Take a screenshot of a spreadsheet. You can capture every visible number, every label, every carefully chosen shade of green. Put the screenshot beside the spreadsheet and, for a moment, they can look identical. Then change one of the numbers. In the spreadsheet, the total updates. In the screenshot, you are editing a picture. Somewhere between the two, the relationship between the numbers disappeared.
That difference explains why there are so many file formats, why a few become familiar to almost everyone, and why conversion can be more consequential than choosing a different option in a menu. A file format is a set of rules for representing information: pixels, words, musical notes, measurements, or relationships between parts of a machine. The extension is only a label on the outside.
What are formats for?
Consider a black circle on a white background. A raster image describes a grid of coloured samples. An SVG can describe an instruction: draw a black circle with this centre and radius. Both can produce the same picture at one size, but they preserve different possibilities. The vector description keeps the geometry explicit; the raster description keeps the sampled appearance explicit. The format follows the work someone expects to do next.
Music makes the distinction sharper. A recording stores a representation of sound. A MIDI file stores performance instructions: which notes begin when, how long they last, and how they should be played. That lets a musician change tempo or move a note, but it does not preserve the particular sound of a particular performance. You may need the instructions, the recording, or both. The MIDI Association explains this distinction.
Medical imaging adds another requirement. A scan is not only a picture; it has an orientation, a position, and a relationship between pixels and physical distance. DICOM carries that information with the image data. A picture that looks right on a screen does not, by itself, tell you how many millimetres something measures. DICOM defines the image geometry.
Why do they keep multiplying?
Two pressures pull in opposite directions. A widely supported format becomes more valuable because more people and applications can open it. A specialist format appears when existing agreements cannot express something a group needs to preserve, calculate, or change. Those pressures coexist with corporate competition, historical accidents, and standards work. Even perfect cooperation would not make a musical note and a medical image require the same description.
The familiar formats are also biased by what reaches us. We receive a finished song, report, or photograph; the working file may stay with the people who made it. A project format used by a small team can be central to millions of public outputs. What looks like the long tail from your laptop may be the centre of someone else’s working day.
Why is “convert” an ambiguous verb?
Sometimes conversion preserves the information you care about. Sometimes the target format has nowhere to put it. Export a spreadsheet to CSV and you can retain displayed values in a table while losing formulas, formatting, charts, and relationships. The export can be excellent for moving data and a poor substitute for the workbook.
Even “lossless” needs a reference point. PNG compression can recover every input pixel exactly. A spreadsheet turned into a PNG can therefore be losslessly compressed while losing every formula before compression begins. Reconstruction can be useful without being recovery: an AI may infer a plausible formula from a screenshot, but where several originals fit the same picture, the picture cannot establish which one was used.
The expensive image that loads faster
Specialist formats often become interesting when their layout changes the cost of reading. A Cloud Optimized GeoTIFF stores an enormous image in tiles and adds smaller overviews. A viewer can request only the byte ranges needed for the current view over ordinary HTTP. The extra overview pixels cost roughly 1/3 of the original uncompressed pixels: 1/4 + 1/16 + 1/64 + … . A larger file can reduce transfer, computation, and waiting time. COG’s design is documented here, alongside HTTP range requests.
The same idea appears in columnar data formats. Apache Parquet arranges values so a reader can inspect only the columns and row groups relevant to a query, using statistics to skip data that cannot match. The format spends effort while writing so a later reader does less work. Parquet describes its file layout and encodings.
How many are there, and which matter?
There is no meaningful single census. The UK National Archives’ PRONOM registry is a carefully documented collection, not every private application format ever made. Versions, profiles, containers, and dialects make the boundary fuzzy, so a count is only useful when its rules are stated. PRONOM explains what its registry records.
“Important” can mean familiar, widely stored, essential to a profession, or likely to remain readable. A format can be obscure to the public and indispensable to a laboratory. Preservation therefore considers documentation, adoption, dependencies, and the availability of tools, not popularity alone. The Library of Congress lists these sustainability factors.
The useful question is not “what extension should I choose?” It is “what do I want the next person to be able to do?”
The short head of formats gives us convenient defaults. The long tail records the jobs that defaults cannot quite do. Keeping both a rich working file and a widely supported export is often a way of preserving different futures.