This is a correctness risk in the central argument, distinct from the taxonomy-completeness concern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper is a survey/tutorial on energy-aware protocol design for zero-energy devices (ZEDs). It structures the design space into three main pillars: energy-information (EI) acquisition (measurement points, methods, overhead, forecasting), operation tasks (sensing, computation, communication, actuation), and energy-usage/storage models (source/load abstractions, harvest-use interaction modes, granularity, storage constraints). It then reviews representative energy-aware protocols and closes with a research agenda. The central thesis is that realistic protocol design must explicitly model EI acquisition overhead, task-level energy variability, and storage dynamics, because even small modeling inaccuracies can make ZED protocols unworkable.","tokens_in":45327,"tokens_out":2999,"duration_ms":41328,"significance":"If the central thesis holds, this is a valuable synthesis: it provides a useful taxonomy of EI acquisition points and methods, concrete overhead models (Eqs. (1), (2), (6)), a systematic task-energy characterization (Tables III and IV), a formal energy-evolution framework (Eq. (12) and Fig. 7), and several in-house examples grounded in prototypes. The paper also explicitly names open challenges, including sensitivity analysis and comparative evaluation. However, the significance is conditional: the paper is primarily a taxonomy and agenda, and its prescriptive conclusion—that the field 'must' shift to overhead-aware, storage-aware models—is not yet supported by quantitative evidence that small modeling errors actually cause feasibility collapse rather than mere performance degradation.","major_comments":[{"comment":"The load-bearing claim that 'even small modeling inaccuracies may result in unworkable protocols' is asserted rather than demonstrated. The supporting examples show performance degradation or parameter-dependent trade-offs, not a feasibility threshold: Example 3 (Fig. 11) shows a concave completion-rate curve, Example 5 (Fig. 14) shows average-AoI differences, and Example 6 (Fig. 15) shows that dynamic RF combining is 'prejudicial' only when the exploration phase is long. None of these establishes that small modeling errors push a realistic ZED from workable to unworkable. Moreover, Table VIII itself lists 'sensitivity of modeling assumptions' as an open challenge, conceding that the impact of these factors has not been quantified. I recommend either adding a quantitative sensitivity analysis (e.g., sweeping model-parameter errors and showing feasibility-region collapse for representative ZED parameter regimes) or softening the necessity claim to conditional guidance, so that the paper's research agenda does not rest on an unproven premise.","section":"Sections I-B, VI; Remarks 1 and 8"},{"comment":"The illustrative simulations that support the central narrative are not reproducible from the manuscript: no code, data, or full parameter tables are provided, and the figures lack error bars or confidence intervals. For a paper whose main claim is that modeling accuracy is critical, the absence of reproducible evidence for the illustrative performance comparisons weakens the argument. Please provide the simulation code/data or, at minimum, a complete parameter listing and a sensitivity check over the key parameters (E_c, Q, F, B, E_M, p, p', δ).","section":"Examples 3 and 5 (Figs. 11 and 14)"},{"comment":"The statement that purely theoretical studies 'often overestimate performance, or worse, incorrectly claim protocol feasibility' is a strong empirical claim, but Table I is a representative selection rather than a systematic comparison. No quantitative evidence is given that existing idealized models actually overestimate performance in realistic ZED scenarios, nor is there a meta-analysis showing how often such overestimation occurs. This claim is used to justify the paper's central necessity argument, so it should either be backed by a systematic comparison of model predictions against prototype measurements or be reframed as a plausible hypothesis.","section":"Section I-A1 (Table I)"}],"minor_comments":[{"comment":"There is a typo: 'indicatie' should be 'indicate'.","section":"Section II-B1"},{"comment":"The phrase 'acqusision' should be 'acquisition'.","section":"Section II-B"},{"comment":"'PWW dimming' appears to be a typo for 'PWM dimming'.","section":"Section V-D, paragraph on actuation intensity"},{"comment":"The text '10 W . . .' in the figure is unclear; please specify whether 10 W is the EIRP of the RF source or the conducted power, and state the number of Monte Carlo runs used