{"id":"1a015964-cc2d-4d1b-94ef-5fb979cd4692","arxiv_id":"2507.20234","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"SNS DAOs on the Internet Computer maintained or increased voter participation and approval over 20 months, averaging 64% participation and 96.8% approval across 14 DAOs.","lead":"This paper measures how 14 DAOs on the Internet Computer blockchain handle governance, tracing more than 3,000 proposals over 20 months. It finds these DAOs keep high participation and approval over time, unlike many Ethereum-based DAOs that decline.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Participation metric counts delegated voting power, not human engagement; the 'sustained engagement' claim and cross-DAO comparisons are not yet established.","rationale":"The reader's weakest_assumption correctly identifies the metric. I agree; this is the single most load-bearing concern because the paper's novelty is the empirical observation of sustained participation, and that observation is made with a proxy that conflates automated delegation with active engagement. The paper itself provides the necessary data on-chain, so the test is feasible. Alternative concerns—survivorship bias in selecting 14 of 29 SNS DAOs, the unusual cost metric ($11 per proposal derived by dividing all SNS canister costs by executed proposals in one month), and lack of error bars—are real but secondary: even if addressed, the headline claim would still fail if participation is mostly automatic and concentrated. The most direct fix is to report distinct-neuron participation and delegation shares, or to narrow the claim to 'voting-power participation.' The reader's CONDITIONAL verdict remains appropriate; the paper should be accepted only if the authors either supply these supplementary metrics or explicitly reframe the conclusion. No change to the reader's verdict is needed.","tokens_in":12814,"tokens_out":4737,"duration_ms":53096,"concrete_test":"Re-query the governance canisters for all 14 SNS DAOs and, for each proposal, extract the set of distinct neuron IDs that voted (directly or through following) and the voting power of the top 10 neurons. Recompute monthly participation as (a) distinct neurons that voted / total neurons, and (b) voting power not cast through automatic follow votes / total voting power. If (a) is below 20% or declines over time, or if (b) falls while the headline 64% remains flat, the 'sustained engagement' claim and the cross-DAO engagement comparison are not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that SNS DAOs 'exhibit sustained or increasing engagement levels over time' (abstract; §V) rests entirely on the participation metric defined in §III.A as 'the fraction of voting power engaged relative to the total voting power registered.' In the SNS model (§II.A), a neuron can follow other neurons per topic and its voting power is cast automatically when a majority of followees agree; otherwise it abstains. Therefore high and stable 'participation' can be achieved by few active neurons followed by many passive token holders, with no growth in human deliberation. The paper even acknowledges concentration qualitatively in §III.C ('dominance by a few neurons with high voting power') but never reports the fraction of votes cast via following, the distribution of voting power across neurons, or the number of distinct neurons whose voting power actually moved. §IV then compares the resulting ~64% to Ethereum DAO figures that use different denominators (e.g., Barbereau et al. report <1% of token-holders; Messias et al. report votes per delegated tokens), so the claimed superiority in 'community engagement' may be an artifact of metric choice and automatic delegation rather than broader participation. This is the load-bearing weak point because the abstract's most important sentence, the time-trend comparison, and the 'higher activity... faster decisions' conclusion all depend on it.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents an observational study of 14 SNS DAOs on the Internet Computer, based on over 3,000 proposals collected from launch until September 2024. The authors define participation, approval, rejection, and decision-duration metrics, report aggregate averages (participation about 64%, approval about 96.8%, decision duration about 1.14 days), and compare these with prior studies of Ethereum-based DAOs. The central claim, stated in the abstract and conclusion, is that SNS DAOs exhibit sustained or increasing engagement over time, in contrast to the participation decline reported for other DAO frameworks.","tokens_in":13047,"tokens_out":5780,"duration_ms":73666,"significance":"If the findings hold, the paper provides a practically important and policy-relevant counterexample: a low-cost, reward-backed, delegate-able voting framework that maintains high voting-power participation. The study's strengths include a substantial on-chain dataset (over 3,000 proposals across 14 DAOs), clear metric definitions in Section III.A, and a useful comparison to prior DAO governance studies. The on-chain data collection is, in principle, reproducible. However, the current evidence does not establish the headline claims: the participation metric conflates delegated voting power with human