{"id":"1d25c65f-c382-4628-bd3f-69716990f3a4","arxiv_id":"2606.18384","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"SCOPE-FL is a synchronous HFL system that applies the TTC algorithm to client selection for simultaneous Pareto efficiency and strategy-proofness, uses One-Round Reconstruction for Shapley rewards, and runs on blockchain.","lead":"SCOPE-FL proposes a hierarchical federated learning framework that models client selection as a school choice problem solved by the Top Trading Cycle algorithm to achieve both Pareto efficiency and strategy-proofness, with blockchain smart contracts and an approximated Shapley value method for rewards. A smart generalist might read it to see how mechanism design tools can address incentive problems that currently limit practical large-scale collaborative AI training.","discovery_kind":"unclear","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest_assumption already isolates the precise translation step that would have to be verified. With only the abstract available, that step cannot be checked for hidden assumptions or domain mismatch, so the unverdicted status is appropriate and no additional load-bearing flaw is detectable.","tokens_in":1762,"tokens_out":237,"duration_ms":29252,"concrete_test":"Locate the section defining the two-sided market (clients vs. aggregators or servers), the preference construction from latency/cost/contribution values, and the statement of the TTC outcome; confirm that the preference domain satisfies the conditions under which TTC is known to be PE and SP (e.g., strict rankings, no indifferences).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on modeling HFL client selection as a two-sided school choice instance whose TTC solution inherits standard PE and SP guarantees, with blockchain enforcing the mechanism in practice. Because the full manuscript text was not supplied beyond the abstract, no internal inconsistency, missing assumption, or mapping error can be located in the argument itself.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes SCOPE-FL, a synchronous hierarchical federated learning (HFL) system that models client selection as a two-sided school choice problem solved by the Top Trading Cycle (TTC) algorithm to simultaneously guarantee Pareto efficiency (PE) and strategy-proofness (SP). Rewards are distributed via a One-Round Reconstruction (OR) approximation to Shapley values, and the full mechanism is executed through blockchain smart contracts to enforce the properties in practice. Experiments on MNIST, Fashion-MNIST, and CIFAR-10 claim superior model accuracy, convergence rate, and reward efficiency compared to DA, IAS, and other baselines, with communication latency comparable to DA and lower blockchain overhead at scale.","tokens_in":1841,"tokens_out":605,"duration_ms":19958,"significance":"If the TTC mapping from HFL attributes (costs, latencies, contributions) to school-choice preferences preserves the standard PE/SP guarantees and the blockchain layer actually enforces truthful reporting, the work would supply a theoretically grounded, incentive-compatible client selection mechanism for HFL. The combination of mechanism design with blockchain execution and an efficient Shapley approximation is potentially valuable for practical federated systems if the modeling assumptions hold.","major_comments":[{"comment":"Abstract and client-selection section: the claim that modeling HFL client selection directly as a two-sided school-choice instance allows TTC to inherit its standard PE and SP properties is asserted without a derivation or explicit verification that the concrete HFL costs, latencies, and model-update contributions satisfy the domain restrictions under which TTC is known to be PE and SP.","section":"Abstract / client selection formulation"},{"comment":"Evaluation section: outperformance is reported on three datasets without error bars, statistical tests, or ablation of the TTC mapping versus the OR reward approximation, so it is impossible to determine whether the claimed gains in accuracy and reward efficiency are robust or attributable to the proposed mechanism.","section":"Evaluation"},{"comment":"Blockchain enforcement paragraph: the assertion that smart-contract execution supplies the tamper-proof environment required for SP to hold in practice is not accompanied by any formal argument or implementation detail showing that the on-chain TTC execution prevents the misreporting that SP is meant to deter.","section":"Blockchain layer"}],"minor_comments":[{"comment":"Abstract: 'pareto efficient' should be 'Pareto-efficient' for standard capitalization.","section":"Abstract"},{"comment":"Abstract: the statement that blockchain overhead is 'significantly lower than DA at scale' should specify the participant counts or network sizes at which the comparison holds.