Gov-Acc Pipeline.

Monitoring Limits in DAO Governance: Capacity Breakpoints and Endogenous Concentration ∗ Guy Tchuente March 13, 2026 Abstract Decentralized autonomous organizations (DAOs) are designed to disperse control, yet recent evidence shows that effective governance is often concentrated in a small number of participants. This note studies one simple mechanism behind that pattern. Because decentralized governance is monitor-intensive, rising proposal flow may even- tually outpace the capacity of broad-based participation. Using a DAO–quarter panel, I estimate a fixed-effects kink model with DAO and quarter fixed effects and find a statistically significant decline in the marginal responsiveness of active voters once pro- posal activity crosses an interior threshold. I then study realized voting concentration using kink specifications with data-driven cutoffs. Across specifications, decentraliza- tion gains do not persist indefinitely once governance workload becomes sufficiently high, and load-based measures show especially clear evidence of a transition toward more concentrated realized control. The results provide reduced-form evidence consis- tent with a “too big to monitor” mechanism in DAO governance: when proposal flow grows faster than broad participation can keep up, effective control may drift toward a smaller set of highly active participants. Keywords: DAOs; decentralized governance; monitoring; capacity constraints. JEL Codes: D71, D83, G34, C23, C24. ∗Purdue University. Email: gtchuent@purdue.edu. 1 6202 raM 11 ]NG.noce[ 1v22211.3062:viXra 1 Introduction Decentralized autonomous organizations (DAOs) are often presented as a governance tech- nology capable of dispersing authority through member voting rather than centralized man- agerial control. A growing literature documents both the rapid expansion of DAO activity and the institutional diversity of on-chain governance, including variation in voting rules, delegation mechanisms, quorum requirements, and implementation architectures (Han et al., 2025; Nassif and Savva, 2024; Bellavitis et al., 2023; Sharma et al., 2024). Yet decen- tralization is not costless. For collective decision-making to remain genuinely broad-based, participants must monitor governance activity: they must read proposals, assess tradeoffs, follow discussion, and cast informed votes. As proposal flow rises, this monitoring task may become increasingly burdensome, creating scope for participation fatigue, reliance on del- egates, and the effective concentration of influence even when formal voting rights remain widely distributed (Ammons and Makridis, 2025; Zhang, 2025; Bongaerts et al., 2025). This note is part of a broader research program on the limits of decentralized monitoring. In earlier work, I develop and test a “too big to monitor” framework showing that decen- tralized monitoring systems may function well at modest scale but can break down once the volume of entities or decisions to be monitored exceeds the capacity of dispersed monitors (Tchuente, 2025, 2026). The first papers in that series study more traditional regulatory and service-delivery environments. The present note extends the same logic to DAOs, a setting that is institutionally very different and therefore especially revealing. In DAOs, monitoring is voluntary, governance is token-mediated, participation may be pseudonymous, and au- thority is designed to be decentralized from the outset. If a too-big-to-monitor mechanism arises even in this environment, it suggests that capacity limits are a more general constraint on decentralized governance than the earlier applications alone might imply. The note also speaks directly to recent evidence on concentration in DAO governance. Appel and Grennan (2023) show that many DAOs are effectively controlled by a very small number of entities, while Appel and Grennan (2026) emphasize that DAO governance out- comes depend critically on institutional design rather than on decentralization alone. This paper provides a simple empirical micro-foundation for those findings. The core idea is that scale in a DAO should be understood not only as the number of proposals submitted, but as the monitoring burden those proposals impose on the voting community. If monitoring capacity is limited, active participation may initially rise with proposal volume but eventu- ally fail to keep pace. Beyond that point, additional governance activity no longer attracts commensurate broad participation, and the effective conduct of governance may shift toward a smaller set of highly active participants. In this sense, the relevant consequence of scale is 2 not merely lower turnout, but the endogenous reconcentration of influence within a formally decentralized system. Empirically, I