engagement, the time-trend claim rests on visual inspection without statistical tests, and the cross-ecosystem comparisons use different denominators. These issues are load-bearing because they directly affect the abstract's main sentence, the comparison in Section IV, and the conclusion.","major_comments":[{"comment":"The participation metric is defined as the fraction of voting power engaged relative to total registered voting power. In the SNS model described in Section II.A, a neuron votes automatically when a majority of its followees agree, and otherwise abstains. Therefore, high and stable participation can be produced by a small number of active neurons followed by many passive token holders, without broad human deliberation. The paper acknowledges concentration only qualitatively in Section III.C ('dominance by a few neurons with high voting power'), but it never reports the fraction of votes cast via following, the distribution of voting power across neurons, or the number of distinct neurons whose voting power actually moved. This matters because the abstract and conclusion repeatedly interpret the participation rate as 'community engagement.' Please either refine the terminology and claims, or provide additional data on following behavior and voting-power concentration that would let the reader assess whether the rate reflects broad engagement.","section":"Section III.A and II.A"},{"comment":"The claim that 'participation rates show an overall increase' and that 'as of September 2024, all of them are higher than at launch' is based on visual inspection of monthly average participation rates. The paper reports no confidence intervals, no trend test, and no correction for multiple DAO comparisons, so the central dynamic claim is not statistically established. In addition, Section III.F.1 says 'all of them are higher than at launch,' while Section IV says 'for most cases the participation rates are higher than the initial months,' an internal inconsistency. Please provide per-DAO trend statistics (e.g., a linear or monotonic trend test with standard errors), show all DAOs rather than a subset, and reconcile the 'all' versus 'most' statements.","section":"Section III.F.1 and Figure 6"},{"comment":"The headline aggregate values (participation 64.34%, approval 96.8%, decision duration 1.14 days) are not explicitly defined as unweighted DAO averages or proposal-weighted aggregates. Because proposal counts vary widely across DAOs (OpenChat alone has 966 proposals, while others have far fewer, and Table I reports no proposal count per DAO), the aggregation choice can materially change these numbers. Please define the aggregation rule explicitly and report per-DAO and per-proposal distributions, including variances or standard errors.","section":"Sections III.B, III.C, and III.E"},{"comment":"The comparison of SNS participation (about 64%) with Ethereum DAO participation figures is not apples-to-apples. The SNS rate is voting power relative to total registered voting power, whereas the cited studies use different denominators: Barbereau et al. report exercised voting rights relative to token-holders, Feichtinger et al. report percentages with yet another basis, and Messias et al. use votes or delegated tokens relative to total delegated tokens. These differences in denominator are part of what drives the apparent gap. Please either recompute comparable metrics from the raw data or clearly state the metric differences and discuss how they affect the cross-platform conclusion.","section":"Section IV"},{"comment":"The cost comparison is computed on different accounting bases: the SNS cost of 'around 11 USD' is the total cost of all SNS canister operations (including ledger transfers and upgrades) divided by the number of executed proposals, whereas the cited Ethereum costs are per-proposal or per-vote gas costs for governance transactions. This makes the 'lower costs' claim difficult to evaluate. Please provide a defined cost model with the same scope (e.g., proposal submission plus voting) and report costs per proposal and per vote for both ecosystems.","section":"Section IV, governance cost comparison"}],"minor_comments":[{"comment":"There are several typos and infelicities: 'adn' (Section III.F.2), 'ICSwap' (Section III.D), 'suggestions suggestions' (Section III.D), 'government' instead of 'governance' (Section III.F.2), and 'see 6 for a plot' missing the word 'Figure' (Section III.F.1).","section":"Throughout"},{"comment":"The x-axis labels with DAO names are very small and likely unreadable in print; consider rotating labels, using larger fonts, or abbreviating names consistently.","section":"Figures 1-4"},{"comment":"Table I lists age, treasury, and neuron count but not the number of proposals per DAO; adding a proposal count column would help interpret the weighted versus unweighted aggregation issue and the activity comparisons.","section":"Table I"},{"comment":"The 'harmless' proposal category is used in Figures 1-4 before its composition is explained; state explicitly in Section III.A or III.D that harmless proposals are primarily Motion and SNS-Upgrade proposals.","section":"Section III.D"},{"comment":"The sentence 'The oldest SNS DAO investigated, OpenChat and Yral, shows moderate