{"id":"db78497b-da21-4846-8d22-31b64d2c78d7","arxiv_id":"2504.14959","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A configuration anonymization framework that injects realistic fake network devices and repairs routes to hide true network size while preserving forwarding behavior.","lead":"NetCloak is a system that anonymizes network configuration files by adding fake routers and hosts so the true size of a network stays hidden, while trying to keep traffic routing behavior unchanged. It matters because real network configurations are needed for research and debugging, but organizations rarely share them for fear of leaking internal scale and structure.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"k-DMA threat model leaks network scale: knowing the original degree sequence reveals the router count, so NetCloak's stated privacy model cannot support the central 'conceals scale' claim.","rationale":"The paper is a well-structured systems contribution: the embedding algorithm, k-DMA definitions, mimicry-based configuration generation, and layered SMT/iterative repair are clearly described, and the evaluation covers multiple topologies. The reader's CONDITIONAL verdict is reasonable from an evaluation standpoint. However, my stress-test pass finds a more fundamental issue that the reader's weakest-assumption statement only partially captures. The central claim is that NetCloak 'effectively conceals network size' or 'obfuscate[s] true scale.' The formal privacy mechanism offered for this is k-DMA, and Section 4.2 and Definitions 3.1-3.2 state the adversary 'knows only the original degree sequence.' The original degree sequence for the router graph G has length equal to the number of original routers, so the claimed secret (network scale) is an explicit input to the adversary. k-DMA constrains the number of candidate nodes per degree rank but places no constraint on the total number of real nodes; it is an identity-anonymization notion, not a scale-hiding notion. Therefore the abstract's headline privacy claim does not follow from the paper's formal model, regardless of how well the topology expansion or config mimicry works. This is an internal inconsistency, not a matter of out-of-consensus hardness assumptions. A concrete, cheap test is to run the stated adversary on any dataset (e.g., network A) with just the original degree sequence; the true router count is immediately output. The reader's auxiliary-signal concern (configuration style, host counts, traffic) is real but secondary; even without auxiliary signals, the model leaks scale. I recommend REJECT for the central claim as stated, though the paper could be reframed as an identity-anonymization system under degree-knowledge adversaries, which would be a weaker but defensible contribution.","tokens_in":24894,"tokens_out":8071,"duration_ms":72576,"concrete_test":"Analytical check on a paper dataset, e.g., network A: compute the original router degree sequence D used in Section 4.2 and note that |D| equals the number of routers. Then examine Definitions 3.1 and 3.2 to confirm they only require counts of nodes with degree at least deg(v_i) in the anonymized graph and do not restrict the number of original nodes. Finally, simulate the stated adversary by giving it D and the anonymized topology, and measure whether it can report the true router count with probability 1. If yes, the k-DMA guarantee does not deliver scale concealment, and the abstract's central claim is unsupported.","verdict_should_be":"REJECT","load_bearing_attack":"NetCloak's central claim is that it effectively conceals network size by injecting synthetic routers and hosts. The formal guarantee invoked for this is k-degree mapping anonymity (Definitions 3.1-3.2, Section 4.2). The definitions and Section 4.2 explicitly assume the adversary knows the original network's degree sequence: 'k-degree mapping anonymity assumes attackers know only the original degree sequence.' But the degree sequence of the router graph G=(V,E) has exactly |V| entries, i.e., the number of routers. An adversary granted that knowledge already knows the true network scale with certainty. k-DMA only limits which anonymized nodes can correspond to which original nodes based on degree; it imposes no constraint on the total number of real routers. Thus, under the paper's own threat model, injecting synthetic nodes cannot conceal scale because the secret is an input to the attacker. This is an internal inconsistency, not a disagreement with an external consensus: the motivating scenario (Section 2.3) is precisely that an adversary should not infer organizational size from the shared configuration, and the abstract promises to 'obfuscate true scale.' The formal model gives the adversary the scale as background knowledge. No amount of topological rationality or configuration mimicry fixes this mismatch; even a perfect degree-matched expansion leaves |D| = n_original directly readable. The reader's concern about auxiliary signals is valid but secondary: the stated adversary model already leaks the protected attribute.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents NetCloak, a configuration-anonymization framework that injects synthetic routers and hosts into a network topology to obscure network scale while preserving end-to-end forwarding behavior. The system selects a reference topology, embeds the original graph into it, applies a new k-degree mapping anonymity (k-DMA) scheme, generates fake configurations via template mimicry, and repairs routing using a combination of SMT-based intra-AS synthesis and iterative inter-AS filtering. Experiments on real and emulated campus/data-center topologies report a 73.5% improvement in topological rationality, a 29.0% increase in configuration similarity, and roughly 60% faster route repair relative to baselines. The paper is positioned as an extension of ConfMask that overcomes its inability to add router nodes.","tokens_in":25115,"tokens_out":5938,"duration_ms":54346,"significance":"If