Methodology

How a photograph becomes a record.

Cataloguing is where an archive earns trust or loses it. Here is the whole pipeline, including the part most archives are quiet about: the line between what we assert and what we merely propose.

From file to record

  • Ingest. A master is captured at archival quality and vaulted in cold storage. Public previews and licensing derivatives are generated from it — the master itself is never the thing on the page.
  • Provenance. We record what's actually known about the collection — era, place, maker, how it reached us — at the level it's known. "Various" and "unknown" are valid, recorded answers, not blanks to be filled with confidence later.
  • Enrichment. Where stored facts exist, the record asserts them, scaled to how sure we are. Where they're thin, the writing shifts to what can be seen rather than claimed.
  • Review & publish. Nothing is public until it's reviewed. Withdrawal leaves a visible tombstone, not a silent deletion.

Asserted vs. proposed

This is the rule the rest of it hangs on. A machine-suggested identification — a name, a place, a date the model "recognizes" — is never asserted as fact. It is routed to a proposed — verify state and labelled as such until a human confirms it. Stored, sourced facts are stated plainly. Guesses are dressed as guesses. We fabricate nothing to fill a gap; the gap is part of the record.

Dates, eras, and citations

Where a single year isn't known, the record carries an era span derived from what the works themselves support, rather than a falsely precise date. Every record generates a citation in three scholar-ready formats — see Citations.