{"id":"b6ab4252-fc48-4ddf-974b-8f587ae0b802","arxiv_id":"1909.00939","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"An empirical study of 995 Ethereum DApps and 29.8 million 2018 transactions shows heavy popularity concentration, low open-source rates, a majority of single-contract apps, and significant unused prepaid gas.","lead":"This paper studies the ecosystem of Ethereum-based decentralized applications, or DApps, using 995 apps and nearly 30 million transactions from 2018. It maps which apps are popular, how they are built, and what they cost to run, giving a data-backed starting point for users, developers, and blockchain vendors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Developer-supplied contract addresses may bias the 995-DApp sample; no validity check is reported for the 43% of DApps excluded.","rationale":"I agree with the reader that the dataset filter is the weakest link. The paper makes an explicit universe claim (Ethereum DApp market) but samples from a directory and then further restricts to DApps whose developers supply contract addresses. The missing validity check is the natural point of attack. I also noticed an internal inconsistency in the reported transaction volume: Section III says '9,057,344,360 Ethers' while Table I reports 9,057,344.360 ETH; this factor-of-1000 discrepancy is not explained and suggests the data-cleaning pipeline may contain unit or comma errors, reinforcing the need for a released, verified dataset. The reader's CONDITIONAL verdict is appropriate: the study is a plausible first look but not yet a robust baseline. My recommended verdict is UNCHANGED because the concern is exactly the one the reader already identified; the paper should add a validity section and ideally release the data before the findings are treated as definitive.","tokens_in":18008,"tokens_out":6609,"duration_ms":65391,"concrete_test":"Compute the category and first-published-date distributions for the 995 included and 754 excluded DApps from the State of the DApps listings, and test for significant differences (e.g., chi-square or Kolmogorov-Smirnov). If the excluded DApps are, for instance, significantly more likely to be in the High-risk or Gambling categories or to have been published after mid-2018, the Section III filter is selective; the aggregate statistics in Sections IV-VI would then need to be reported as conditional on the filtered sample, and claims about the 'Ethereum DApp market' revised. A null result on these observables would reduce, though not eliminate, the concern.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III of the paper states that from 1,749 Ethereum DApps listed on State of the DApps, the authors 'filter those DApps that developers do not provide the addresses of smart contracts for,' leaving 995 DApps. The abstract and conclusion use this 995-DApp set to present 'the first comprehensive empirical study' of the Ethereum DApp ecosystem, and every aggregate finding in Sections IV-VI is phrased as a property of that ecosystem. However, whether a developer supplies contract addresses is a self-selection that is plausibly correlated with the outcome variables studied: actively maintained, open-source, and popular DApps are more likely to publish addresses, while abandoned or deliberately opaque DApps are less likely to do so. The paper provides no comparison between the 995 included and the 754 excluded DApps on any available attribute, such as category, publication date, or presence of an external repository. Without such a check, it is possible that the reported 15.7% fully-open-source figure, the 75% single-contract figure, and the Pareto-shaped popularity distribution are artifacts of filtering rather than properties of the ecosystem. Because the central claim is a descriptive baseline, biased sampling directly undermines its validity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a descriptive empirical study of Ethereum-based decentralized applications (DApps). The authors collect 995 Ethereum DApps listed on State of the DApps as of January 2019, along with 5,158 associated smart contracts and 29,846,075 on-chain transactions from 2018. They analyze three research questions: the popularity distribution of DApps, development practices such as open-source status and smart-contract usage patterns, and the deployment and execution costs of smart contracts. The main reported findings are that DApp popularity follows a Pareto distribution with financial and gambling applications dominating; only 15.7% of DApps are fully open source; about 75% of DApps use a single smart contract; and typical contract executions leave large amounts of prepaid gas unused. The paper concludes with stakeholder-oriented implications for users, developers, and blockchain vendors.","tokens_in":18251,"tokens_out":5642,"duration_ms":60020,"significance":"If the methodology is sound, this is a useful descriptive baseline for the early Ethereum DApp ecosystem, and it is one of the first papers in software engineering to characterize DApps at this scale. The study's strengths include direct measurement of on-chain data, a broad coverage of categories, and practically actionable observations such as the gas-cost distributions. However, the absence of a released dataset, the unvalidated sampling