What our engine got wrong, and what it got right

Method and limits

How a case is scored

Each case is a pair of names an office has ruled on: the name that was applied for, and the earlier mark. We run the engine on the pair. A known conflict counts as caught if the engine rates it Some similarity or High similarity. A control, where the office found no likely confusion, counts as correct if the engine rates it No close match.

Each result comes with a 95% interval (Wilson score interval), computed by script from the counts on this page. With small samples the interval is wide, and it is the honest number to read: for example, 16 caught out of 16 is consistent with a true catch rate anywhere from about 81% to 100%.

What this test does not show yet

• The development set was used to build the engine. Rules were added after we saw some of these cases, so it cannot tell you how the engine does on names it has never seen. That is what the independent holdout test is for.
• Every development case so far comes from the US Trademark Trial and Appeal Board. The EU, UK, WIPO and Brazil registers have no test cases yet.
• The test compares each pair of names directly. It does not yet test the full search (finding the earlier mark among millions of records, or checking goods and services).
• Some links go to reputable summaries of a decision rather than the office's own record. Each link is labelled.
• An office's ruling depends on goods, services, evidence and more than the names alone. Our engine looks at names, so Some similarity means "worth a lawyer's look", not a prediction of the ruling.

Each case is listed to measure our engine, not to comment on any applicant or mark owner. If you see an error, write to hello@lightfathom.com and we will correct it under our corrections policy.

Preliminary check, not legal advice. Lightfathom is a search and research tool, not a law firm.