The paper also doesn't talk about whether or not the initial assumptions were true. E.g.:

"The ability to delegate voting power allows liquid democracy to scale to large groups as well as representative democracy does. People who have others delegating to them are much like representatives in representative democracy systems, becoming specialists for group decisions and voting with greater power. For example in Figure 2, many people delegate directly or indirectly to person H, making H a powerful delegate. Liquid democracy systems tend to be meritocracies. As a person becomes more of an expert on issues facing the group, more people trust and delegate to them. The increased voting power can incentivize them to further increase their expertise and effectiveness. The opposite effect occurs when the delegate becomes less effective. This feedback process gives accountability and curbs abuses of power.

"The ability to change delegations at any time eliminates many election-related distractions and distortions. There is no election-cycle and delegates must continuously prove their worth. And, they can do so in incremental manner, instead of wasting time and energy on all-or-nothing elections."

So OK. Did people actually take back their delegations? How often did people override and vote directly? Did the optimism work out in practice?

From the paper: "Issues with category "food" have the highest number of votes cast, over 73,000, and the largest number of issues, at 150."

If you think about it, food is probably the most personal and least likely thing you'd delegate, as you have your own preferences and it doesn't require expertise in the subject. If it were for more serious things in which expertise is required, it's more likely people would delegate votes; for example, if there was a vote on the California drought using this system, I'd probably delegate my vote to a hydrologist, as they know more about the subject than I can reasonably gather.

But on the flip side, if food voting is driving it and explains why delegation is a rounding error over multiple years of use, then their 'successful' experience is of little import for people who are interested in whether it would be good for voting on things more serious than burgers.

It just needs to be tested on more complex decisions, possibly in a larger network. Most people don't feel the need to delegate choosing what kind of food they like to eat. Now if google starts using this system to decide what kind of phone they are going to make next, things will get interesting.