filter, and the questionable regression interpretation currently limit reproducibility and the strength of the causal/implicational claims. The paper is likely to be of interest to the empirical software engineering and blockchain communities, provided the identified validity concerns are addressed.","major_comments":[{"comment":"The study starts from 1,749 Ethereum DApps on State of the DApps and keeps only the 995 DApps for which developers supplied smart-contract addresses, but it then presents all aggregate results as properties of the Ethereum DApp ecosystem. The paper provides no validity check comparing the 995 included DApps with the 754 excluded ones on any available attribute, such as category, publication date, or presence of an external repository. Because supplying an address is a self-selected developer action that is plausibly correlated with activity, open-source behavior, and maintenance quality, the reported figures (open-source percentage, single-contract percentage, Pareto popularity distribution) could be artifacts of this filter. Please add a comparison of included and excluded DApps and a discussion of the potential selection bias, or explicitly reframe the conclusions as applying only to DApps with published contract addresses.","section":"Section III"},{"comment":"The conclusion that NoF (number of functions) has more influence than LoC (lines of code) on deployment cost is based on raw regression coefficients with different units: 37,141.26 gas per function versus 1,514.68 gas per line of code. Such unstandardized coefficients are not directly comparable, especially because the univariate model with LoC alone achieves a higher R-squared (0.4066) than the model with NoF alone (0.2956). To support Finding F7 and the implication that 'reducing the number of functions could help more,' the paper should report standardized coefficients (beta weights), common-scale effect sizes, or an explicit model-comparison test, and should also check for multicollinearity between NoF and LoC.","section":"Section VI-A"},{"comment":"The paper uses the association in Figure 8 to advise developers that open-sourcing smart contracts 'could improve the popularity of DApps.' This is a causal/implicational claim drawn from a raw correlation, and open-source status is likely confounded with DApp category, age, and development effort. The current evidence does not support the recommendation as stated. Please add a controlled or matched comparison (for example, within category and publication cohort) or soften the implication to an associational statement.","section":"Section V-A"}],"minor_comments":[{"comment":"The text says '9,057,344,360 Ethers' while Table I reports '9,057,344.360 ETH.' These are inconsistent by a factor of 1,000; please reconcile the value and the notation.","section":"Section III"},{"comment":"The sentence 'At the level of t¡0.01' appears to be a typo and should read 'p < 0.01.'","section":"Section VI-A"},{"comment":"The regression analysis does not state the sample size used for fitting the models, the number of contracts with source code and ABI data, or any residual diagnostics. Since only 2,568 of the 5,158 contracts have retrievable source code, please specify the effective sample size and whether the reported regressions are restricted to that subset.","section":"Section VI-A"},{"comment":"The agas metric weights per-function median gas by invocation count, but for functions with very few invocations the median is an unstable estimate. Please report the distribution of invocation counts and consider requiring a minimum invocation count before including a function in Eq. (1), or provide confidence intervals for the pattern comparisons in Figure 15.","section":"Section VI-B"},{"comment":"The x-axes of these figures appear to mix a linear origin (0) with logarithmic powers (10^1, 10^2, ...). Please use a consistent axis scale or explain how zero is represented on a logarithmic axis.","section":"Figures 3b, 4b, 5b"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for an empirical software engineering venue, but the authors should be aware that the 'first comprehensive study' claim may be contested by prior or concurrent empirical blockchain work; a broader related-work search is advisable. I would also encourage the authors to release the dataset and filtering/crawling scripts, since this is a measurement paper and the lack of data availability is a significant reproducibility barrier for the community."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it delivers what it promises: a broad, descriptive baseline of Ethereum DApps, built from 995 DApps and 29.8 million transactions. That is genuinely new and useful for the blockchain software engineering community. Second, the central weak spot is exactly where the stress-test note lands: the authors filter from 1,749 to 995 DApps by keeping only those with developer-supplied contract addresses, and they provide no validity check on the excluded 754. If missing addresses correlates with abandonment, opacity, or unpopularity—which it plausibly does—the aggregate findings like the 15.7% fully-open-source figure and the Pareto curves could be biased. The paper does not even compare the two groups on category or publication date, so the reader cannot tell how much the filter matters.