How to read the Lab

Everything here is one of three things. We label which, every time.

Label	What it means	What it does NOT mean
Observed	A pattern seen on real product pages in our library of 715 tagged examples	That the pattern works, or that the brand tested it
Hypothesis	An experiment we built, with what we expect to change, for which buyer, measured how	That it will win. Most experiments don't.
Result	An A/B test on real store traffic, with sample size, duration and confidence	That it will win on your store, audience or price point

Experiments

Built on Duskwell, a fictional brand, so no real brand is imitated. Its prices, reviews and numbers are illustrative.
Each experiment states its hypothesis, the test we'd run and the metric we'd watch.
Each carries a stage: Sketch → Prototype → Field test → Proven → Installed. "Prototype" means built and working, not proven.
Copy follows supplement claim rules: structure/function language only ("supports"), never "cures, treats or prevents".

715 examples from real PDPs, each tagged by component, pattern, persuasion lever, buyer type, the objection it answers, and a quality score.
Tags are our judgement. Quality scores rate the execution, not the brand.
Brands are named only where the brand is clearly visible. Screenshots are shown as commentary on DTC Audit, never as our own work.
A pattern being common doesn't make it good, and a pattern being rare doesn't make it bad. Rare usually means untested.

Real tests from real stores, published only with the client's written permission or fully anonymised (no brand, no identifying screenshots, rounded numbers).
We publish losing and flat tests too. A result is only as good as its sample, duration and context, so we show all three.

Redesigns of real brands are labelled "Unofficial concept. Not affiliated with [brand]." They have no working checkout and use rewritten copy.
If you're a brand and want a concept or screenshot taken down, email us and it's gone within 48 hours.