All things considered, the USPTO patent database is well curated. Millions of patent documents (issued patents and application publications) are available for search and download. The documents include the complete application (title, abstract, specification, and claims), along with various important dates (filing, publication, and issue dates) and lists of references submitted or cited during prosecution. Not bad.

The data, however, is not flawless, and the more searches we do, the more frequently we find errors. For example, Dorothy returned a patent titled “Remotely Aligned Surgical Drill Guide,” while performing a search for waterproof textiles. I initially thought this was a bad reference, despite what appeared to be relatively good results overall. Some of our synonyms are a little weird. Dorothy thinks “sausage” is a synonym for “candle,” for instance. Corrections will be in place next week.

DorothyAI's head of quality assurance reviews results "by hand."

Nonetheless, it’s important to understand why a surgical drill appears in search results for waterproof textiles. I read the abstract:

A method is proposed for the application of a finishing layer to a textile support material. By means of the novel method, a water repellent or oil repellent layer, a so-called finishing layer, is applied to a textile support material selected from the group of fibers, tissues, and fabrics. The water repellent or oil repellent finishing layer comprises at least two water repellent or oil repellent components wherein a first component comprises one or more dispersants and a second component comprises one or more dispersed phases or colloids, and wherein the dispersant and the dispersed phase are present in the gel state...

Clearly relevant!

Why is a patent directed to a textile finishing layer titled “Remotely Aligned Surgical Drill Guide”? Jay (DorothyAI’s CTO) immediately assumed it was an extreme example of gamesmanship. I don’t completely discount the gamesmanship explanation, but in my opinion, it’s more likely a clerical error.

The application was filed in 2001, when most patent applications were still filed by mail (others were hand delivered). The USPTO’s electronic filing system (EFS) did not go on-line until October 1, 2004. When you’re dealing with millions of pages of filing documents every day, papers occasionally get shuffled or numbers inputted by hand get transposed. Numerous things could have happened that caused the entwinement of unrelated families of patents titled, “Finishing of Textile Fibers, Tissues, and Fabrics” and “Remotely Aligned Surgical Drill Guide.” Whatever the reason, this error is present in our database, Google’s database, the USPTO database, and others.

The moral of this story: You can’t judge a patent by its title. You can’t judge a patent by its abstract either, by the way. As I explained to my intern, the specification is important. You should know what’s in there. That’s why we’re putting so much effort into searching the entire patent publication and identifying “key phrases.”