Why fewer delegates show up for the tight votes

In issue 001, we found something we could not immediately explain. Across our 1,571 proposals with vote data, median voters-per-proposal has two peaks, one at 20-30% and a higher one at 60-70%, with a valley between them where the 40-60% contested band sits. Tight vote proposals draw fewer voters than the proposals around them, in the opposite direction from what rational pivot-probability would predict.

We laid out three candidates. Information cost: tight votes are tight because they are technical and not popular. Whale neutralization: in a 50-50 vote whales on each side cancel out and retail disengages. Selection effect: tight votes are tight BECAUSE turnout is low, in which case the causation runs the other way around.

We did some testing and here's what we found.

First, let's explain the shape. The chart below plots median voters-per-proposal against FOR-share, bucketed in 10-point steps. Two peaks bracket the contested band: a left peak at 20-30% (median 201 voters) and a higher right peak at 60-70% (median 218 voters). The 70-90% partial-win region also draws high turnout. The 40-60% contested band actually sits in the valley between the two peaks at a median of 74. It is the most counterintuitive thing on the chart. This essentially shows us that the proposals where a single vote is most likely to matter draw fewer voters than proposals already mostly decided.

01The contested band is the gauge band (highly technical)

Throughout the rest of this issue we compare the 40-60% contested band against the 60-90% partial-win band for two reasons. First, those are the only two bands with enough proposals to support per-band statistics (n=39 and n=122). The 30-40% sub-band shares the contested band's low-turnout signature (median 79 voters, similar to the 68-75 inside 40-60%) but with only 7 proposals it is too thin to draw conclusions from. Second, 60-90% is the natural shoulder above the dip, where turnout peaks before the consensus zone above 90% takes over.

We took every proposal in the 40-60% contested band and the 60-90% partial-win band and categorized each one by topic using its title. The categories are rough but consistently fall into these buckets. They are gauge changes, treasury allocations, listings, parameter tweaks, governance meta, technical migrations, and so on. Anything we could not classify falls into "other."

And behold. Gauge proposals are 25.6% of the contested band and 1.6% of the partial-win band, a 24-point gap and the largest in the dataset by a wide margin. Parameter changes (caps, thresholds, risk settings) are the second contested-band-skewed category at 15.4% versus 8.2%. Both gauge and parameter proposals are highly technical.

If gauge proposals were just rare in mainstream governance, you would expect a low non-zero share in the high-turnout bands. Instead gauge proposals are 0% of the 20-30% left peak (n=11) and 0% of the 60-70% right peak (n=18). They simply don't appear where there is higher voter turnout.

The contested band proposals read highly technical: [BIP-57a] Introduce Gauge Framework v1, [BIP-7] Enable auraBAL Gauge, Decide on Gauge Unexpected Behavior, Increase GNO cap from Cap3 to Cap4, What should be the LTV/Liquidation Threshold level for Aave V2?, Raise the proposal quorum threshold. Whereas the partial-win band proposals read like mainstream policy meetings: service-provider renewals, partnership integrations, treasury committee setups, exchange listings, oversight-committee creations.

Gauge proposals in particular are what makes the 40-60% band specialist territory. To vote intelligently on a gauge change you need to understand veBAL mechanics, the gauge weight system, the specific pools at issue, the bribe market dynamics, and your own protocol's exposure. So casual delegates can't vote with confidence and the proposal is decided by the dozen or two people who understand it. What we're really seeing is two opposing factions of specialists, both showing up, that produce the 40-60% contested-band outcome.

Risk parameter changes are also highly technical. LTV adjustments, borrow caps, or liquidation thresholds require understanding the protocol's risk model, the asset's liquidity profile, historical liquidation behavior, and correlation structure.

02Whales aren't canceling each other out

The whale-neutralization idea suggested that the 40-60% band should have higher voting-power concentration than the 60-90% band. In other words, intuition would suggest that in a close vote, large holders on each side neutralize each other and it ultimately becomes a vote between a few whales and in turn smaller voters disengage. If this were true, one would expect higher Gini coefficients, larger top-1 voter shares, and larger top-3 voter shares in the contested band.

Instead, the contested band actually has lower Gini concentration than the partial-win band (0.857 vs 0.923), and the top-1 and top-3 voter shares are roughly the same magnitude across both bands. The contested band has less whale concentration, not more, and whale dynamics don't differentiate the two bands.

03Low turnout does not produce tight outcomes

The selection-effect idea was that the causation might run backwards meaning tight votes are tight simply because only a few people are voting. If a proposal had broader participation, the latent majority preference would assert itself and the outcome would move away from 50/50. We computed the correlation between voter count (log-transformed) and outcome closeness (the absolute distance of FOR-share from 0.5). If selection-effect were true, it would show a positive correlation, e.g. more voters AND outcomes further from 50/50.

However, the actual correlation is -0.087, just slightly negative. So, low turnout is definitely not what makes tight votes tight.

04What this could mean for delegates and analytics tooling

The contested 40-60% band is not where the most exciting governance debates live. The 60-70% sub-band, where we get a lot more voters, (a median 218 voters), is where mainstream-readable governance happens. These proposals, such as service-provider renewals, partnership decisions, oversight committees, listings of well-known assets, get people to show up. These are the proposals where most delegates have a view or can form one quickly, and show up. Whereas, the contested band are the proposals where it's much more difficult for mainstream delegates to form any kind of view due to the highly technical nature of the proposals.

Some takeaways for delegates and analytics tooling.

Tight does not necessarily mean important to your audience. The obvious and logical framing of "look at the close votes" is misleading. The close votes are those that many delegates either didn't or couldn't engage with. They may still be high-stakes, but they are high-stakes for a specific audience.

Filtering by topic is pretty important. The contested band is useful for finding coordination patterns (as in issue 001), but it is not super useful when trying to understand "what the DAO thinks." For the second use case, the 60-90% band is the more honest signal.

If you cover governance, look into covering gauges. The biggest need we're seeing in DAO governance writing right now is accessible coverage of gauge mechanics, parameter changes, and the technical proposals where contested outcomes actually happen. These votes are actually super important to the protocols' economics, but they're the votes that most delegates aren't equipped to evaluate. We believe there is a reader market for "what's actually at stake in this gauge proposal" that nobody is serving well right now.

05One more thing, rejections look different

While running the concentration analysis, the decisive rejection band stood out for an unrelated reason. Decisive rejections (FOR-share below 40%) have a top-1 voting-power share that dwarfs every other category: median 46.5%. The median proposal that fails decisively is not failing because of broad opposition. It is failing because one large holder (a whale) is voting against, often supported by a few peers (top-3 median 82.0%).

But, alas, this is for issue 003. A publication-quality dataset of 'decisive rejections and what is actually happening' probably exists, and we think there is something useful there for delegates to know a little more about how their DAO is actually working.