How the Briefing knows what it says

Source. Patent data comes from the European Patent Office’s Open Patent Services, which serves its worldwide bibliographic database and its database of events recorded by patent offices. Records are read through that service only. Espacenet is linked from every family so that a reader can check the record, its claims and its description; it is never read by a program.

Discovery. The world’s patents are not crawled. Three kinds of narrow search are run, and they overlap on purpose: everything filed by 26 organisations the site profiles; the classification groups that are about the field by definition (computer input from nervous-system activity, invasive electrodes for brain recording, brain cortex electrodes); and the field’s wording together with the wording of its hardware, such as packaging, telemetry or insertion with a neural subject. Recall comes from the overlap.

Who filed. A name in a register is not an identity. Each organisation is searched under the names offices hold for it, including transliterations and earlier names, and a family is linked to it only when the applicant’s name matches a pattern that was checked by hand. 8 organisations are deliberately not searched by name, among them Merge Labs and Araya, because records under their name belong to someone else, or because their portfolios are almost entirely outside the field. Applicants are shown as recorded; an applicant is not always the present owner.

What is counted. Every family found passes the same published rules, whichever search found it. Independent signals add up: the field named in the title or abstract, a class that is about the field, a neural subject together with a part of the stack, a profiled applicant, a citation of a counted family. A family is counted when they agree. Neural technology that is plainly something else, such as a stimulator that reads nothing, a hearing implant or a sleep monitor, is kept as adjacent and not counted. What the rules cannot settle is left uncounted, and goes to an editor where it belongs to a profiled organisation. Of the families held now, 7,312 are counted, 951 are adjacent and 901 are uncertain.

The unit. One invention is published many times: as an application and as a grant, in several offices. All of those are one family, and a family is counted once, in the year of its earliest priority, which is when it was first filed anywhere. Publication counts are never used as a measure.

Layers and themes. A family is placed in the 11 layers and 24 themes by rules that read its title, its abstract and its classification codes. A theme needs the wording in the title, a class, or wording the abstract returns to; one passing word is not enough. Themes say what a filing is about. They are not a reading of its claims, and the claims are what define legal scope.

Claims. Where an editor has read a family’s claims and written what they appear to address, that note is marked as an editorial interpretation and appears only after review. No claim interpretation is generated and published automatically, and none states what a patent covers in law. Conclusions about competition or freedom to operate are not drawn at all.

Status. A family is shown as granted when an office published a grant for one of its members, read from the kind of publication, and only for offices where that reading is unambiguous. Granted does not mean in force. Events offices recorded, such as lapses, withdrawals and changes of owner, are shown in the offices’ own words on a family’s page where they have been read, and are not interpreted.

Limits. Applications are published about eighteen months after they are first filed, so the latest 3 priority years are incomplete and are drawn apart in every chart. Discovery is targeted and will miss families that use neither the field’s classes nor its wording. Utility models and filings made only in one country are less completely indexed than international ones. Names are matched conservatively, so a portfolio can be undercounted where an organisation files under a name not yet known here. Patent counts measure filing activity. They are not a measure of technical quality, of the value of a portfolio or of commercial success.

The patent landscape