the central claim held, NetCloak would be a useful step toward privacy-preserving sharing of realistic network configurations, extending prior work beyond link/host manipulation to node-level scale obfuscation. The paper has clear strengths: it presents a complete implemented system built on ConfMask and NetComplete, evaluates on a diverse set of 12 networks, candidly discusses limitations in Section 8, and identifies a concrete gap in prior anonymization tools. The threat-model problem described below, however, means the headline privacy claim is not currently supported by the formal framework; the engineering and evaluation contributions are real but the core privacy assertion needs substantial revision.","major_comments":[{"comment":"The formal threat model is internally inconsistent with the paper's central claim of concealing network scale. Definition 3.2 and Section 4.2 state that k-degree mapping anonymity assumes the attacker knows only the original degree sequence. That sequence has exactly |V| entries, so an attacker granted this knowledge already knows the original router count with certainty. Consequently, injecting synthetic nodes cannot conceal scale from the stated adversary; the k-DMA guarantee only limits which anonymized nodes can be mapped to which original nodes, not the total number of real routers. The abstract and Section 2.3 promise that NetCloak 'obfuscates true scale,' but the formal model gives the scale to the attacker as background knowledge. The evaluation also never measures how well an adversary can estimate the original node count from the anonymized graph. This is a load-bearing mismatch: either the threat model should be revised to an adversary who does not know the original degree sequence (with an empirical scale-hiding evaluation), or the 'conceals network size' claim should be replaced by a narrower claim about hiding node identities and degree mappings.","section":"§2.3, §3.1, §4.2"},{"comment":"Topological rationality is defined as the K-S distance between the degree sequence of the anonymized graph and that of the reference graph G_ref, and the embedding algorithm in Section 4.1 explicitly minimizes this exact distance during construction (node mapping, edge completion, and edge rearrangement all target degree-sequence closeness to G_ref). The reported 73.5% improvement in topological rationality over baselines is therefore partly by construction: the metric is the optimization objective of the proposed method but not of the compared baselines. To make the claim meaningful, the paper should report additional structural fidelity metrics that the algorithm does not directly optimize (e.g., clustering coefficient, diameter, joint degree distribution, or spectral properties), or explicitly frame the 73.5% as 'degree-sequence agreement with a chosen reference.'","section":"Definition 3.3, §4.1, §6.3"}],"minor_comments":[{"comment":"The configuration similarity metric computes, for each fake configuration, the maximum similarity to any real configuration and then averages these maxima. Since fake configurations are generated by selecting a real router as a template, this metric is favorable by construction; reporting the average similarity over real configurations (or a leave-one-out baseline) would give a less biased view of style fidelity.","section":"§6.2"},{"comment":"The definition of 'K-S distance between degree sequences' should specify that it is the Kolmogorov–Smirnov statistic between the empirical cumulative distribution functions of the two degree sequences, and how ties or differing sequence lengths are handled.","section":"Definition 3.3"},{"comment":"Line 19 of Algorithm 1 contains a typo: 'create G_emb with all all edges of G' should read 'all edges of G.'","section":"Algorithm 1"},{"comment":"The discussion of differential privacy refers to the 'post-processing immunity theorem'; the standard name is the post-processing property of differential privacy. The argument is correct in substance, but the terminology should be aligned with the literature.","section":"§8"},{"comment":"Reference [23], the Netconan anonymizer, lists the URL as https://internet2.edu/; this appears to be an incorrect URL for the Netconan tool and should be corrected.","section":"References"},{"comment":"The RMSE values for node-addition accuracy (4.31 for embedding, 7.20 for replica) are reported without describing how the target set was chosen or whether the differences are statistically significant; a brief clarification of the RMSE computation would improve reproducibility.","section":"§6.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid systems effort and the engineering is credible, but the central privacy claim is not supported by the formal model as written. The k-DMA threat model gives the adversary the original degree sequence, whose length immediately reveals the router count; this is a logical contradiction with the abstract's 'conceals network size' promise. I would not recommend rejection because the flaw is fixable: the authors could revise the threat model (e.g., an adversary who observes only the anonymized graph and has no prior degree-sequence knowledge) and add a direct evaluation of scale-hiding, or they could reframe the contribution as node-identity mapping anonymity rather than scale obfuscation. The rationality metric being the algorithm's own objective is a second issue that should be addressed with independent structural metrics. If the authors make those changes, the paper could be acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take on NetCloak. The core system is genuinely new: it's the first configuration anonymizer I know that injects entire synthetic routers and hosts, not just links, and the mimicry-based config generation plus layered SMT/iterative repair is a plausible engineering contribution. The k-DMA definition is a real addition to the graph anon literature, and the evaluation covers a reasonable spread of topologies. Credit where due: the authors also honestly report that SMT repair is worse under default link costs.