construct a DAO–quarter panel from proposal and voting activity and study whether decentralized participation exhibits a measurable capacity threshold. Rather than treating concentration as a static property of token ownership, I ask a dynamic ques- tion: does decentralized participation cease to scale beyond a certain governance workload, and do concentration measures shift systematically in the same region? This framing com- plements the earlier papers in the series by moving from public and regulatory monitoring environments to decentralized digital governance, and it complements the DAO literature by focusing not only on whether control is concentrated, but also on one process through which concentration may emerge endogenously. Contributions. Thisnotemakesthreecontributions. First,itextendsthebroadertoo-big- to-monitor research agenda to decentralized digital organizations and provides an empirical test of monitoring-capacity limits in a setting where authority is already formally decen- tralized. Second, it estimates a capacity cutoff in the relationship between proposal volume and active voter participation using a fixed-effects kink specification with a data-driven breakpoint, and then studies whether concentration outcomes—including the Herfindahl– Hirschman Index and top-3 voting control—shift in the same region, both as functions of proposal scale and of monitoring-load measures such as proposals per voter. Third, it offers a compact empirical micro-foundation for the concentration patterns documented by Appel and Grennan (2023, 2026): when governance workload grows faster than broad participation capacity, effective control may drift toward a smaller set of highly active participants. More broadly, the note provides a transparent empirical pipeline that can be extended to hetero- geneity in delegation, quorum rules, and DAO design, linking naturally to recent work on governance architecture and value creation in DAOs (Bellavitis and Momtaz, 2025; Zhang, 2025). 2 Data and Measurement TheanalysisusesaDAO–quarterpanelconstructedfromproposal-andvote-levelgovernance records. Each observation corresponds to DAO i in calendar quarter t. The empirical objective is to relate three objects measured within the same DAO–quarter: the scale of governance activity, the breadth of realized participation, and the concentration of effective voting control. This measurement strategy follows the spirit of Appel and Grennan (2023), who study DAO governance using proposal and vote records and show that a small number 3 of entities often control most decisions. It also complements Appel and Grennan (2026), who emphasize that DAO governance outcomes depend on institutional design and broad participation rather than on decentralization in name alone. 2.1 Variable construction The panel is built by aggregating proposal and voting activity to the DAO–quarter level. Let i index DAOs and t quarters. Governance scale. Let P denote the number of governance proposals brought to a vote it in DAO i during quarter t. This is the paper’s baseline measure of governance workload. The main running variable is x = ln(1+P ), it it whichallowsproportionalcomparisonsacrossDAOswhilepreservingobservationswithsmall proposal counts. Participation capacity. Let V denote the number of active voters in DAO i during it quarter t, defined as the number of distinct voting entities (or, when entity-level aggregation is unavailable, distinct voting addresses) that cast at least one vote on at least one proposal during that quarter. In the note, V is interpreted as a revealed measure of monitoring and it participation capacity: it captures how many participants actually engage in governance when proposals arrive. The main outcome in the capacity regressions is y = ln(V ). it it Monitoring load. A central premise of the note is that governance scale matters not only in levels but also relative to the breadth of active participation. I therefore define the DAO–quarter monitoring-load measure as proposals per active voter: P it load ≡ , ℓ ≡ ln(1+load ). it it it V it This variable is not used in the baseline capacity regression, which focuses on how active participation scales with proposal volume, but it is used in the concentration analysis as a compact measure of governance burden per active participant. For robustness, I also use N , the number of voters recorded in DAO i during quarter it t, to construct an alternative workload measure based on proposals per voter rather than proposals per active voter. 