Curation by method is still curation

Part of the Far Wide Radio reservoir is organised into folders called unique/ and rare/. They are not hand-picked. A motif extraction pipeline produced them: it read each cube, found recurring figures, and sorted them by how unusual they were within the piece.

On the first production pass we treated those folders as raw material — output of a program, not choices — and reached past them for the files someone had marked by hand. The reasoning felt sound. Hand-marked files carry taste; algorithm output carries statistics.

That was wrong, and the correction is the useful part of this note.

The motif pipeline’s selection is the curation. Somebody wrote it. Somebody chose what counts as a motif, what window to look in, what “unique” means relative to “rare”, and where the thresholds sit. Those choices are taste — they have simply been frozen into an instrument instead of exercised one file at a time. The instrument that selected the motifs has taste encoded in it.

Once you accept that, the archive reads differently. The PREFERRED_ prefixes, the unique/ and rare/ splits, the numbered passes in the DAW session names: these are all curation fossils. They are the record of decisions already made, some by hand and some by machine, and programming the collection meant reading those fossils rather than inventing taste on top of them.

The practical result was the Field Series — one program per cube, structured directly by what the pipeline had already sorted: the unique suite first, then the modes field, then the rare suite as the close. That structure was not designed. It was recovered. It had been sitting in the folder layout for months.

The general version, which we suspect applies well outside audio: when you inherit an archive that someone organised, the organisation is data. Treating directory structure as mere plumbing throws away the most compressed record you have of what its author thought mattered.