\n\nBut there's a load-bearing problem. The paper's headline promise is 'concealing network size.' The formal privacy guarantee, k-DMA (Defs 3.1-3.2, Sec 4.2), explicitly assumes the adversary knows the original degree sequence. A degree sequence of a graph with n routers has exactly n entries. So the adversary already knows the true scale with certainty. Injecting fake nodes under that model cannot obfuscate scale; the secret is an input to the attacker. That's not a quibble about auxiliary signals—it's an internal inconsistency with the stated threat model. The reader's concern about configuration style and traffic patterns is secondary; the stated model already leaks the protected attribute.\n\nThe other weaknesses are proportionally softer. 'Topological rationality' is defined as K-S distance to the reference graph that the embedding algorithm already minimizes against, so the 73.5% gain is partly by construction. Configuration fidelity is measured as max similarity to the real configs used as templates. Headline numbers are point estimates without error bars. None of these are fatal on their own, but they compound the central privacy problem.\n\nWho is this paper for? Researchers building config-sharing platforms or synthetic network generation. The system could still be useful for enlarging datasets even if it doesn't provide the advertised scale privacy. But the abstract's claim needs to be rewritten, and the threat model needs a fix: either drop the scale-concealment claim or give the adversary less knowledge.\n\nMy recommendation: this deserves a serious referee—it has real new machinery and an honest evaluation—but it should not be accepted without major revision. The scale leak alone warrants a reject-and-resubmit.","headline":"k-DMA's threat model leaks the network scale the paper promises to hide, so the central privacy claim fails; the system engineering is real but needs a major rewrite.","tokens_in":25691,"tokens_out":3155,"would_cite":false,"duration_ms":27001,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"NetCloak claims to conceal a network's true size by embedding it in a larger synthetic topology while leaving every real forwarding path intact.","keywords":["network configuration anonymization","network scale concealment","topology expansion","graph embedding","k-degree mapping anonymity","configuration mimicry","routing repair","functional equivalence"],"falsifier":"Run a deanonymization game: give an adversary the anonymized configuration plus a candidate original network with fewer routers, and ask it to decide whether the original router count equals the small candidate or the expanded count. The scale-concealment claim fails if the adversary succeeds by counting host subnets, matching organizational records, or exploiting stylistic fingerprints, rather than by using the degree sequence that $k$-DMA is designed to protect.","tokens_in":24636,"feed_emoji":"🕸️","tokens_out":7484,"duration_ms":65272,"temperature":0.7,"pith_summary":"NetCloak is a configuration-anonymization system whose goal is to let a network owner share real router configuration files without revealing the network's true size. It expands the topology by embedding the original graph as a subgraph of a larger reference topology, then writes configuration files for the added routers and hosts by imitating real devices, so fake nodes do not stand out stylistically. It then repairs routing so that every original host-to-host path is preserved, meaning the anonymized files remain usable for troubleshooting, research, and verification. Against the baselines, the paper reports a 73.5% improvement in topological rationality, a 29.0% increase in configuration similarity, and a 60.0% reduction in route-repair time under randomized link costs. The stakes: organizations could publish realistic but padded configurations, and researchers could obtain larger, realistic datasets without collecting sensitive real ones.","feed_headline":"Fake routers hide a network's true size while routing stays intact","feed_subtitle":"Embedding a real topology into a larger reference graph lets organizations share configs without leaking true scale.","key_machinery":"The load-bearing machinery is the graph-embedding expansion plus the $k$-degree mapping anonymity condition. Embedding is a maximum bipartite matching between original nodes and reference nodes of no smaller degree, followed by Havel-Hakimi-style residual degree completion and greedy edge rearrangement; it determines which nodes exist and how many links each one has. $k$-DMA is the privacy condition that makes those added nodes protective: because every original node has at least $k$ peers of equal or higher degree in the anonymized graph, degree information alone cannot identify the real devices. The functional carrier is the layered repair: inside an AS, SMT constraints encode the required shortest paths, with a correction distinguishing a unique primary path from ECMP alternatives, and between ASes, iterative comparison of border-router forwarding tables inserts filters until both views agree. Mimicry-based configuration generation supplies the realism carrier: each fake router inherits the command structure, stanza ordering, and naming conventions of its most similar real router.","core_discovery":"The central proposal is that obscuring a network's scale is compatible with preserving its function. NetCloak takes the original router-level graph and embeds it in a larger graph drawn from a real-world reference topology: each original node is matched to a reference node of at least equal degree, the original edges are kept, and the remaining degree budget is filled with Havel-Hakimi-style edge addition and greedy rewiring so that the anonymized degree sequence closely tracks the reference. A new privacy definition, $k$-degree mapping anonymity, then