4 Governance concentration. The note studies two outcome measures of concentration, both constructed from realized voting outcomes within a DAO–quarter rather than from token holdings alone. This choice is deliberate: the question is who effectively participates in and controls decisions, not simply who holds tokens. Let s denote entity j’s share of realized voting power cast in DAO i during quarter t, jit aggregated across all proposals in that quarter. Using these within-quarter vote shares, I construct: 1. the Herfindahl–Hirschman Index (cid:88) HHI = s2 , it jit j which increases as realized voting influence becomes more concentrated; and 2. the Top-3 control share (cid:88) Top3 = s , it jit j∈T(3) it where T (3) denotes the three entities with the largest realized voting shares in DAO i, it quarter t. These measures are closely aligned with the control-oriented perspective in Appel and Grennan (2023), who emphasize that DAO decisions are often effectively determined by a very small number of participants. In the present paper, the contribution is dynamic: rather than treating concentration as a fixed feature of DAO governance, I ask whether concentration changes systematically once governance workload becomes sufficiently high relative to broad participation. 2.2 Sample and estimation samples Theanalysisfocusesonthepost-2020period. Idistinguishbetweenthefullpost-2020sample and narrower nested estimation samples required by specific specifications. The full sample contains all DAO-quarter observations that survive the baseline cleaning rules. The capacity sample additionally requires nonmissing proposals and active voters so that ln(1+P ) and it ln(V ) are observed. The harmonized concentration sample additionally requires nonmissing it monitoring-loadandconcentrationmeasures. Inthecurrentworkingpanel, thefirstobserved quarter is 2020q2. Table 1 reports sample coverage and descriptive statistics for these nested samples. I use three nested samples. 5 1. The full post-2020 sample contains all DAO–quarters satisfying the baseline data- cleaning rules. 2. The capacity sample requires no missing P and V , so that both ln(1+P ) and ln(V ) it it it it are observed. 3. The harmonized concentration sample additionally requires nonmissing ℓ = ln(1 + it P /V ), HHI , and Top3 . This harmonized restriction ensures that comparisons it it it it across the concentration specifications are not driven by outcome-specific missingness. Table 1 reports sample coverage and descriptive statistics. The table shows that the sample contains 686 DAO–quarters from 136 DAOs over 10 calendar quarters, with estimat- ing samples of 680 DAO–quarters for the capacity analysis and 679 DAO–quarters for the harmonized concentration analysis. Proposal activity is highly skewed, active participation varies substantially across DAOs and over time, and both HHI and the Top-3 control share indicate sizable dispersion in realized voting concentration. These patterns are consistent with the broader DAO evidence in Appel and Grennan (2023) that formal decentralization often coexists with substantial concentration of effective control. Table 1: Sample overview and descriptive statistics (POST-2020) Fullsample Capacitysample Harmonizedconcentrationsample Variable N Mean SD Median N Mean SD Median N Mean SD Median Proposals 680 15.384 32.542 7.000 680 15.384 32.542 7.000 679 15.405 32.561 7.000 Activevoters 686 642.327 1851.843 123.500 680 647.107 1859.264 128.500 679 646.788 1860.616 128.000 ln(1+proposals) 680 2.127 1.034 2.079 680 2.127 1.034 2.079 679 2.129 1.033 2.079 ln(activevoters) 686 4.851 1.841 4.816 680 4.860 1.843 4.856 . . . . Proposals/activevoters . . . . . . . . 679 0.230 0.813 0.055 ln(1+proposals/activevoters) . . . . . . . . 679 0.139 0.283 0.054 HHI . . . . . . . . 679 0.300 0.218 0.236 Top-3controlshare . . . . . . . . 679 0.695 0.376 0.889 Sampleoverview DAO-quarters 686 680 679 DAOs 136 135 135 Calendarquarters 10 10 10 Firstquarter 2020q2 2020q2 2020q2 Lastquarter 2022q3 2022q3 2022q3 Notes: The full sample contains all DAO-quarter observations in the post-2020 working panel that survive the baseline cleaning rules. Reported N may vary across rows within a sample column because descriptivestatisticsarecalculatedusingnonmissingobservationsforeachvariable. Thecapacitysample additionally requires nonmissing ln(active voters) and ln(1+proposals). The harmonized concentration sample additionally requires nonmissing monitoring-load measures, HHI, and Top-3 control share. 3 Empirical Strategy This note studies whether broad participation in DAO governance exhibits a measurable capacity limit and whether realized voting concentration changes in the same range of gov- 6 ernance workload. The empirical design is intentionally parsimonious. All specifications are estimated on DAO–quarter panels with DAO fixed effects and quarter fixed effects, and standard errors are clustered at the DAO level. The results should therefore be interpreted as reduced-form evidence based on within-DAO changes over time rather than as fully causal estimates. 