Housing test properties

Context

One property proves a path works once. It cannot separate a framework that generalises from a framework that happened to fit one niche, or from a builder who happened to know that niche.

StayRentals.com and HomeRule.com exist to make that separation. Both are housing verticals, both non-YMYL, and both were built through the same open-source framework as the flagship property. Two properties in adjacent but distinct verticals, built the same way, is the smallest arrangement that can show whether the framework carries the standard or whether the standard was coming from somewhere else.

They are described together on purpose. Separately they are two small sites; together with the flagship they are the argument.

Owned

Both properties end to end, as sole builder: brand, architecture, content system, quality gates, and deployment. Same role, same tooling, deliberately the same process.

Architecture

Shared with the flagship property, which is the experimental control: a Next.js static build, schema-first programmatic content, and the audit layer running as a publication gate rather than as a report.

Holding the architecture constant is what makes the comparison mean anything. If each property had been built the way that suited it, the fact that all three work would say nothing about the framework, because the variable under test would have changed along with everything else.

What did differ was the schema and the vocabulary. Housing has different entities than the flagship niche does, different page types that a reader expects, and different structured data. The framework had to absorb that without special cases, and the honest report is that the vertical-specific work lived in configuration and content schemas rather than in forked code.

The audit layer transferred without modification, which was the surprise worth recording. The classes of defect a generator produces turned out to be properties of generation rather than properties of the niche, so the gates that caught structural and linking problems on one property caught the same shapes on the others.

Decisions

• Two properties, not one more. One additional property would have doubled the evidence for effort; two in distinct verticals is what tests generalisation rather than repetition.
• Non-YMYL verticals, deliberately. The flagship carries the regulated-content case. These isolate the generalisation question from the compliance question.
• Architecture held constant. The point of a control is that it does not move.
• Vertical differences pushed into schema and configuration. Rejected forking the code per property, which would have made the framework three frameworks.

Outcome

Both properties are live and were built through the framework without forked code, and the audit layer moved across verticals unchanged.

That is the evidence the pair was built to produce. Generalisation shows up as the architecture holding constant and the quality gates transferring, which is a structural result rather than a traffic one.

Revision

I would have picked verticals that were further apart. Two housing verticals share more vocabulary than I expected, so the test was weaker than intended: some of what transferred may have transferred because the domains are adjacent rather than because the framework generalises. A property in an unrelated category would have been the harder and more honest test, which is what the next ones are for.