says that an attacker who knows only the original degree sequence, not the anonymized one, cannot single out any original node: under the strong form, the $i$-th highest-degree original node has at least $k+i-1$ nodes of at least its degree in the output. Configuration generation mimics real routers and applies the same routing-policy filters to fake hosts, and a layered repair step—SMT constraints for intra-AS paths, iterative forwarding-table filtering between ASes—restores the original end-to-end routes. The paper claims this combination conceals network size while satisfying strong functional equivalence, so control-plane and verification tools see a plausible larger network with intact real behavior.","pith_inferences":["The scale-concealment guarantee is only as strong as the stated threat model: an adversary who exploits host subnets, VLANs, organizational charts, or traffic volumes is outside the model, and those signals would likely fingerprint real nodes; a natural extension is to anonymize those layers too.","The rationality metric only measures the degree distribution, the 1K property of the dK-series, so the embedding could in principle be upgraded to preserve joint-degree (2K) and triangle (3K) correlations; the paper notes this direction but does not implement it.","The choice of reference topology is a hidden parameter: when no reference is close to the target size, embedding overshoots the requested node count, so building synthetic reference topologies at arbitrary scales is a testable way to make the method fully flexible.","A direct way to test the privacy claim without relying on graph properties alone would be an end-to-end deanonymization game: given the anonymized configs, can a classifier tell real devices from injected ones using style, naming, and policy patterns?"],"forward_implications":["Organizations can publish anonymized configurations without revealing router count, so outsiders cannot infer organizational size or data-center capacity from topology degrees.","Real configurations can be expanded into larger datasets, giving verification, synthesis, and repair research more realistic inputs than purely synthetic ones.","Because strong functional equivalence is preserved, existing control-plane simulators and verifiers can be run on the anonymized files and will see the original forwarding behavior.","The $k$-DMA notion requires far fewer added edges than classical $k$-degree anonymity, so anonymized networks stay structurally plausible instead of accumulating telltale fake links.","The SMT-for-intra-AS, iterative-for-inter-AS split makes repair time stable under randomized link costs, where a purely iterative method can take many rounds."],"supporting_citations":[{"why":"Supplies the baseline edge-adding anonymizer, the functional-equivalence definitions, and the iterative repair method that NetCloak extends and compares against.","marker":"[46]"},{"why":"Supplies the SMT/CEGIS configuration-synthesis machinery and the template-synthesis baseline whose configuration style NetCloak's mimicry beats.","marker":"[14]"},{"why":"Supplies the SMT shortest-path constraint encoding that NetCloak reuses and corrects by distinguishing primary paths from ECMP paths.","marker":"[17]"},{"why":"Defines classical $k$-degree anonymity, the notion NetCloak replaces with $k$-degree mapping anonymity to avoid excessive edge edits.","marker":"[31]"},{"why":"Supplies the $k$-fold replication strawman for node addition, whose structural regularity NetCloak's embedding is designed to avoid.","marker":"[44]"},{"why":"Supplies the $k$-automorphism framework that motivates replication-based expansion and illustrates its detectability limits.","marker":"[52]"}],"fun_headline_variants":["Fake routers mask network size, real routes stay intact","NetCloak hides true scale with synthetic routers","Obfuscate network size without breaking routing","Synthetic nodes conceal your network's real footprint","Scale camouflage: add fake gear, keep routes true"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole privacy guarantee rests on the assumption that the adversary knows only the original network's degree sequence, and that no other observable signal—host counts, naming conventions, organizational data, traffic patterns, or configuration style—can separate the injected routers and hosts from real ones.","fun_headline_variants_meta":{"raw":{"variants":["Fake routers mask network size, real routes stay intact","NetCloak hides true scale with synthetic routers","Obfuscate network size without breaking routing","Synthetic nodes conceal your network's real footprint","Scale camouflage: add fake gear, keep routes true"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000193,"raw_usage":{"total_tokens":1403,"prompt_tokens":1049,"completion_tokens":354,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":665,"completion_tokens_details":{"reasoning_tokens":279}},"tokens_in":665,"tokens_out":354,"duration_ms":3531,"temperature":1.0,"reasoning_tokens":279,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:36:25.968365+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a deanonymization game: give an adversary the anonymized configuration plus a candidate original network with fewer routers, and ask it to decide whether the original router count equals the small candidate or the expanded count. The scale-concealment claim fails if the adversary succeeds by counting host subnets, matching organizational records, or exploiting stylistic fingerprints, rather than by using the degree sequence that $k$-DMA is designed to protect.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the SMT/CEGIS configuration-synthesis machinery and the template-synthesis baseline whose configuration style NetCloak's mimicry beats."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the $k$-fold replication strawman for node addition, whose structural regularity NetCloak's embedding is designed to avoid."}],"review_version":1}