3.1 Capacity kink model To estimate when decentralized participation becomes capacity constrained, I fit a fixed- effects kink regression of active participation on proposal workload: y = α +γ +β x +β (x −c) +ε , (1) it i t 1 it 2 it + it where x = ln(1+P ), y = ln(V ), it it it it P is the number of proposals in DAO i during quarter t, V is the number of active voters, it it α are DAO fixed effects, γ are quarter fixed effects, and i t (x −c) ≡ max{x −c,0} it + it is the kink term. In this specification, β is the slope below the cutoff and β + β is the slope above 1 1 2 the cutoff. A negative β indicates that active participation becomes less responsive to 2 additional proposal flow once workload crosses the threshold c, which is consistent with a monitoring-capacity or attention-capacity constraint. 3.2 Data-driven breakpoint selection I select the breakpoint c by grid search over candidate values spanning the interior of the running-variabledistribution,specificallythe10th–90thpercentilesofx . Foreachcandidate it cutoff, I estimate (1) and compute the residual sum of squares. The estimated cutoff is the value that minimizes this criterion. This procedure is descriptive rather than structural. It is designed to locate the point at which the relationship between proposal workload and active participation changes most sharply in the data. To assess sensitivity to sampling variation, the appendix reports boot- strap evidence based on cluster resampling at the DAO level. 7 3.3 Concentration transitions with free cutoffs The second part of the analysis asks whether realized voting concentration also exhibits a transition as governance workload rises. Rather than imposing the participation cutoff on concentration outcomes, the baseline concentration specifications allow each outcome to choose its own breakpoint. For a generic concentration outcome z , I estimate it z = α +γ +δ r +δ (r −c ) +u , (2) it i t 1 it 2 it z + it where z is either the Herfindahl–Hirschman Index HHI or the Top-3 control share Top3 , it it it r is the relevant running variable, and c is selected separately for each outcome by the it z same RSS-minimizing grid-search procedure. I use two running variables. First, I use proposal scale directly: r = ln(1+P ). it it These specifications ask whether concentration changes once raw proposal volume becomes sufficiently large. Second, I use monitoring load per active participant: (cid:18) (cid:19) P it r = ℓ = ln 1+ . it it V it These specifications ask whether concentration changes when governance burden rises rela- tive to the breadth of realized participation. Allowing HHI and Top-3 to choose their own cutoffs is important in this setting. The purpose of the note is not to impose a common threshold across all outcomes, but to examine whether different indicators of effective control display a similar regime-change pattern as governance workload increases. Robustness to alternative load definitions Because the active-load measure P /V uses realized participation in the denominator, I it it also report robustness checks using an alternative load measure based on the broader voting base: (cid:18) (cid:19) P ℓNV = ln 1+ it , it N it 8 where N is the number of voters recorded in DAO i during quarter t. I then re-estimate it the concentration kink regressions for HHI and Top3 using ℓNV as the running variable it it it and allowing each regression to choose its own cutoff. This robustness exercise helps show that the main concentration patterns are not driven solely by the use of active voters in the denominator. Interpretation The identifying variation in all specifications comes from within-DAO changes in proposal workload and monitoring load over time, after absorbing DAO fixed effects and common quarter shocks. This design does not imply that proposal flow is exogenous. Periods with unusually many proposals may also coincide with disputes, treasury events, strategic mobi- lization, or other DAO-specific shocks. For that reason, I interpret the estimates as disci- plined descriptive evidence consistent with a too-big-to-monitor mechanism, not as definitive proof that proposal volume alone causes lower participation or greater concentration. The purpose of the empirical strategy is narrower: to document whether governance workload, participationsaturation, andrealizedconcentrationmovetogetherinawaythatisconsistent with a monitoring-capacity interpretation. 4 Results This section proceeds in three steps. I first estimate the participation-capacity breakpoint. I then examine whether realized concentration changes as governance activity rises, using both proposal scale and proposals per active voter. Finally, I assess robustness to an alternative monitoring-load definition and report bootstrap evidence on cutoff uncertainty. Throughout, the results are interpreted as reduced-form evidence on within-DAO regime changes rather than as fully causal estimates. 