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The provided text is essentially the abstract; without the full manuscript the central mapping and experimental claims cannot be verified, which is the primary reason for the major-revision recommendation rather than a stronger verdict."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments. We address each major comment below and indicate planned revisions.","responses":[{"response":"We acknowledge that the manuscript asserts inheritance of PE and SP from TTC without an explicit derivation verifying that HFL-derived preferences (from costs, latencies, and contributions) satisfy the required domain conditions such as strict ordering. In the revision we will add a subsection deriving the preference mapping and confirming that the HFL instance meets the standard TTC assumptions, including a short proof sketch.","revision_made":"yes","referee_comment":"[Abstract / client selection formulation] Abstract and client-selection section: the claim that modeling HFL client selection directly as a two-sided school-choice instance allows TTC to inherit its standard PE and SP properties is asserted without a derivation or explicit verification that the concrete HFL costs, latencies, and model-update contributions satisfy the domain restrictions under which TTC is known to be PE and SP."},{"response":"The referee correctly notes the absence of error bars, statistical tests, and ablations. We will revise the evaluation section to report results with error bars from repeated runs, include statistical significance tests, and add an ablation isolating the TTC client-selection component from the OR reward approximation.","revision_made":"yes","referee_comment":"[Evaluation] Evaluation section: outperformance is reported on three datasets without error bars, statistical tests, or ablation of the TTC mapping versus the OR reward approximation, so it is impossible to determine whether the claimed gains in accuracy and reward efficiency are robust or attributable to the proposed mechanism."},{"response":"We agree that a dedicated argument linking on-chain TTC execution to prevention of misreporting is missing. The revised manuscript will expand the blockchain section with a concise formal argument showing how smart-contract immutability and consensus enforce truthful preference reporting, thereby realizing the SP guarantee in practice.","revision_made":"yes","referee_comment":"[Blockchain layer] Blockchain enforcement paragraph: the assertion that smart-contract execution supplies the tamper-proof environment required for SP to hold in practice is not accompanied by any formal argument or implementation detail showing that the on-chain TTC execution prevents the misreporting that SP is meant to deter."}],"tokens_in":1481,"tokens_out":479,"duration_ms":28127,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing here is a concrete system that treats client selection in hierarchical federated learning as a two-sided matching problem and solves it with the Top Trading Cycle algorithm to get both Pareto efficiency and strategy-proofness at once, then runs the whole thing on smart contracts with a one-round Shapley approximation for payouts. That combination is new enough to be worth noting for people who care about incentives in distributed training.\n\nWhat the paper does cleanly is name the welfare loss from stability-focused selection rules and the resulting misreporting. The abstract also reports better accuracy, convergence, and reward efficiency than DA and IAS on the three standard image datasets while keeping latency comparable.\n\nThe soft spots are exactly where the reader flagged. The claim that TTC properties carry over once you plug in real HFL costs, latencies, and update contributions is asserted rather than derived in the visible text. No ablation of the mapping, no error bars on the reported gains, and no check that the blockchain layer actually prevents the strategic behavior it is supposed to deter. The OR-Shapley step is presented as scalable but without any stated bound on the approximation error relative to true Shapley values.\n\nThis is for researchers already working on mechanism design inside federated or hierarchical learning who want to see one end-to-end attempt at incentive-compatible selection. It is not yet ready for broader claims about fixing FL incentives in general.