rate of participation' is grammatically ambiguous; OpenChat and Yral are two separate DAOs and should be presented as such.","section":"Section III.B"},{"comment":"No data or code availability statement is provided; since the study is empirical, sharing the query scripts and aggregated datasets would strengthen reproducibility and transparency.","section":"Reproducibility"},{"comment":"One co-author is affiliated with DFINITY, and reference [15] includes a co-author of this paper; the manuscript would benefit from an explicit conflict-of-interest or self-citation disclosure.","section":"Conflicts of interest"}],"recommendation":"major_revision","confidential_remarks":"The paper is a useful empirical study, but the current version requires substantial revision to support its headline claims. The participation-metric concern raised in the stress-test is real: with automatic following, the metric can reflect a small set of active voters rather than broad engagement, and the paper's own qualitative acknowledgement in Section III.C does not resolve it. The dynamic claim also needs formal trend testing and a reconciliation of the 'all' versus 'most' statements. The lack of a conflict-of-interest statement is notable given the DFINITY affiliation and the self-citation to [15]; this should be disclosed to the editor. None of these issues appear irreparable, so major revision is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a useful descriptive dataset, not a proof that SNS DAOs are more democratic. The novelty is real — it is the first systematic governance metrics study of the SNS ecosystem, covering 3,000+ proposals from 14 DAOs over 20 months. The raw observations (64% average participation, 96.8% approval, ~1.1 day decisions) are worth having, and the paper defines its metrics clearly, pulling directly from on-chain governance canisters. The descriptive statistics are probably accurate as measurements of voting-power participation.\n\nThe load-bearing problem is the participation metric. It counts voting power that votes, not people. SNS neurons can follow other neurons per topic, and votes are cast automatically when a majority of followees agree, so a few active neurons can generate high 'participation' without broad human engagement. The paper acknowledges the possibility of dominance by a few neurons in §III.C but never quantifies following, voting-power concentration, or distinct voters. So the abstract's headline — 'sustained or increasing engagement' — is not established as human engagement; it is established as voting-power participation, which is weaker. The comparison to Ethereum DAOs also mixes denominators: Barbereau et al. report <1% of token-holders, Messias et al. use delegated tokens, while this paper uses registered voting power. That is an apples-to-oranges comparison, and the claimed 'greater voter engagement' should be softened or redefined.\n\nOther soft spots, in proportion: no error bars, confidence intervals, or trend tests — the 'increasing participation' claim rests on eyeballing monthly averages in Figure 6. Survivorship bias is real: only DAOs that survived 6+ months are included, which could inflate the sustained-engagement narrative. No artifact release, which hurts reproducibility. Self-citation of [15] (including co-author Schmid) is minor and not improper.\n\nWhat the paper does well beyond the dataset: it separates proposal types, shows that rejection trends concentrate in Motion and treasury-transfer proposals, and makes a genuine effort to engage prior work even where the metrics do not align.\n\nThis is for DAO governance researchers interested in ICP/SNS or in cross-platform comparisons. With revision — redefining or supplementing the participation metric, adding statistical tests, and releasing data — it could be a solid contribution. As is, it is a good descriptive report with an over-extended conclusion.\n\nI would send it out: the empirical core is novel and likely correct, and the issues are fixable in revision. I would ask for a major revision that fixes the engagement claim before publication.","headline":"Useful first SNS governance dataset, but the 'sustained engagement' claim overreaches what a voting-power metric can support.","tokens_in":13574,"tokens_out":2815,"would_cite":true,"duration_ms":29105,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"SNS DAOs on the Internet Computer sustain participation where other DAO platforms decline.","keywords":["DAO governance","Internet Computer","SNS","participation rates","liquid democracy","voting delegation","on-chain governance","empirical study"],"falsifier":"Recompute participation using distinct principals or human-linked accounts instead of voting power; if the number of distinct voters declines while voting-power participation stays high, the sustained-engagement claim fails. Also test whether the top 1% of neurons control a majority of voting power; if so, the 64% figure may reflect automated votes from a few large holders.","tokens_in":12624,"feed_emoji":"🗳️","tokens_out":6739,"duration_ms":68419,"temperature":0.7,"pith_summary":"This paper asks whether a particular DAO governance design can avoid the voter fatigue and participation decay observed in other blockchain communities. Studying 