\n\nWhat the paper does well: the dataset is large, the descriptive statistics are direct and mostly transparent, and the authors separate deployment cost from execution cost, which is a sensible distinction. The identification of the leader-member, equivalent, and factory smart-contract patterns is a reasonable qualitative contribution. The gas-left analysis, showing that users typically pre-pay far more gas than is used, is a concrete, actionable observation. The paper also earns credit for grounding its work in a well-known directory and using Etherscan as a cross-check for simulated executions.\n\nThe soft spots, in proportion: the sampling-bias issue is real and is the most important one. The regression comparison between NoF and LoC is a second genuine problem: the authors compare unstandardized coefficients with different units, so the claim that NoF is more influential than LoC is not supported. That claim should be softened or re-analyzed. Third, the paper sometimes drifts from correlation to causation, particularly when suggesting open source may improve popularity. That is a minor overreach, not a fatal one. Fourth, the dataset is not released, which hurts reproducibility in a field where others could to re-run the analysis. The gas-threshold recommendation of sending 141,213 gas is derived from the same data it is validated on, which is mild self-reference; the recommendation is fine as a summary statistic but not as an independent prescription.\n\nWho benefits: readers working on DApp engineering, blockchain platform design, or smart contract cost modeling will get value from this as a baseline. It deserves a serious referee, but the referee should push for a validity check on the filter, a corrected regression interpretation, and ideally a released dataset.\n\nMy recommendation: engage with it, but treat the specific percentages and the NoF-vs-LoC conclusion as provisional until the filter bias is addressed.","headline":"A useful descriptive baseline of the early Ethereum DApp ecosystem, but the unvalidated sample filter and the regression overclaim need attention before the findings are treated as robust.","tokens_in":18734,"tokens_out":1047,"would_cite":true,"duration_ms":12726,"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":"This paper presents the first comprehensive empirical study of Ethereum DApps, analyzing 995 DApps and 29,846,075 transactions from 2018, and derives quantitative patterns for popularity, development practices, and running costs.","keywords":["decentralized applications","Ethereum","smart contracts","empirical study","transaction analysis","gas cost","open source","DApp ecosystem"],"falsifier":"Repeat the same 2018 transaction analysis for the Ethereum DApps listed in the same directory but excluded because they lacked published contract addresses; if their popularity distribution, category mix, or gas usage differs systematically from the 995 DApps, the paper's aggregate findings would not generalize to the directory as a whole.","tokens_in":17834,"feed_emoji":"📊","tokens_out":7663,"duration_ms":67181,"temperature":0.7,"pith_summary":"This paper presents the first comprehensive empirical study of blockchain-based decentralized applications, using 995 Ethereum DApps and 29,846,075 transactions from 2018. It finds that popularity follows a Pareto distribution: a small group of finance, exchange, and gambling DApps accounts for most users, transactions, and volume, while most DApps are barely used. It also reports that only 15.7% of DApps are fully open source, about 75% rely on a single smart contract, and in the median contract execution half of the prepaid gas is left unused. These findings give users, developers, and platform vendors an evidence-based picture of how the early Ethereum DApp ecosystem actually behaves.","feed_headline":"Ethereum's DApp economy mapped: 995 apps, 29.8M transactions","feed_subtitle":"Popularity concentrates in finance and gambling; most DApps run on one contract and leave prepaid gas unused","key_machinery":"The central object is the assembled dataset: 995 DApps with developer-supplied smart-contract addresses, their 5,158 contracts, and all 29,846,075 transactions initiated in 2018. The analytic machinery consists of three pieces: popularity metrics (unique users, transactions, and transaction volume), a classification of multi-contract usage patterns (leader-member, where an entry contract invokes member contracts; equivalent, where contracts are independent; and factory, where one contract deploys similar child contracts), and a gas-accounting analysis that separates deployment from execution, compares gas sent with gas used, and regresses deployment cost on lines of code and number of functions. The dataset carries the empirical claims, while the patterns and gas metrics supply the structure for the findings.","core_discovery":"The paper's central claim is that the early Ethereum DApp ecosystem is not a uniform market but follows strong empirical regularities. Popularity is highly concentrated: fewer than 5% of DApps carry about 80% of transactions, and most DApps have fewer than 1,000 users and fewer than 10,000 transactions over the year. Financially oriented categories dominate, with Exchanges and Finance together accounting