4.1 Capacity breakpoint: participation stops scaling proportion- ally with proposal volume Table 2 reports the baseline capacity specification, which relates ln(V ) to ln(1+P ) with it it DAO and quarter fixed effects. The breakpoint is chosen by grid search to minimize the residual sum of squares in the kink specification. The estimated cutoff is cˆ = 2.3441 cap in ln(1 + proposals), which corresponds to approximately e2.3441 − 1 ≈ 9.4 proposals in a DAO-quarter. 9 The main result is a statistically significant decline in the marginal responsiveness of activeparticipationonceproposalflowcrossesthisthreshold. Belowthecutoff, theestimated slope is 1.104; above the cutoff it falls to 0.601. Thus, proposal activity continues to attract participation beyond the threshold, but at a substantially lower marginal rate. This is the core empirical signature of the note’s monitoring-capacity mechanism: broad participation rises with governance workload at low levels of activity, but eventually fails to keep pace. Figure 1 reinforces this interpretation. Panel (a) plots the RSS objective over candidate cutoffs and shows a well-defined interior minimum. Panel (b) provides a binned residual plot around the estimated cutoff and makes the change in slope visually transparent. .6 .4 .2 0 -.2 (a) RSS over candidate cutoffs )EF retrauq + EF OAD( )sretov evitca(nl laudiseR -1 -.5 0 .5 1 ln(1+proposals) - c_cap (b) Residualized participation around cˆ cap Figure 1: Participation capacity breakpoint Notes: Panel (a) plots the residual sum of squares from the fixed-effects kink regression for candi- date cutoffs in ln(1+proposals). Panel (b) shows a binned residual plot of ln(active voters) after partialling out DAO and quarter fixed effects, centered at the estimated breakpoint. 4.2 Governance concentration under monitoring load and pro- posal scale I next turn from participation to realized concentration. I begin with a more mechanism- oriented measure of governance burden, ln(1+P /V ), which scales proposal flow by active it it participation. This specification is especially relevant for the paper’s interpretation, because it more directly captures monitoring load per participant. Table 3 shows that the concentration–load relationship is strongly nonlinear. For HHI, the estimated slope is 0.978 below the cutoff and 0.200 above it. For the Top-3 control share, the corresponding slopes are 5.091 and 0.193. In both cases, the kink is highly statistically significant. The pattern is therefore not a reversal but a sharp flattening: as monitoring load rises, concentration increases steeply at low levels of burden and then continues to rise much more slowly beyond the estimated threshold. 10 Table 2: Capacity breakpoint: participation stops scaling proportionally with proposal vol- ume (1) Linear (2) Kink ln(active voters) ln(active voters) ln(1+proposals) 0.882*** 1.104*** (0.089) (0.113) (ln(1+proposals)−cˆ ) -0.503** cap + (0.198) Observations 680 680 Estimated cutoff cˆ 2.3441 cap Slope below cutoff 1.104 Slope above cutoff 0.601 p-value (kink) 0.0123 Notes: The dependent variable is ln(active voters). All specifications include DAO fixed effects and quarter fixed effects. Standard errors, in parentheses, are clustered at the DAO level. The breakpointisselectedbyRSS-minimizinggridsearch. Sample: capacityestimationsamplefromthe post-2020 working panel. Significance: * p<0.10, ** p<0.05, *** p<0.01. Figure 2 provides a visual counterpart. The binned residual plots show a steep positive relationship at low levels of active monitoring load and a much flatter relationship after the outcome-specific cutoff. This is consistent with a saturation mechanism in which increases in governance burden initially coincide with more concentrated realized control, but the marginal effect weakens once the DAO is already operating under high monitoring strain. I then return to proposal scale itself, using ln(1 + P ) as the running variable and al- it lowing each concentration outcome to choose its own breakpoint. This free-cutoff design is intentionally descriptive. Its purpose is not to force concentration to bend exactly where par- ticipation bends, but to ask whether concentration also exhibits a regime change as proposal activity rises. Table 4 shows that for HHI, the concentration–scale relationship reverses sign. At low-to-moderate proposal volume, concentration declines with scale: the estimated slope below the cutoff is −0.046. Beyond the estimated threshold, however, the relationship turns positive, withanabove-cutoffslopeof0.029. Thekinkisstatisticallysignificant(p = 0.0321), which suggests that the apparent decentralization gains from scale do not persist indefinitely. For the Top-3 control share, the point estimates tell a similar qualitative story. The slope is negative below the estimated cutoff and positive above it, but the kink is less precisely estimated. Taken together, these proposal-scale results are consistent with a gradual transi- tion away from broad-based governance as activity becomes more demanding, although the evidence is stronger for HHI than for Top-3 concentration. 