14 Service Nervous System (SNS) DAOs on the Internet Computer over 20 months and more than 3,000 proposals, it reports an average participation rate of about 64% of voting power, an average approval rate of 96.8%, and an average decision time of about 1.14 days. Its headline finding is that, unlike Ethereum-based DAOs such as Compound, Uniswap, and ENS, these SNS DAOs show sustained or increasing engagement over time rather than decline. The authors attribute this to the SNS design: free voting under a reverse gas model, token-locked neurons with rewards, and per-topic vote delegation. If the result holds, it offers a concrete empirical counterexample to the pattern of decaying participation in DAO governance and a possible template for other ecosystems.","feed_headline":"SNS DAOs keep voters engaged as other DAOs fade","feed_subtitle":"A 20-month look at 3,000 proposals finds SNS DAOs hold ~64% participation while other DAO platforms decline.","key_machinery":"The central object is the Service Nervous System (SNS), the Internet Computer's DAO framework: a shared governance canister codebase that each community parameterizes. The load-bearing mechanisms are the reverse gas model (developers prepay computation so voting costs users nothing), neuron-based voting power that grows with token lockup and age, per-topic delegation ('following') where a neuron's vote is cast automatically when a majority of its chosen followees agree, and token rewards proportional to voting participation. These mechanisms together are what the paper credits for high, sustained participation, near-universal approval, and fast decisions; it uses them to explain the contrast with gas-fee-burdened, delegation-poor Ethereum DAOs.","core_discovery":"The central discovery is that the Internet Computer's SNS governance framework produces persistently high community engagement, whereas other DAO platforms studied in the literature show participation decay. Across 14 SNS DAOs ranging from DeFi to gaming and meme coins, the authors measure an average participation rate of 64.34% of total voting power, an average approval rate of 96.8%, and an average decision-making duration of 1.14 days, with no monotonic decline over time. The paper argues this stems from mechanisms unique to SNS: a reverse gas model that makes voting free for users, rewards distributed in native tokens proportional to voting power, topic-specific delegation ('following') that lets neurons delegate to experts, and low proposal costs around $11 compared to thousands on Ethereum. The authors interpret these results as evidence that SNS-style governance—fully on-chain, low-cost, delegatable, and reward-backed—can sustain democratic involvement and serve as a replicable model for other blockchain ecosystems.","pith_inferences":["The participation metric—share of voting power that votes—can be inflated by delegation and by a few large neurons; testing participation by distinct neuron owners or human wallets could reveal that the 'sustained engagement' is largely automated voting, a possibility the paper acknowledges only qualitatively.","The same incentive structure that sustains participation (rewards proportional to voting power, voting power growing with lockup) may entrench early large holders, so the fast, high-approval governance could trade breadth for oligopoly; a power-law analysis of neuron sizes would test this.","The SNS finding suggests a testable design rule for other chains: eliminating per-vote transaction fees and adding per-topic delegation should raise participation; a natural experiment would be a fork or layer-2 DAO adopting the reverse gas model and measuring the participation slope over time.","The lower participation on critical proposals suggests that requiring more effort for high-stakes votes may backfire; a design that lowers the effort for critical votes, such as better delegation defaults, could raise engagement on the decisions that matter most."],"forward_implications":["If the central claim is correct, SNS-style low-cost, reward-backed, delegatable voting is a workable antidote to voter fatigue in DAOs, sustaining engagement where fee-based on-chain voting declines.","The combination of fast decisions (1.14 days on average) and high participation suggests that cost and delegation design, not community size or proposal volume, are the main levers of governance agility.","The near-universal approval rates (96.8%) and low rejection of non-critical proposals indicate high alignment or potentially low scrutiny; the paper itself flags the need to study whether a few high-power neurons dominate small DAOs.","The proposal-activity comparison (OpenChat at about 48 proposals per month versus Uniswap at 1.7) implies SNS communities can sustain continuous development governance without overwhelming voters.","The paper's cost figures (roughly $11 per proposal versus $594 to $20,000 on Ethereum) imply that governance cost is a first-order determinant of participation, a direct design lesson for other ecosystems."],"supporting_citations":[{"why":"Supplies the Ethereum DAO participation rates (Compound 34%, Uniswap 31.4%, ENS 39.2%, Gitcoin 28.6%) and proposal-cost estimates that SNS results are benchmarked against.","marker":"[6]"},{"why":"Provides