An open notebook on how the web works.
BusinessIndex Lab is the research arm of the agency. We study how the best sites work, abstract their patterns into a manipulable graph, and apply what we learn to every site we build. What follows is a page from the notebook.
Pattern languages were Christopher Alexander's 1977 insight: design wisdom is most useful when named, connected, and addressable. The pieces have existed for fifty years. What didn't exist until now was a system that could study real-world sites as a daily practice, name the patterns, draw the edges, and run experiments on the graph.
AI labor changes what a pattern language can be.
Three stages on a closed loop.
Observe
We study how real-world sites work — what they do, when patterns repeat, which prior opinions they support or contradict. A five-tier maturity ladder governs how observations earn confidence. The discipline runs daily.
Abstract
Recurring findings get extracted into named pattern records — each with intent, forces, preconditions, consequences, and typed relations to other patterns. Compositions name the sequences. Formal operations run on the graph to surface hypotheses for the wild to verify.
Apply
The build layer queries the catalog by intent and force rather than reusing a template. cafe-section-funnel goes into a Korean cafe homepage; freemium-radical-honesty-cta goes into an indie SaaS landing. Sites we ship become evidence; patterns earn promotion or revision from how they actually perform in the wild. The application stage closes the loop.
What the engine makes.
Four worked examples across categories. Designed by the engine, not deployed for clients. A fair preview of what we'd build for you.
Including the parts that don't work yet.
The schema is at v0.1.1. Seventeen open candidates for the next version. Two hypotheses rejected last cycle. One relation has zero entries because its semantics don't fit the evidence. We surface this because labs that hide failure modes are doing brand management, not research.
Structured records — six kinds, six typed relations, a five-tier confidence ladder.
Combine, invert, substitute, exaggerate, contrast — plus three composition-level moves.
A pattern earns its record by passing five tests: intent, recurrence, boundary, substitution, forces.
Seventeen open candidates for the next schema version — places where the current shape doesn't quite hold what we've observed.
Predictions that didn't hold up against real-world sites. Archived with rationale rather than quietly removed — failures are data.
The research arm of BusinessIndex. An open notebook on how the web works.