11 .1 .05 0 -.05 )EF retrauq + EF OAD( IHH laudiseR 0 .5 1 1.5 2 2.5 ln(1 + proposals/active voters) .1 .05 0 -.05 -.1 )EF retrauq + EF OAD( lortnoc 3poT laudiseR 0 .5 1 1.5 2 2.5 ln(1 + proposals/active voters) Figure 2: Governance concentration and active monitoring load Notes: The figure plots binned residual relationships between concentration outcomes and ln(1+ proposals/active voters) after partialling out DAO and quarter fixed effects. Each panel is centered at the outcome-specific cutoff selected by the corresponding kink regression. The visual pattern is oneofsteepincreaseatlowmonitoringloadfollowedbysubstantialflatteningbeyondtheestimated threshold. Table 3: Governance concentration and active monitoring load HHI Top-3 control share (1) Linear (2) Kink (3) Linear (4) Kink ln(1+proposals/active voters) 0.371*** 0.978*** 0.298*** 5.091*** (0.106) (0.176) (0.094) (0.993) (ln(1+proposals/active voters)−cˆ ) -0.778*** -4.898*** z + (0.192) (0.997) Observations 679 679 679 679 Estimated cutoff cˆ 0.3037 0.0608 z Slope below cutoff 0.978 5.091 Slope above cutoff 0.200 0.193 p-value (kink) 0.0001 0.0000 Notes: The dependent variable is either the Herfindahl–Hirschman Index (HHI) of realized voting con- centration or the Top-3 control share. Monitoring load is defined as ln(1+proposals/active voters). All specifications include DAO fixed effects and quarter fixed effects. Standard errors, in parentheses, are clustered at the DAO level. The breakpoint is selected separately for each outcome by RSS-minimizing grid search. Sample: harmonized concentration sample from the post-2020 working panel. Significance: * p<0.10, ** p<0.05, *** p<0.01. 12 Figure 3 provides a visual counterpart. The binned residual plots again suggest a gradual transition rather than a discrete jump, which is what one would expect from a saturation mechanism in which monitoring capacity becomes progressively strained. .04 .02 0 -.02 -.04 )EF retrauq + EF OAD( IHH laudiseR 1 2 3 4 5 6 ln(1+proposals) .05 0 -.05 )EF retrauq + EF OAD( lortnoc 3poT laudiseR 1 2 3 4 5 6 ln(1+proposals) Figure 3: Governance concentration and proposal scale Notes: The figure plots binned residual relationships between concentration outcomes and ln(1+ proposals)afterpartiallingoutDAOandquarterfixedeffects. Eachpanelusestheoutcome-specific cutoff selected by the corresponding kink regression. 4.3 Robustness: alternative monitoring-load definition A natural concern is that monitoring burden may look mechanically high when proposals are scaled by realized participation. To address this, Table 5 replaces proposals per active voter with proposals per recorded voters, (cid:18) (cid:19) P ℓNV = ln 1+ it , it N it where N is the number of voters recorded in the DAO-quarter. This is a more conservative it denominator, since it scales proposal flow by a broader voting base. The qualitative pattern is the same. For both HHI and the Top-3 control share, con- centration rises steeply with this alternative load measure at low levels of burden and then flattens sharply after the estimated cutoff. For HHI, the slope falls from 3.635 below the cutoff to 0.155 above it. For the Top-3 control share, the corresponding slopes are 2.418 and 13 Table 4: Governance concentration and proposal scale HHI Top-3 control share (1) Linear (2) Kink (3) Linear (4) Kink ln(1 + proposals) -0.026 -0.046** -0.027 -0.039 (0.016) (0.021) (0.021) (0.025) (ln(1 + proposals) − cˆ ) 0.075** 0.097 z + (0.035) (0.073) Observations 679 679 679 679 Estimated cutoff cˆ 2.8653 3.5264 z Slope below cutoff -0.046 -0.039 Slope above cutoff 0.029 0.058 p-value (kink) 0.0321 0.1825 Notes: The dependent variable is either the Herfindahl–Hirschman Index (HHI) of realized voting con- centrationortheTop-3controlshare. Therunningvariableisln(1+proposals). Allspecificationsinclude DAO fixed effects and quarter fixed effects. Standard errors, in parentheses, are clustered at the DAO level. The breakpoint is selected separately for each outcome by RSS-minimizing grid search. Sample: harmonizedconcentrationsamplefromthepost-2020workingpanel. Significance: *p<0.10,**p<0.05, *** p<0.01. 0.121. In both cases, the kink is highly statistically significant (p = 0.0004). These results indicate that the link between governance burden and concentration is not an artifact of using active voters in the denominator. 