Written for humans, read by machines
Nearly everything the world publishes is written for a person to read. News is written for readers. Filings are written for regulators. Registries, schedules, menus, and notices all assume the same thing: a pair of eyes, a little patience, and one page at a time.
For most of history that assumption was safe. The reader was always a person, so the format never had to be more than legible. Prose, tables, documents, and pages did the job, because eyes and patience were the only interface the record needed.
That assumption is no longer true. The largest reader of the public record today is software. Applications check it, models train on it, and agents act on it, continuously and at a scale no human readership ever approached. Machines do not skim. They query.
And they read our formats badly. Software that meets human formats has to parse prose, guess at tables, and reconcile the same fact told a hundred slightly different ways, and then it watches its hard-won copy of the world begin aging the moment it is assembled. Every team that needs current information ends up rebuilding the same brittle pipeline, one source at a time.
A fact you cannot query is a fact you do not have.
The gap is structural. Publishing optimizes for the writer and the human reader: expressive, irregular, wrapped in context. Software needs the opposite: the same shape every time, current values, explicit fields, and a way to ask precise questions. Neither side is wrong. They speak different languages, and the distance between them grows with every model and agent that comes online.
Structured data is the translation layer. The work is not glamorous: reading millions of sources, resolving the conflicts between them, agreeing on a shape, and keeping the whole thing current every minute of every day. When it is done well, nobody notices. Things downstream simply work.
It also compounds. Once one slice of the record is structured, everything built on it gets simpler, and the next slice gets easier to reason about. News, places, markets: each is a different dialect of the same problem.
That is the thesis Relative is built on. We back the teams doing this translation well, and we intend to keep at it for a long time.