Skip to Content
vnext · upcoming Unreleased Read the changelog →
Blog9 signups in 5 months

I shipped a bug tool for 5 months and got 9 signups. Here’s what the data said.

Sep 21, 2026 · Kenji Sakuramoto


The repo went public on 16 April 2026. On 20 September I ran the queries I should have been running every Monday. Here is what came back, and then what I think it means.

The numbers

  • 9 console signups, all time. Three in April, five in May, one in June, zero in July, August and September.
  • 4 external projects created by those signups. 0 of them ever sent a report. The API keys were minted on day one and never seen again.
  • 16 bug reports in the database, all from my own apps.
  • 26,000 end-user sessions, all on my own apps.
  • 3 GitHub stars.
  • 100 repository views from 25 unique visitors in the last 14 days.
  • 0 issues from anyone but me. 1 external pull request.

Against that: 30 published npm packages, and an MCP server listed in the official registry, Glama, mcp.so, Smithery, cursor.directory and awesome-mcp-servers.

I had built every surface a launch is supposed to land on. Nothing had landed on them.

Three things I found when I looked

The launch never ran. In June I wrote a five-day launch runbook: Show HN on Tuesday, Product Hunt on Wednesday, Reddit across the week, a retro on Friday. The five features it was built around shipped on schedule. The posts did not. There is no Show HN, no Product Hunt listing and no Reddit thread for this project anywhere. The public launch page said otherwise until this week; it now says this.

GitHub OAuth was off the whole time. The console has a “Continue with GitHub” button for a developer audience. The provider was never enabled in the auth settings of the live project, so the button never rendered, and every one of the nine signups used email and password. I found out by reading the live auth config, not from any dashboard.

The console was not measuring itself. Mushi’s own console is supposed to run the Mushi SDK, so that “did the owner open the diagnosis” is a number. The deployed console had never been given its own SDK key. The build pipeline did not pass it. Dogfooding was silently off, and every funnel cell between “signed up” and “paid” was blank.

What the data said

Put together, the three findings are one sentence: I built distribution surfaces, never ran the distribution, and could not see the funnel. The registries and packages were real work, and I would do them again. But they are where people arrive, not why they come, and without measurement I could not tell the difference between “nobody came” and “people came and left”.

The four external projects are the part that stings. Four strangers signed up, created a project and got a key. None of them installed the SDK, or if they did, it never sent anything. I do not know which, because the step between “key minted” and “first report” was not instrumented.

What changed

One number. The goal is now activated external projects per week: a project owned by someone who is not me receives its first real bug report, and the owner opens the diagnosis. Not signups (nine people did that and none came back), not stars, not downloads.

Measured with our own SDK. The SDK grew a track() call and the console grew a Users & Funnels page, and Mushi’s own landing, docs and console emit events through it. The company funnel is a query now. Zero is a value it returns; “unmeasured” is not.

First run rebuilt. A new signup should see a diagnosis of a realistic report in under a minute, before installing anything, and then be walked to the install. That is the step the four external projects fell through, so it is the step that got rebuilt first.

What happens next

A Show HN, once a public gate list is true: the demo loads with no signup in a private window, a fresh app reaches a diagnosis in under five minutes on camera, the docs search returns no dead links, the legal pages exist, and the activation query returns a number. The launch page lists the gate so anyone can check it.

Whatever the numbers are afterwards, they go in the next post. If you have shipped an app mostly written by an AI editor and hear about bugs from users, I would like to know what you actually do with those messages. The Discussions board  is the place; I reply on weekdays.

Last updated on