4.4 Bootstrap uncertainty for estimated cutoffs Because all breakpoints are selected in a data-driven way, it is useful to assess how sensitive they are to sampling variation. Table 6 reports DAO-level cluster-bootstrap distributions for theestimatedcutoffs. Theparticipationcutoffisreasonablystable: thebootstrapmedianfor cˆ is2.093,witha95%percentileintervalof[1.205, 3.556]. Bycontrast,theexactlocationof cap theconcentrationcutoffsisestimatedlessprecisely,especiallywhenrawproposalscaleisused as the running variable. The load-based concentration cutoffs are more tightly concentrated, which is consistent with the paper’s mechanism: governance burden per participant appears to be more informative than proposal count alone. I therefore view the bootstrap exercise as supporting the existence of regime change while cautioningagainstover-interpretingtheprecisenumericallocationofanysingleconcentration threshold. 14 Table 5: Robustness: governance concentration and alternative monitoring load HHI Top-3 control share (1) Linear (2) Kink (3) Linear (4) Kink ln (cid:0) 1+ proposals (cid:1) 0.163*** 3.635*** 0.160*** 2.418*** number of voters (0.036) (0.952) (0.054) (0.622) (ln(1+proposals/number of voters)−cˆ ) -3.480*** -2.298*** z + (0.959) (0.637) Observations 608 608 608 608 Estimated cutoff cˆ 0.0335 0.0944 z Slope below cutoff 3.635 2.418 Slope above cutoff 0.155 0.121 p-value (kink) 0.0004 0.0004 Notes: Monitoring load is defined as proposals divided by number of voters. All specifications include DAO fixed effects and quarter fixed effects. Standard errors, in parentheses, are clustered at the DAO level. The cutoff is selected separately for each outcome by RSS-minimizing grid search. Sample: DAO-quarters with nonmissing alternative load, concentration outcomes, and fixed effects. Significance: * p<0.10, ** p<0.05, *** p<0.01. Table 6: Bootstrap uncertainty for estimated cutoffs Cutoff Reps. Mean P50 P2.5 P97.5 c : capacity cutoff in ln(1+proposals) 300 2.242 2.093 1.205 3.556 cap c : implied load cutoff 300 0.020 0.012 0.003 0.057 load,cap c : HHI cutoff vs. proposals 300 2.837 3.122 0.822 3.761 HHI,P c : Top-3 cutoff vs. proposals 300 2.075 1.757 0.741 3.725 Top3,P c : HHI cutoff vs. active load 300 0.292 0.290 0.035 0.421 HHI,L c : Top-3 cutoff vs. active load 300 0.075 0.055 0.039 0.278 Top3,L Notes: DAO-levelclusterbootstrap. Ineachreplication, DAOsareresampledwithreplacement and the full cutoff-selection procedure is repeated. P50 denotes the bootstrap median. The capacity cutoff is estimated more stably than the concentration cutoffs, and the load-based concentration cutoffs are more tightly concentrated than the proposal-based cutoffs. 15

Table 5: Robustness: governance concentration and alternative monitoring load HHI Top-3 control share (1) Linear (2) Kink (3) Linear (4) Kink ln (cid:0) 1+ proposals (cid:1) 0.163*** 3.635*** 0.160*** 2.418*** number of voters (0.036) (0.952) (0.054) (0.622) (ln(1+proposals/number of voters)−cˆ ) -3.480*** -2.298*** z + (0.959) (0.637) Observations 608 608 608 608 Estimated cutoff cˆ 0.0335 0.0944 z Slope below cutoff 3.635 2.418 Slope above cutoff 0.155 0.121 p-value (kink) 0.0004 0.0004 Notes: Monitoring load is defined as proposals divided by number of voters. All specifications include DAO fixed effects and quarter fixed effects. Standard errors, in parentheses, are clustered at the DAO level. The cutoff is selected separately for each outcome by RSS-minimizing grid search. Sample: DAO-quarters with nonmissing alternative load, concentration outcomes, and fixed effects. Significance: * p<0.10, ** p<0.05, *** p<0.01. Table 6: Bootstrap uncertainty for estimated cutoffs Cutoff Reps. Mean P50 P2.5 P97.5 c : capacity cutoff in ln(1+proposals) 300 2.242 2.093 1.205 3.556 cap c : implied load cutoff 300 0.020 0.012 0.003 0.057 load,cap c : HHI cutoff vs. proposals 300 2.837 3.122 0.822 3.761 HHI,P c : Top-3 cutoff vs. proposals 300 2.075 1.757 0.741 3.725 Top3,P c : HHI cutoff vs. active load 300 0.292 0.290 0.035 0.421 HHI,L c : Top-3 cutoff vs. active load 300 0.075 0.055 0.039 0.278 Top3,L Notes: DAO-levelclusterbootstrap. Ineachreplication, DAOsareresampledwithreplacement and the full cutoff-selection procedure is repeated. P50 denotes the bootstrap median. The capacity cutoff is estimated more stably than the concentration cutoffs, and the load-based concentration cutoffs are more tightly concentrated than the proposal-based cutoffs. 