Wirly.com

Context

Wirly is the property that had to answer a question about the tooling rather than about a market: can a one-line brand idea reach a launched, ranking site through the open-source framework, with a single person driving it?

Anything short of a real launch would not have answered it. A demo site proves the generator runs. A live site in a competitive niche proves the whole path holds: naming, architecture, content, quality gates, deployment, and the audits that catch what the generator got wrong.

The niche is a YMYL one, which raised the bar rather than the difficulty. Pages that touch a reader's health or money have to be right, so the quality layer could not be the part that got cut to make a launch date.

Owned

Everything. Brand, information architecture, stack, content system, quality gates, deploy, and the decision about what was good enough to publish. Sole builder, with agents doing drafting work under briefs I wrote and reviewed.

Architecture

Next.js 15 with a static build, Supabase for the data that had to be queryable, Vercel for hosting, Cloudflare for DNS and edge. The same shape as this site, for the same reason: a static site has almost no runtime surface to get wrong, and the parts that need to be dynamic can be named explicitly rather than by default.

Content was generated programmatically against a schema, not written page by page. That is the only way one person reaches this page count, and it is also the failure mode: a generator applied to a YMYL niche will produce fluent, confident, wrong pages unless something is standing in the way.

What stood in the way was an audit layer running as a gate rather than as a report. Audits covered structure, internal linking, schema validity, accessibility, and content quality, and a failure blocked publication rather than filing a ticket. The distinction is the whole point. An audit that produces a report produces a backlog; an audit that blocks produces a standard.

The audit layer earned its place. It caught structural problems the generator introduced at scale, which is exactly the class of defect that is invisible on a page and obvious across a thousand of them, and the sort a human spot-check does not find because a spot-check reads one page well.

Decisions

• A real launch in a competitive niche over a safe demonstration. A demo would have proved the generator runs and nothing about whether the path holds.
• Programmatic generation with a blocking audit layer. Generation without gates in a YMYL niche is the irresponsible version of this project.
• Static build over a dynamic application. Fewer moving parts at run time, and the dynamic parts have to be argued for.
• Schema-first content. Structure before prose, so the generator produces something checkable rather than something that merely reads well.

Outcome

Launched March 2026 and tracking 2,800+ Semrush-tracked keywords^{Source note 1} within five months, in a competitive YMYL niche.

Read that as a statement about the path, not about the property. The keyword count is evidence that the framework produced a site search engines treated as legitimate, from a standing start, at one person's hand.

Revision

I would have run the audit layer against a deliberately broken corpus before pointing it at the real one. It caught real problems, but I could not have told you in advance which classes of defect it was blind to, and seeding known faults is how you find out. I would also have frozen the content schema later than I did: the early version encoded assumptions about the niche that the first audit pass then argued with.