\n\nI would send it to referees. The core idea is worth checking even if the current write-up leaves the central mapping and the experimental controls unexamined.","headline":"SCOPE-FL maps HFL client selection to TTC school choice plus blockchain and OR-Shapley, but the abstract gives no evidence the theoretical guarantees survive the translation or that the experiments are robust.","tokens_in":2358,"tokens_out":395,"would_cite":false,"duration_ms":19384,"reading_group":"maybe","serious_thinker":"unclear","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"SCOPE-FL models hierarchical federated learning client selection as a school choice problem and solves it with the Top Trading Cycle algorithm on blockchain to guarantee both Pareto efficiency and strategy-proofness.","keywords":["hierarchical federated learning","client selection","Pareto efficiency","strategy-proofness","Top Trading Cycle","blockchain smart contracts","Shapley value approximation"],"falsifier":"A controlled experiment in which clients submit false preference reports and the resulting allocation is shown to be Pareto dominated by another feasible allocation that would have been chosen under truthful reporting.","tokens_in":2663,"feed_emoji":"🔗","tokens_out":651,"duration_ms":30865,"temperature":0.7,"pith_summary":"The paper proposes SCOPE-FL to fix strategic inefficiencies in hierarchical federated learning client selection. Current methods favor stability and allow participants to misreport preferences, producing worse overall outcomes. SCOPE-FL recasts selection as a two-sided school choice problem solved by the Top Trading Cycle algorithm, which delivers both Pareto efficiency and strategy-proofness. The full process runs inside blockchain smart contracts for tamper-proof enforcement, paired with an approximated Shapley value for proportional rewards. Experiments on MNIST, Fashion-MNIST, and CIFAR-10 show gains in accuracy, convergence speed, and reward fairness over prior methods while keeping latency comparable.","feed_headline":"TTC algorithm on blockchain delivers efficient truthful client selection","feed_subtitle":"By recasting selection as school choice and running it tamper-proof, the system raises accuracy and convergence without strategic misreprese","key_machinery":"The Top Trading Cycle algorithm applied to a two-sided school choice formulation of client selection, executed on blockchain smart contracts to enforce the properties in practice.","core_discovery":"SCOPE-FL formulates client selection as a two-sided school choice problem solved through the Top Trading Cycle (TTC) algorithm that simultaneously guarantees PE and SP. For reward distribution, SCOPE-FL employs a scalable Shapley value approximation based on One-Round Reconstruction (OR), ensuring compensation proportional to each client's contribution. The entire mechanism executes via blockchain smart contracts, providing the tamper-proof environment required for the SP guarantees to hold in practice.","pith_inferences":["The same matching approach could be tested in other resource-allocation settings where participants hold private information about costs.","Dynamic client availability over multiple rounds would require an extension of the static school-choice model.","Tamper-proof execution opens the possibility of third-party audits of selection fairness in regulated environments."],"forward_implications":["Clients have no incentive to misrepresent preferences, so reported data matches true costs and capabilities.","Resource allocations reach Pareto efficiency, improving overall model accuracy and convergence rate.","Rewards scale with contribution via the one-round Shapley approximation without requiring full recomputation.","Blockchain execution keeps communication latency similar to existing methods while lowering overhead at larger scales."],"fun_headline_variants":["TTC on blockchain for strategy proof Pareto efficient HFL","School choice TTC solves for PE and SP in federated learning","Blockchain enables strategy proof client selection with TTC","Scalable Shapley approximation rewards client contributions in HFL","Two sided school choice via TTC for optimal Pareto federated learning"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"That the abstract properties of TTC in school choice problems survive translation into the concrete costs, latencies, and contribution values of actual federated learning participants.","fun_headline_variants_meta":{"raw":{"variants":["TTC on blockchain for strategy proof Pareto efficient HFL","School choice TTC solves for PE and SP in federated learning","Blockchain enables strategy proof client selection with TTC","Scalable Shapley approximation rewards client contributions in HFL","Two sided school choice via TTC for optimal Pareto federated learning"]},"model":"grok-4.3","cost_usd":0.005435,"raw_usage":{"total_tokens":2627,"prompt_tokens":691,"num_sources_used":0,"completion_tokens":78,"cost_in_usd_ticks":54349500,"prompt_tokens_details":{"text_tokens":691,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1858,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":691,"tokens_out":78,"duration_ms":16698,"temperature":1.0,"reasoning_tokens":1858,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T01:21:56.804558+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled experiment in which clients submit false preference reports and the resulting allocation is shown to be Pareto dominated by another feasible allocation that would have been chosen under truthful reporting.","supporting_citations":[],"review_version":1}