15 5 Conclusion This note studies a simple limit of decentralized governance: broad participation may not scale indefinitely with governance workload. Using DAO–quarter data, I show that active participation rises with proposal activity up to an interior breakpoint and then becomes sub- stantially less responsive. This pattern is consistent with a monitoring-capacity constraint: as proposal flow increases, the burden of following, evaluating, and voting on governance decisions begins to outpace the capacity of dispersed participants. I then show that realized voting concentration exhibits related transition patterns. When governance burden becomes sufficiently high, concentration outcomes cease to evolve in the same way they do at lower levels of activity. This is especially clear when workload is measured relative to the breadth of participation. Overall, the results suggest that decentralization gains become harder to sustain once governance activity becomes sufficiently demanding. I present a simple mechanism linking the concentration facts documented by Appel and Grennan (2023, 2026) to the scale of governance activity itself: when proposal flow grows faster than broad participation can keep up, effective control may drift toward a smaller set of highly active participants. The paper is intentionally modest in scope. The estimates are reduced-form and should be read as descriptive evidence on within-DAO regime changes rather than as definitive causal effects of proposal flow. Even so, the patterns are informative. They suggest that formal decentralization does not eliminate organizational capacity constraints; instead, those constraints may reappear as participation saturation and renewed concentration of influence. More broadly, the note extends the “too big to monitor” framework to decentralized digi- tal organizations (Tchuente, 2025, 2026). The main implication is not that DAOs inevitably centralize, but that decentralized governance may require institutional responses—such as delegation design, proposal screening, agenda management, or other participation-saving mechanisms—once governance workload grows sufficiently large. References Ammons, John and Christos Makridis,“FromIdealstoInstitutions: TransactionCosts, Risk, and Governance in Decentralized Autonomous Organizations,” SSRN Scholarly Pa- per 5377155, SSRN August 2025. Appel, Ian and Jillian Grennan, “Control of Decentralized Autonomous Organizations,” AEA Papers and Proceedings, 2023, 113, 182–185. 16 and , “DAO Governance: Decentralized But Not So Disorganized,” SSRN Scholarly Paper 6065407, SSRN January 2026. Bellavitis, Cristiano and Paul P. Momtaz, “Voting governance and value creation in decentralizedautonomousorganizations(DAOs),” Journal of Business Venturing Insights, 2025, 24, e00537. , Christian Fisch, and Paul P. Momtaz, “The rise of decentralized autonomous organizations (DAOs): a first empirical glimpse,” Venture Capital, 2023. Bongaerts, Dion, Frank De Jong, Joost Driessen, Igor Makarov, and Nikolai Reshchikov, “Vote Delegation in DeFi Governance,” 2025. Han, Jungsuk, Jongsub Lee, and Tao Li, “A review of DAO governance: Recent literature and emerging trends,” Journal of Corporate Finance, 2025, 91, 102734. Nassif, Basilio and Andreas Savva, “Decentralized Autonomous Organizations: Gover- nance and Implementation,” Applied Sciences, 2024, 14 (16), 7007. Sharma, Tanusree, Yujin Potter, Kornrapat Pongmala, Henry Wang, Andrew Miller, Dawn Song, and Yang Wang, “Future of Algorithmic Organization: Large- Scale Analysis of Decentralized Autonomous Organizations (DAOs),” 2024. Tchuente, Guy, “Too Big to Monitor? Network Scale and the Breakdown of Decentralized Monitoring,” arXiv preprint arXiv:2511.23320, 2025. , “Scale and Capacity Limits in Decentralized FDA Food-Safety Enforcement,” arXiv preprint arXiv:2602.12392, 2026. Zhang, Junzi, “Governance in Decentralized Finance (DeFi): A Comparative Study of On-Chain Voting and Delegation,” Working Paper 33639, National Bureau of Economic Research April 2025. A Appendix: Suggested figure to motivate the setting and data source Figure 4 provides a simple motivating fact in the DAO–quarter panel: proposal volume rises sharply over time, while average voting concentration remains substantial. This pattern 17 .45 .4 .35 .3 .25 IHH naeM 2000 1500 1000 500 0 slasoporP fo rebmuN 2020q2 2020q4 2021q2 2021q4 2022q2 2022q4 Quarter Figure 4: Trends in proposals and voting concentration (motivating evidence) is consistent with related evidence on concentration in DAO governance from Appel and Grennan (2023). Thedatausedinthispaperareconstructedfromthereplicationmaterialsassociatedwith Appel and Grennan (2023), together with proposal- and vote-level governance files used to build the DAO-quarter panel. From these source files, I construct measures of proposal volume, active voter participation, monitoring load, and realized voting concentration. The final estimation samples are obtained by applying the sample restrictions described in the paper. 18