for the average net harvested power curves.","section":"Example 6, Fig. 15"},{"comment":"The energy formula for the LED example is typeset without a multiplication symbol between I_F and t, which may confuse readers; please add the missing operator.","section":"Table IV"},{"comment":"The phrase 'V oltage behavior' contains an unintended space; please fix the capitalization and spacing.","section":"Section IV-A3"},{"comment":"The description of per-instruction-level energy tracking would benefit from a citation to a specific intermittent-computing checkpointing system, since the current text refers only to 'some intermittent computing platforms'.","section":"Section IV-C, paragraph on per-instruction granularity"}],"recommendation":"major_revision","confidential_remarks":"This is a broad survey with a strong authorial footprint: many of the cited prototype examples are the authors' own works ([36], [39], [83], [85]), and the literature selection is representative rather than systematic. That is not disqualifying for a vision/foundations paper, but the central necessity claim should be supported by quantitative sensitivity evidence or explicitly downgraded to a research hypothesis. If the authors can supply a small sensitivity study demonstrating a feasibility-region transition for a realistic ZED parameter set, the paper would be considerably stronger."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: read this if you work on zero-energy device protocols; it's a solid organizing framework, not a breakthrough. The new thing is the decomposition of EI acquisition (ETO/ESE/LCP and four measurement methods with energy-overhead formulas), the SHC/CHC/HHC harvest-use modes, and the granularity ladder per-cycle to per-instruction. Those are real refinements, and the paper is at its best when it ties protocols to circuit-level constraints: comparator vs ADC vs coulomb-counter overhead, load models CR/CI/CP, storage leakage and efficiency parameters. The actuator and task tables are also useful reference material.\n\nIt is a survey/tutorial, so don't look for a new theorem or dataset. The examples are mostly didactic re-runs of the authors' own prototypes (solar ARIMA logger, TinyML model selection, batteryless NB-IoT, dynamic RF combining). That is not a defect by itself — the cited prototypes are real and relevant — but it does mean the taxonomy is illustrated rather than validated.\n\nThe soft spot I want to flag is the one the stress-test note raises: Section I-B and the conclusions say 'even small modeling inaccuracies may result in unworkable protocols,' which is the paper's motivation for demanding overhead-aware models. That claim is asserted, not demonstrated. Example 3 shows a concave completion-rate curve, Example 6 shows dynamic combining can be 'prejudicial' with long exploration, but neither shows a feasibility cliff. And Table VIII openly lists 'sensitivity of modeling assumptions' as an open challenge. So the prescriptive conclusion is a reasonable research hypothesis, not a proven necessity. The paper would be stronger if it said exactly that. This doesn't sink the paper, because the paper's value is the classification and research agenda, not the proof of that one claim.\n\nMinor: the two illustrative simulations (Examples 3 and 5) come without code or data, which makes them hard to reproduce. Also, the literature selection is representative rather than systematic, so I wouldn't call it a comprehensive survey. Those are fixable.\n\nBottom line: a serious referee should engage. It deserves publication as a tutorial/foundation paper after the authors soften the necessity claim and make the simulations reproducible. I'd cite the taxonomy if I wrote a ZED protocol paper, and I'd bring it to a reading group as a way to align vocabulary.","headline":"A genuinely useful taxonomy paper, not a new result; the motivating claim about modeling inaccuracy is plausible but unproven, exactly as the paper's own open-challenges table admits.","tokens_in":45865,"tokens_out":2066,"would_cite":true,"duration_ms":23887,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Energy-aware protocols for zero-energy devices only work if their models include the cost of measuring energy itself, task-level energy variability, and the physics of energy storage.","keywords":["zero-energy devices","energy harvesting","energy-aware protocols","energy information acquisition","intermittent computing","energy storage dynamics","energy forecasting","Internet of Things"],"falsifier":"Instrument a capacitor-based ZED with a real ambient source, run a scheduler that assumes free and perfect energy knowledge plus fixed task costs, and compare its measured task-completion