Challenge

Day 3: I Picked an Idea That Didn't Pass My Own Test

Four ideas, eight questions, one scorecard. The winner scored 12 to 14 out of 16, one of its answers is still blank, and it's on trial until day 7.

By Nghia Vo · September 25, 2026

On day 1 I said I'd score four ideas against eight questions and pick whichever one won. This is that scorecard.

The short version: one idea won, but it didn't clear the bar I set for myself, and one of its answers is still a blank. I'm picking it anyway, as a trial with an end date.

The four ideas

1. People who can't code, building real things with Claude
2. Bloggers and creators stuck picking a niche
3. Sellers in countries where Stripe doesn't work
4. Getting Claude to write in your own voice

The scorecard

Each question gets a 2 if the answer is yes and I can point to where I saw it. A 1 means partly, or the evidence points both ways. A 0 means no. A 1↑ means 1 for now, and more evidence could make it 2. A ? means I haven't checked, and I'm not filling it in with a guess. So every idea gets two totals: the low one counts each ? as 0 and each 1↑ as 1, the high one counts both as 2.

On question 2, the other three ideas have a 1↑. I haven't counted their X posts by hand yet, and ideas 2 and 3 still have Reddit titles I haven't read. For idea 4 I read every Reddit title. The high totals already count each 1↑ as 2.

Idea 3 is out because of question 1. The people posting about it are in Uganda, Nepal, India and Zimbabwe. Same problem I have, but they aren't who I'm trying to sell to.

My bar was 12 out of 16. Idea 1 gets 12 on the questions I've actually checked. But question 1 is a blank, and a 0 there knocks an idea out no matter what the total says. So 12 isn't a pass. I don't know yet whether the people I'd sell to are mostly in the US.

My first draft of question 5 said "X or Reddit". I don't have a Reddit account and I only read it, so I scored the place where I can actually reach people. Counting Reddit gives idea 2 one more point. It still comes second.

A correction to day 1

On day 1 I wrote that 71 percent of the posts about idea 1 were questions, or someone saying where they got stuck. That number came from a script sorting titles by their first few words. Nobody read them.

This week I read them. Titles only, not the posts. For ideas 1 and 2, a script first pulled out titles with words like "stuck" or "help", then I read those one by one. For idea 1, I read 170 and flagged 23 that show signs of someone stuck on something they hadn't solved yet, like "The no code wall is real and I am completely stuck on my database logic". For idea 2, I read all 105 and flagged 17. For idea 4, I read all 92 Reddit titles and flagged 13. That's why the other three ideas lost a point on question 2.

On day 1 I also wrote that 13 of idea 4's 92 posts were people giving the fix away free. That came from the same kind of script. Reading the titles, I can only stand behind at least five.

A title that shows signs of trouble is still a title. It doesn't tell me that person would pay for anything.

My favorite lost

Idea 4 was the one I wanted. It got a 0 on question 8. In the same year of Reddit posts, at least five showed a free way to fix how Claude writes, and two of those were about getting it to sound like you. It's hard to charge for something people are handing out with instructions.

What I'm picking, and what I'm not claiming

Idea 1: people who can't code, building real things with Claude, usually after work. That describes me, which is part of why it scored well on question 7. It's also a reason to be careful.

What I have is three strangers on X who wrote, in their own words, about being stuck on something they hadn't solved. One couldn't get Claude to use the skills they'd installed. One wanted to know what to check before publishing a web app. One said Claude was too slow for them to ship anything fast.

Three people, three different problems. I don't know if any of them can't code, and I'm not claiming they're the people I'd sell to.

Nobody outside my own head has asked me for anything yet.

The guess I'm testing

Here's the gap I keep running into. Something looks like it works, then someone else uses it and it doesn't.

My own example: on day 1 I found out my signup form had been saying thank you to broken email addresses that never made it onto my list. No error anywhere, on my side or theirs.

So the guess is that people who can't code would want a short check they can run before a stranger finds the problem for them. Something like ten minutes. Ten minutes is how long the check would take. It's not a promise that it finds anything.

That guess comes from my own mistakes. I haven't found a stranger stuck on this exact thing. I'm not building anything to sell this week. First I want to see whether anyone recognizes the problem.

The rule for day 7

Rule 3 of this challenge says no switching the idea halfway. Days 1 to 7 are for picking, so until day 7 this is a trial. I can switch once, and only before I build anything. From day 8 it's locked.

Here's what decides it. By Tuesday, September 29, which is day 7, I need two people I don't know to do something real: join the email list, or tell me where they're stuck on this. Two separate people, each with an email address or an X account I can see. One person doing both counts once. Anonymous link clicks don't count toward the two. I'll report them next to it, separately.

Fewer than two people, and I switch ideas and say so here.

The numbers today

Follow along

If you can't code and you've shipped something with Claude, I'd like to know how you check that it works when someone else uses it. Reply to me on X.

If you'd rather get these reports by email, here's the box. The free PDF version of guide one comes with it.