How we test

Every guide on this site carries a label saying what kind of evidence sits behind it. There are three labels, there is no fourth, and there is no default. A page with no label renders no label rather than implying we tested something.

The three evidence labels

Hands-on tested

We used the gear ourselves. Photos, measurements or video back it up. This label is rare on purpose. It applies to the Breville Barista Express, Barista Pro, Infuser and Smart Grinder Pro, the AeroPress Go, a Dolce Gusto machine, and the coffee subscriptions from Driftaway, Atlas and Lifeboost.

Editorially researched

Specifications, verified owner reports and our own experience in the category. No machine on our counter. Most buying guides here sit at this level, and they say so in the byline rather than in a footnote nobody reads.

Refreshed from verified specs

Existing guidance rechecked against current manufacturer specifications and current availability. Used when a guide was sound but the market moved underneath it.

The label is enforced, not remembered

This is the part that matters. Trust signals fail when they depend on a writer remembering to add them, so on this site they do not depend on that.

The theme reads the evidence label off the page and filters the page’s own text against it. A guide marked editorially researched cannot contain a sentence claiming we pulled shots on the machine, because the filter strips that phrasing before the page renders. There are 16 automated tests covering that filter alone, including the awkward cases where the claim is buried in a nested block.

The same applies to affiliate links and disclosure. Every outbound commercial link is forced to rel="sponsored nofollow" at render time, and the disclosure is injected before the first call to action if the writer did not put one there. It is not possible to publish a page here that earns us money without saying so.

How a buying guide gets built

1. Availability first. We check the live retail catalogue and record the date. A product that is out of stock or discontinued does not make a list, however good it is. The date sits in the byline as a last-verified stamp, which is a separate field from the WordPress modified date and does not move when we fix a typo.
2. Criteria, weighted and published. Each guide states what it is judging and how much each factor counts. On the espresso machine guide, for example, what ships in the portafilter carries 35 percent, temperature control 25, serviceability 20, owner evidence 20. You can disagree with the weighting, which is the point of publishing it.
3. One sorting criterion that actually decides. Most categories have a single question that separates the real products from the ones that only photograph well. For espresso machines it is whether the machine ships with a pressurized basket. We identify that question per category and sort on it rather than averaging twenty specifications into mush.
4. Owner evidence. Verified purchase reports, read for failure patterns rather than star averages. A four-star product with a recurring gasket complaint is worse than a four-star product with scattered shipping complaints.
5. Missing figures stay missing. Where a manufacturer does not publish a number, we record that it is unpublished. We do not estimate it and present the estimate as a spec.
6. Review before publish. Chea Hong (Caden) Lee reads the guide against its sources and flags overstatement. Named in the byline when it happened.

Scores

Scores run to one decimal and come out of the weighted criteria published on each page. They are comparable within a category and meaningless across categories. A 8.7 grinder and a 8.7 espresso machine were judged on different things.

Video attribution

Coffee sites routinely embed other people’s YouTube videos and let their schema imply they made them. We had that problem too, at scale, and it is now fixed structurally.

The site keeps a registry recording the real owning channel for every video ID it embeds, resolved from YouTube itself. Schema, the video sitemap and the on-page video modules all read from that registry, and nothing else is allowed to decide whether a video is ours. Of the 484 videos embedded across this site, 30 are Tim’s. The video sitemap now lists those 30 and nothing else.

When we get it wrong

We log the correction on the page itself with the date, and we leave the record of what was wrong. Pages that have been materially changed carry a changelog. If you spot an error, tell us and we would rather fix it than argue about it.