rate and brownout count against a scheduler that budgets the actual energy-information cost. If the idealized scheduler sustains its promised performance, the claim that small modeling inaccuracies make protocols unworkable collapses; if it fails as predicted, the claim is supported, and the measured completion-rate curve versus sampling period should show the concave peak that the paper's Example 3 predicts.","tokens_in":44937,"feed_emoji":"⚡","tokens_out":12982,"duration_ms":136983,"temperature":0.7,"pith_summary":"This paper argues that energy-aware protocol design for zero-energy devices (ZEDs), IoT devices powered solely by harvested ambient energy, has been built on models that are too clean, and that this is not a minor shortcoming. Because a ZED's energy budget is razor-thin, even small modeling errors, such as ignoring the energy spent to measure energy, treating every task as consuming the same power, or assuming ideal storage, can turn a supposedly feasible protocol into an unworkable one. The paper therefore provides a structured map of the components a realistic model needs: where and how a device can learn about its energy state, at what overhead; how sensing, computation, communication, and actuation differ in energy cost, granularity, and timeliness; and how storage physics and harvest-use timing constrain what protocols can do. Its aim is to redirect the field from idealized energy-state assumptions toward overhead-aware, storage-aware, task-aware models, and it illustrates the payoff with worked examples spanning MAC scheduling, TinyML inference, task deferral, RF energy combining, and batteryless NB-IoT.","feed_headline":"Energy-model slips can make zero-energy IoT protocols unworkable","feed_subtitle":"A survey argues realistic protocol design must count the cost of sensing energy, task variation, and storage physics.","key_machinery":"The load-bearing machinery is the energy state evolution equation, $E(t) = E(t_0) + \\eta_1 E_H(t_0,t) - (1/\\eta_2) E_L(t_0,t) - E_{\\mathrm{leak}}(t_0,t)$, with capacity bounds $0 \\le E(t) \\le E_M$, which ties every protocol decision to physical energy accounting, together with the taxonomies that feed it: the three EI measurement points and four acquisition methods, the three harvest-use interaction modes (sequential harvest-then-consume, concurrent harvest-and-consume, and hybrid two-buffer harvest-consume), and the four energy-usage granularities (per-cycle, per-task, per-phase, per-instruction). The measurement cost models, for example $P_c = g(L) I_{sb} V_{dd}$ for comparator banks, $E_c = P_{\\mathrm{an}} t_m + C_s V_{dd}^2$ for ADC sampling with capacitance scaling as $2^N$, and $E_c = (t-t_0) P_{\\mathrm{id}} + \\lfloor E_H/\\Delta E \\rfloor E_v$ for energy-integrated accumulation, are what let the paper quantify the overhead that idealized models omit. The granularity-axis-plus-atomicity rule, that each task class must be matched to the granularity at which energy enforcement happens, with atomic tasks designed idempotent and energy-bounded, is the design principle that translates the models into protocol guidance.","core_discovery":"The central claim is that the realism of the underlying energy model, not the cleverness of the scheduling policy, is what decides whether an energy-aware ZED protocol works in practice. Concretely, the paper maintains that three aspects routinely oversimplified in the literature must be modeled explicitly: the energy and time cost of acquiring energy information (EI), the heterogeneity of task-level energy behavior, and the nonlinear dynamics and imperfections of energy storage. To make this operational, it dissects EI acquisition into three measurement points, the energy transducer output, the storage element, and the load consumption points, and four acquisition methods, comparator-based monitoring, information sampling, energy-integrated accumulation, and indirect sensing or time-to-event monitoring, each with its own overhead and error model. It then characterizes tasks by energy profile, execution granularity, and timeliness, and formalizes energy evolution through a state equation balancing harvested, consumed, and leaked energy with storage efficiency factors and capacity bounds. The paper's own examples, such as an age-of-information MAC where a threshold-only partially-aware policy can beat a fully-aware one once measurement cost is counted, and a dynamic RF-combining scheme whose gains evaporate if the exploration phase is too costly, are offered as evidence that including these factors changes protocol conclusions.","pith_inferences":["A natural extension the paper leaves implicit: because measured EI fidelity has a cost, each operating regime