It has been almost a month since I wrote “Why Projects in the Prediction-Market Ecosystem Are Hard to Build.” Here is a record and summary of the project’s progress over the past month.
Picking up where the previous post left off: what happened to Virae’s planned paper trading competition? It finally got underway after a two-week delay, with KOLs brought in to repost the promotion.
In other words, the promotion and testing originally scheduled to start on August 10 were pushed back to August 21.
What did we do during those two weeks? Mainly, we added a Guide mode to the website for beginners, the kind that walks you through each operation step by step. Someone thought the original website and interface were too “Pro,” suitable only for people already familiar with prediction markets, and that beginners would find them confusing and complicated.
How did that turn out? History keeps repeating itself: neither the original Pro version nor the new Guide version had any real users.
Let me unpack two problems with the logic here.
The reason we had to build Guide mode was that someone did not understand how to use the product, thought its complexity had caused him to lose money, and demanded improvements across the entire product.
The question is, why did that have to hold up the paper trading competition? Paper trading, as the name suggests, does not use real money. It uses a virtual balance provided by the platform. However confusing the project’s Agents might be, or however easily they might cause people to lose money, why should that block this competition?
Of course, you could ask: what if the paper trading competition converts some participants into real-money users, and they lose money? This is another case of crying wolf. We have been through this more than once: “Suppose lots and lots of users come. How much commitment and product quality do we need to show before the product is ready to promote?”
But what was paper trading supposed to achieve? It was meant to find out whether anyone was interested in agentic trading for prediction markets. We were not yet at the stage of trying to convert users at scale.
Even if real users showed up, how would we determine whether they preferred the Pro version or the Guide version? Would Guide’s success conclusively establish Pro’s failure? Conversely, would Guide’s failure prove that Pro would fail too?
This is not a matter of personal preference. It is a matter of controlled experiments. We learned about controlled experiments in elementary school, but perhaps these experienced, money-making adults have forgotten all about them. If the project ultimately succeeds, will it be thanks to the two weeks spent building Guide mode, or because the project’s direction was attractive to begin with?
Anyway, after that two-week delay, the Paper Trading competition is now scheduled to end tomorrow, September 7. This will also be the first real test of conversion: will anyone try live trading because of the returns they saw in paper trading?
Adding Guide mode also changed how I work. Previously, whenever I changed a feature, I would post about it in the group chat, or ask ChatGPT to generate a daily summary for me to share there.
But someone not only lost money on Virae because he did not understand how to use it, he also slowed down the entire project and made us spend extra time and effort on improvements. That made one thing painfully clear to me: nobody reads the feature updates posted in the group. He was the one who asked for those updates in the first place, and then he did not read them.
So I stopped posting summaries in the group. The last time I sent a feature update summary there was August 10.
I did keep up my own habit of writing summaries and keeping records. This page contains the changes I made to Virae from July 12 through September 2:
https://smallyunet.github.io/virae-logs/
After September 2, the summaries really did stop, because improvements to Virae’s features stopped too.
If you have read “Why the DM KOL GTM Strategy Is Necessarily Correct,” you know that I am not particularly optimistic about this strategy.
The week of August 22–28 was a confusing one for the team. By then, the so-called Guide version had been added, and promotion of the paper trading competition had been handed over to our operations colleague. Development and product were left in a state of not knowing what to do next.
In our meeting on Friday, August 28, we said that everyone could continue DMing KOLs. The plan was to contact 50 KOLs over one week.
After a weekend of doing nothing, on Monday, August 31, I proposed that we stop everything else: product should stop designing, engineering should stop building features, and everyone should focus entirely on DMing KOLs, with a goal of messaging 500 KOLs over two weeks. (I was not optimistic, but since everyone said it worked, we might as well try.)
Let me explain my motivation and reasoning. We would set a cutoff for our losses and see whether a sufficiently large sample over a sufficiently long period could tell us whether Virae was heading in the right direction. The sample size means the number of KOLs contacted: how many people would we need to contact before we could judge whether the audience and market were interested in Virae’s product direction? The time limit means how long the boss was willing to spend on this exercise of validating the project’s direction.
Of course, this raises several questions.
To avoid letting my own shortcomings in experience, methods, or communication skills skew the results, I have kept proposing a matrix approach to contacting KOLs. In other words, each of us would contact the same KOL independently, on our own schedule and with our own wording. That could compensate for individual weaknesses. If a KOL ignores one person, the problem might be with that person. If the KOL ignores all ten people, then the problem is with the KOL.
