The testbed: a directory portfolio

The first thing the company runs is a business I built by hand and understand completely: a portfolio of niche directory sites for underserved Polish verticals. I chose it on purpose. It is low stakes, it has real complexity, and I own all of it. If the factory cannot run a directory portfolio I know top to bottom, it has no business running anything larger. That test is why this project exists in this section, and it is why this page is honest about being mid-experiment rather than dressed up as a win.

What it is, plainly

Niche directories for verticals the big players ignore. One vertical is live: funeral homes, in its monitoring phase. Several more are queued behind it, selected against a scoring rubric. The whole thing runs on a seven-phase playbook, from picking the niche through data, build, launch, and growth, and the playbook is a compounding document: each vertical updates it, so the next one is faster.

Underneath sit the scripts that will become the departments. A data pipeline that crawls and extracts provider records. A content step that generates the directory pages and runs them past review. An outreach step that emails providers to claim and correct their listings. A growth loop that proposes a change, measures it against one number, and keeps it only if it improved. All of this exists and runs today. I run it.

What the factory changes

Nothing about the work, everything about who holds the loop. Today I am the thread between the phases: I run the extraction, read the result, decide it is good enough, carry it to content, carry that to outreach, check the numbers, decide what to change. The factory moves that thread out of my head and into the wiki, and runs it on a schedule. The scripts do not get smarter. I stop being the part that carries results between them.

The one number per phase

The playbook already insists on a discipline that maps cleanly onto the growth department: test, do not guess, and validate against real signal before investing more. Every phase answers to one number. In launch it is indexation and early traffic, whether the pages exist to search engines and whether anyone arrives. In growth it is provider engagement, whether the people listed claim and correct their entries. The growth department's entire job is to move the number that the current phase is about, one change at a time, keeping only what measurably helped. I am stating what those numbers are. I am deliberately not stating values for them, because the loop has not run autonomously yet and I will not invent the readings before they exist.

What I am watching for

I can be specific about the failures I expect, because they are the reasons the safeguards and the verify steps exist. Stale state in the wiki, read as current truth, sending a department off a picture of the world that is a week old. An outreach run that emails the wrong list. A generated page that passes the style check and still reads like a machine wrote it. A deploy that reported success and did not actually go live. Each of these is a case the company is built to catch, and the only way to learn whether it does is to run it and see. The interesting result is not that the system worked. It is which of these broke, and whether the gate caught it before I did.

The honest status

This is the testbed in mid-build. The funeral vertical is live and proves the work holds up by hand. The factory operating that work on its own is what I am turning on now. This page is the setup and the test, not the result. It gets its numbers and its post-mortem once the loop has run real cycles, and I would rather leave that section empty and dated than fill it with a plan wearing the costume of an outcome. When there is evidence, it goes here. Until then, this is a promise I have not yet kept, marked as one.