has an optimal measurement period and resolution, so the protocol design problem can be framed as a joint stopping-time problem, when to stop sensing energy and commit to an action, rather than as scheduling given free information.","The concave task-completion curves in the paper's Example 3 suggest a testable law: for any ZED there is an optimal EI-sampling frequency balancing measurement cost against decision staleness, and this optimum shifts with the ratio of measurement energy to task energy, a prediction a testbed could verify quantitatively.","The hybrid two-buffer architecture points toward a design exercise the paper does not solve: co-optimizing the split of storage capacity between the low-latency buffer and the bulk buffer against the task mix, treating the split as a protocol parameter rather than a fixed hardware choice.","If the paper's central claim is right, benchmark comparisons of ZED protocols are only meaningful when they state the EI overhead model, the storage model, and the task energy profiles used; otherwise two 'energy-aware' protocols are comparing different physical systems under the same name."],"forward_implications":["Protocol designers must treat EI acquisition as a budgeted operation: every voltage reading, comparator event, or forecast step costs energy and time, and a scheduler that ignores this cost can waste the very energy it is trying to protect.","Task classification by energy profile and granularity becomes a prerequisite for scheduling: computational tasks can be chunked and checkpointed across power cycles, while actuation and security-critical bursts must run as atomic, idempotent units that succeed in one uninterrupted shot.","Storage physics, leakage rate, charge and discharge efficiencies, capacity, and harvest-use mode, should enter protocol logic directly, since the sequential, concurrent, and hybrid architectures support different task sizes, latencies, and leakage penalties.","Coarser EI can outperform richer EI once overhead is counted: a single comparator threshold can beat continuous energy sampling in some regimes, so the choice of EI fidelity is itself a design parameter to optimize.","The field's theoretical and hybrid studies should converge on models that parametrize acquisition overhead, storage dynamics, and task variability rather than assuming them away, which the paper frames as its main open research direction."],"supporting_citations":[{"why":"Hybrid deadline-aware scheduler that accounts for energy-information acquisition overhead (1.2% time overhead) and supplies the sequential harvest-then-consume operating mode.","marker":"[18]"},{"why":"Battery-less LoRaWAN node whose capacitor-voltage threshold and offline-measured task costs ground the paper's threshold-based energy-aware sensing and transmission policy.","marker":"[27]"},{"why":"Battery-less TinyML platform supplying PMU-status energy readings and the local-versus-cloud inference context used in the adaptive-model selection discussion.","marker":"[30]"},{"why":"Energy- and age-aware MAC system model with erasure and transmission probability functions that the paper's fully-aware versus partially-aware comparison (Example 5) is built on.","marker":"[32]"},{"why":"Dynamic RF combining study whose exploration-overhead analysis is the paper's Example 6 and a rare theoretical case where EI acquisition cost enters the model.","marker":"[36]"},{"why":"Reconfigurable energy-storage architecture providing the storage-reconfiguration mechanism and the intermittent/capacity-constrained/temporally-constrained task taxonomy.","marker":"[37]"},{"why":"ARIMA-based solar prediction energy management unit that supplies Example 1 and the quantified 1-2 mW error of indirect energy sensing.","marker":"[39]"},{"why":"Battery-free BLE link protocol that infers recharge intervals from on/off timing, grounding the paper's time-to-event indirect EI method.","marker":"[42]"},{"why":"Adaptive TinyML model selection between a small and a large convolutional network, providing Example 2's storage-voltage-based model choice under fluctuating harvesting.","marker":"[83]"},{"why":"Batteryless NB-IoT prototype with dual voltage thresholds that supplies Example 4's throughput-versus-harvested-power and capacitor-size measurements.","marker":"[85]"}],"fun_headline_variants":["Energy sensing costs decide if zero-energy protocols work","Skip the clever policy, nail the energy model for ZEDs","Zero-energy devices rely on honest models, not algorithm tricks","Why energy harvesting protocols fail without sensor-aware models","ZED protocol success hinges on modeling energy info costs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the paper's taxonomies, three energy-information measurement points, four acquisition methods, three harvest-use modes, and four energy-usage granularities, genuinely span the practically important ZED design space, so that a device or protocol whose central constraint falls outside these categories would not be guided toward workable design by the paper's insights.","fun_headline_variants_meta":{"raw":{"variants":["Energy sensing costs decide if zero-energy protocols work","Skip the clever policy, nail the energy model for ZEDs","Zero-energy devices rely on honest models, not algorithm tricks","Why energy harvesting protocols fail without sensor-aware models","ZED protocol success hinges on modeling energy info costs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000142,"raw_usage":{"total_tokens":1186,"prompt_tokens":982,"completion_tokens":204,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":598,"completion_tokens_details":{"reasoning_tokens":126}},"tokens_in":598,"tokens_out":204,"duration_ms":3807,"temperature":1.0,"reasoning_tokens":126,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T11:20:14.864975+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Instrument a capacitor-based ZED with a real ambient source, run a scheduler that assumes free and perfect energy knowledge plus fixed task costs, and compare its measured task-completion rate and brownout count against a scheduler that budgets the actual energy-information cost. If the idealized scheduler sustains its promised performance, the claim that small modeling inaccuracies make protocols unworkable collapses; if it fails as predicted, the claim is supported, and the measured completion-rate curve versus sampling period should show the concave peak that the paper's Example 3 predicts.","supporting_citations":[{"cited_title":"Scheduling computational and energy harvest- ing tasks in deadline-aware intermittent systems,","cited_arxiv_id":null,"evidence_quote":"Hybrid deadline-aware scheduler that accounts for energy-information acquisition overhead (1.2% time overhead) and supplies the sequential harvest-then-consume operating mode."},{"cited_title":"Energy-aware sensing on battery-less LoRaW AN devices with energy harvesting,","cited_arxiv_id":null,"evidence_quote":"Battery-less LoRaWAN node whose capacitor-voltage threshold and offline-measured task costs ground the paper's threshold-based energy-aware sensing and transmission policy."},{"cited_title":"Towards energy-aware tinyML on battery-less IoT devices,","cited_arxiv_id":null,"evidence_quote":"Battery-less TinyML platform supplying PMU-status energy readings and the local-versus-cloud inference context used in the adaptive-model selection discussion."},{"cited_title":"Energy and Age-Aware MAC for Low-Power Massive IoT","cited_arxiv_id":"2502.08344","evidence_quote":"Energy- and age-aware MAC system model with erasure and transmission probability functions that the paper's fully-aware versus partially-aware comparison (Example 5) is built on."},{"cited_title":"Dynamic RF combining for multi-antenna ambient energy harvesting,","cited_arxiv_id":null,"evidence_quote":"Dynamic RF combining study whose exploration-overhead analysis is the paper's Example 6 and a rare theoretical case where EI acquisition cost enters the model."},{"cited_title":"A reconfigurable energy storage architecture for energy-harvesting devices,","cited_arxiv_id":null,"evidence_quote":"Reconfigurable energy-storage architecture providing the storage-reconfiguration mechanism and the intermittent/capacity-constrained/temporally-constrained task taxonomy."},{"cited_title":"An energy management unit for predictive solar energy harvesting IoT,","cited_arxiv_id":null,"evidence_quote":"ARIMA-based solar prediction energy management unit that supplies Example 1 and the quantified 1-2 mW error of indirect energy sensing."},{"cited_title":"Learning to communicate effec- tively between battery-free devices,","cited_arxiv_id":null,"evidence_quote":"Battery-free BLE link protocol that infers recharge intervals from on/off timing, grounding the paper's time-to-event indirect EI method."},{"cited_title":"Energy-aware tinyml model selection on zero energy devices,","cited_arxiv_id":null,"evidence_quote":"Adaptive TinyML model selection between a small and a large convolutional network, providing Example 2's storage-voltage-based model choice under fluctuating harvesting."},{"cited_title":"Batteryless NB-IoT prototype for bidirec- tional communication powered by ambient light,","cited_arxiv_id":null,"evidence_quote":"Batteryless NB-IoT prototype with dual voltage thresholds that supplies Example 4's throughput-versus-harvested-power and capacitor-size measurements."}],"review_version":1}