Suppose we contact 500 KOLs and nobody is interested in Virae. Would that prove the project’s direction is wrong? No, it would not. But at least we would have data on the relationship between the effort invested and the results. That is easy enough to understand, right?
If you say this project will succeed someday, but you do not know when, then working hard, trying different approaches, and persevering are all perfectly normal. People might even be happy to do that.
For a salesperson, if one out of every 500 people is a target customer, finding that one person is simply part of the job. Find that one customer, and the job is done: you get your salary and your bonus. That logic holds together.
But from the perspective of the people running the project, do we need to judge whether its direction is right? If we really get data showing that one in every 500 people in the market is a target user, what does that mean for the project as a whole?
Recently, a restaurant adviser known as Yongge has become popular on short-video platforms. All it takes is a 360-degree camera sweep of the premises for him to judge whether a restaurant can keep operating. He once said something that made sense: many restaurants are already out of business the day they open; they just hang on for a few months before shutting their doors.
My concern follows the same logic. If the project’s direction was wrong from the start, then it had already failed when it launched. It would just take a few months of persistence and effort before we stopped and gave up.
User numbers across the Polymarket ecosystem are public onchain, with roughly one million monthly active addresses. Data from the World Cup this July showed an 8:2 split between users of Polymarket’s official website and users of third-party products. That means third-party projects collectively have roughly 200,000 monthly active users. Now you can see what that one-in-500 ratio could be useful for, right?
I am just an ordinary employee. My role is to earn a salary, so why should I worry about whether the project can succeed?
Unless a project makes you financially independent enough to retire, you will still need to work afterward. And in the job market, experience on a successful project and experience on a failed one are worlds apart.
On a failed project, people will not care even if you did technically demanding work, and explaining its value takes a huge amount of effort. On a successful, well-known project, it is easy to get recognition even if your own work was not particularly substantial. This is a difference I have felt clearly through years of repeated job hunting. I am not saying this out of envy of people who get recognition easily. My own experience includes both unsuccessful projects and projects that were not such failures.
So, to some extent, working on a failed project also uses up future career opportunities.
For years, I have changed jobs frequently. Each time I gave one reason or another, but fundamentally, it was because I had come to see that the project could not succeed.
If I had always been content simply to collect a salary, perhaps my career would have been much more stable, my debt lighter, and my résumé better-looking. Clearly, though, that is not the path I took.
Asking me now to settle for a salary without considering or judging whether the project can succeed conflicts with how I have acted in the past. If I settle for a salary now, what was all that past instability and sacrifice for?
Doesn’t that worry you too?
One week has now passed. I have moved relatively quickly and processed nearly 400 targets. Here is a summary of this week’s results, from a sample of almost 400:
| Metric | People | Rate |
|---|---|---|
| Contacts successfully sent a DM | 284 | — |
| Received an initial reply after a DM | 27 | 9.51% (27/284) |
| Sent a second message after the initial reply | 22 | 81.48% (22/27) |
| Received another reply to the second message | 9 | 40.91% (9/22) |
| DM respondents who explicitly mentioned paid collaboration or quoted a price | 6 | 22.22% (6/27) |
| Contacts reached through comments (103 comments in total) | 103 | — |
| Received a reply to a comment | 5 | 4.85% (5/103) |
| People who actually tried Virae | 0 | 0 |
I do understand that some self-suggestion may be at work here. From the start, I expected the results to be poor, and I hoped other team members would do better than me. But thinking it over, how much better can they really do? Although I do sincerely hope they do better.
If you are sensitive or empathetic, you may already have noticed an important psychological motivation behind setting a cutoff: a lack of confidence in the project’s direction.
What does it mean when a developer on a project has no confidence in its direction? Either my judgment is poor and I cannot grasp the project’s long-term vision, or… There are plenty of personal preferences involved here. Maybe the market and some unknown group of users really do go for this, want to use a project like this, and are willing to pay for it. You may dislike gambling, but if you had the chance to open a profitable casino, you would take it, right? All that matters is that someone likes gambling.
That is what we are doing with Virae now: looking for gamblers, looking for people willing to gamble.
As for me personally—just me—I do not gamble. Does CZ gamble? Does Star Xu gamble? Does Justin Sun gamble? They run the biggest casinos in the world. They do not gamble.
At least for now, the project is not over. There are still at least two things to validate: