Formats multiply when a real workflow asks a representation to do something its predecessor was not designed to do. A camera wants a compact record of sampled light. A map reader wants to fetch one neighbourhood without downloading a continent. A data analyst wants to skip columns and row groups that cannot affect a query. A musician wants notes, timing, and instrument instructions to stay editable. These are different jobs, even when the final screen can look similar.
What pressures create a new format?
Performance is one pressure. Cloud Optimized GeoTIFF arranges image tiles and overviews so an HTTP client can request selected byte ranges. Semantics are another: a medical image needs spatial metadata that a generic photograph does not. Ownership and coordination matter too. PNG emerged as a patent-free alternative to GIF, while Office Open XML packages a document family inside a ZIP-based structure.
Is every variant a new format?
There is no universal boundary. A version, profile, codec, container, or private extension may be counted separately by one registry and together by another. That is why format inventories state their scope. A version can change what a reader may assume; a profile can restrict a broad standard for a particular industry; a codec can change how data is compressed inside a container.
What does a new format cost?
A new agreement creates work beyond writing a parser. Publishers need exporters, readers need tests, archives need identification rules, and people need a migration path. A technically elegant format can remain marginal if it arrives without tools that preserve the surrounding workflow. Conversely, a clumsy format can persist because every institution already knows how to process it.
Formats also multiply because migration is expensive. A specialist team may keep its old format beside a newer export because conversion loses annotations, calibration, or editability. Standards reduce accidental incompatibility, but they do not remove specialization. The long tail is the visible record of people choosing a representation that makes their next operation possible or cheap.
The result is less like a neat family tree than a city map: main roads carry common traffic, while service lanes survive because one building still needs them. A new standard may connect districts without removing every old route.