Compound and Uniswap participation, approval, duration, and proposal-frequency baselines that SNS DAOs are compared with.","marker":"[11]"},{"why":"Establishes Aragon, DAOstack, and DAOhaus voter-participation baselines that frame SNS's relative engagement.","marker":"[5]"},{"why":"Documents low and declining token-holder voting in Ethereum DeFi DAOs, the participation-decay trend the paper claims SNS DAOs avoid.","marker":"[3]"},{"why":"Contributes Snapshot DAO decision-duration and voting statistics used as the time-cost comparison for SNS decisions.","marker":"[20]"},{"why":"Supports the incentive claim that token-based compensation scales governance effort and engagement, justifying SNS reward design.","marker":"[13]"},{"why":"Argues flat token-weighted voting is insufficient and recurring, role-sensitive compensation works, cited in support of SNS rewards.","marker":"[4]"},{"why":"Supports the claim that voting costs and low expected utility discourage participation, which the reverse gas model and following are said to fix.","marker":"[12]"},{"why":"Analyzes liquid democracy economics on the Internet Computer, grounding the delegation and reward discussion.","marker":"[10]"},{"why":"Describes the SNS as fully on-chain DAO infrastructure on the Internet Computer, the framework under study.","marker":"[1]"}],"fun_headline_variants":["SNS DAOs keep 64% voting power engaged while others fade","SNS DAOs: 64% participation, 1.14-day decisions, no engagement decay","Why SNS DAOs sustain voter interest while other platforms decline","SNS DAOs show sustained engagement: 64% turnout, 96.8% approval","Voters stay: SNS DAOs average 64% participation and 1-day decisions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper counts a vote as participation when voting power is cast, so automated votes from delegated neurons and a few large holders can keep participation high without broad human engagement.","fun_headline_variants_meta":{"raw":{"variants":["SNS DAOs keep 64% voting power engaged while others fade","SNS DAOs: 64% participation, 1.14-day decisions, no engagement decay","Why SNS DAOs sustain voter interest while other platforms decline","SNS DAOs show sustained engagement: 64% turnout, 96.8% approval","Voters stay: SNS DAOs average 64% participation and 1-day decisions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000983,"raw_usage":{"total_tokens":4187,"prompt_tokens":979,"completion_tokens":3208,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":3113}},"tokens_in":595,"tokens_out":3208,"duration_ms":23452,"temperature":1.0,"reasoning_tokens":3113,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:40:57.665901+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute participation using distinct principals or human-linked accounts instead of voting power; if the number of distinct voters declines while voting-power participation stays high, the sustained-engagement claim fails. Also test whether the top 1% of neurons control a majority of voting power; if so, the 64% figure may reflect automated votes from a few large holders.","supporting_citations":[{"cited_title":"The hidden shortcomings of (d) aos–an empirical study of on-chain governance","cited_arxiv_id":null,"evidence_quote":"Supplies the Ethereum DAO participation rates (Compound 34%, Uniswap 31.4%, ENS 39.2%, Gitcoin 28.6%) and proposal-cost estimates that SNS results are benchmarked against."},{"cited_title":"Faqir-Rhazoui, J","cited_arxiv_id":null,"evidence_quote":"Establishes Aragon, DAOstack, and DAOhaus voter-participation baselines that frame SNS's relative engagement."},{"cited_title":"Decentralised finance’s unregulated governance: Minority rule in the digital wild west","cited_arxiv_id":null,"evidence_quote":"Documents low and declining token-holder voting in Ethereum DeFi DAOs, the participation-decay trend the paper claims SNS DAOs avoid."},{"cited_title":"An Empirical Study on Snapshot DAOs","cited_arxiv_id":"2211.15993","evidence_quote":"Contributes Snapshot DAO decision-duration and voting statistics used as the time-cost comparison for SNS decisions."},{"cited_title":"Incentives in decen- tralised autonomous organisations","cited_arxiv_id":null,"evidence_quote":"Supports the incentive claim that token-based compensation scales governance effort and engagement, justifying SNS reward design."},{"cited_title":"Compensation in DAOs: A Proposal","cited_arxiv_id":null,"evidence_quote":"Argues flat token-weighted voting is insufficient and recurring, role-sensitive compensation works, cited in support of SNS rewards."},{"cited_title":"Incentive compat- ibility in consensus protocols and daos: A game-theoretic approach","cited_arxiv_id":null,"evidence_quote":"Supports the claim that voting costs and low expected utility discourage participation, which the reverse gas model and following are said to fix."},{"cited_title":"Liu and L","cited_arxiv_id":null,"evidence_quote":"Analyzes liquid democracy economics on the Internet Computer, grounding the delegation and reward discussion."},{"cited_title":"Fully on-chain daos on the internet computer","cited_arxiv_id":null,"evidence_quote":"Describes the SNS as fully on-chain DAO infrastructure on the Internet Computer, the framework under study."}],"review_version":1}