for roughly 87% of transaction volume. On the development side, only 15.7% of DApps are fully open source, 52.3% of DApps have all smart contracts closed, and about 75% of DApps use a single smart contract; among multi-contract DApps the equivalent pattern is the most common. On cost, deployment gas correlates with both lines of code and number of functions, with number of functions the stronger driver, and in the median contract execution half of the prepaid gas is unused, leaving about 100,000 gas locked until confirmation.","pith_inferences":["The filter that keeps only DApps with published contract addresses could bias every aggregate finding; if the excluded DApps are mostly small or centralized, the true market may look even more concentrated, and the open-source percentages could be worse.","Because the study counts only on-chain transactions, DApps whose clients interact with contracts through centralized back ends are undercounted; the popularity rankings measure blockchain-visible usage, not necessarily total user activity.","The gas-left result points to a testable extension: a wallet that sets gas near the 80th-percentile value and measures confirmation refunds would show whether most locked gas can be returned without increasing failed transactions."],"forward_implications":["Users can treat 141,213 gas as a practical ceiling for ordinary contract executions, since 80% of the measured executions used no more than that, and sending more simply locks up the surplus Ether until confirmation.","Developers can lower execution costs by favoring the equivalent pattern, where contracts do not invoke one another, over leader-member or factory arrangements.","Open-sourcing smart contracts is associated with higher transaction counts, giving developers a concrete reason to publish contract source even when the full project remains closed.","Platform vendors can use the measured gas-left distribution to build better default gas limits and client-side gas estimators that return locked Ether to users sooner.","The rapid growth of High-risk and Gambling DApps after early 2018 suggests users need screening or warning mechanisms when choosing DApps."],"supporting_citations":[{"why":"Third-party DApp directory that supplies the population, categories, and publication dates for the dataset.","marker":"[28]"},{"why":"Defines the Ethereum project and smart-contract mechanism that all measurements target.","marker":"[29]"},{"why":"Block explorer that provides contract source code and interfaces, plus an independent check on gas and transaction data.","marker":"[30]"},{"why":"Documents Ponzi schemes on Ethereum and supports the paper's interpretation of the High-risk category.","marker":"[25]"}],"fun_headline_variants":["Most Ethereum DApps are tiny: 5% handle 80% of transactions","DApp study: 995 apps, 29.8M logs, extreme popularity skew","Ethereum DApps: 52% closed-source, 75% single-contract","Blockchain DApps: finance dominates, gas waste rampant","First DApp census: 995 apps, huge inequality, unused gas"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The findings stand or fall on whether the 995 DApps whose developers published smart-contract addresses represent all Ethereum DApps; if the omitted apps differ systematically, every aggregate percentage would shift.","fun_headline_variants_meta":{"raw":{"variants":["Most Ethereum DApps are tiny: 5% handle 80% of transactions","DApp study: 995 apps, 29.8M logs, extreme popularity skew","Ethereum DApps: 52% closed-source, 75% single-contract","Blockchain DApps: finance dominates, gas waste rampant","First DApp census: 995 apps, huge inequality, unused gas"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000839,"raw_usage":{"total_tokens":3643,"prompt_tokens":917,"completion_tokens":2726,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":533,"completion_tokens_details":{"reasoning_tokens":2623}},"tokens_in":533,"tokens_out":2726,"duration_ms":17972,"temperature":1.0,"reasoning_tokens":2623,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:31:14.290438+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the same 2018 transaction analysis for the Ethereum DApps listed in the same directory but excluded because they lacked published contract addresses; if their popularity distribution, category mix, or gas usage differs systematically from the 995 DApps, the paper's aggregate findings would not generalize to the directory as a whole.","supporting_citations":[{"cited_title":"State of the DApps,","cited_arxiv_id":null,"evidence_quote":"Third-party DApp directory that supplies the population, categories, and publication dates for the dataset."},{"cited_title":"Ethereum Project,","cited_arxiv_id":null,"evidence_quote":"Defines the Ethereum project and smart-contract mechanism that all measurements target."},{"cited_title":"Ethereum (ETH) Blockchain Explorer,","cited_arxiv_id":null,"evidence_quote":"Block explorer that provides contract source code and interfaces, plus an independent check on gas and transaction data."},{"cited_title":"Detecting ponzi schemes on ethereum: Towards healthier blockchain technology,","cited_arxiv_id":null,"evidence_quote":"Documents Ponzi schemes on Ethereum and supports the paper's interpretation of the High-risk category."}],"review_version":1}