When a remediation pass on harmful catalogue language goes wrong, the cause is almost always one of three things: a global find-and-replace that ignored context, a failure to separate the creator's original title from the cataloguer's own description, or terminology chosen without the affected community. Fix those root causes and most downstream errors — broken citations, anachronistic titles, angry users — disappear. This guide works through the usual failure modes and the safe fix for each.

Why did my find-and-replace make the catalogue worse?

The classic error. Someone runs a sed or SQL REPLACE across the whole catalogue and discovers a week later that quoted historical titles now read as anachronisms, unrelated words got mangled, and a citation that referenced the old text no longer resolves.

The root cause is scope. A blanket replace cannot tell a slur in a cataloguer's note from the same string inside a creator-supplied title or a quoted contemporary source. Scope the operation instead:

-- WRONG: hits titles, notes, transcriptions, everything
UPDATE records SET description = REPLACE(description, 'old_term', 'new_term');

-- BETTER: restrict to the institutional-voice field, log each hit first
SELECT id, scope_content FROM records
WHERE field = 'scopecontent' AND scope_content LIKE '%old_term%';
-- review the result set by hand, then update only reviewed ids

Never replace inside a title or originaltitle field automatically; those are evidence.

How do I tell harmful description from a harmful original title?

This is the diagnostic that prevents most damage. Ask one question of every flagged record: who wrote these words?

• Creator's voice — the title or caption the photographer, publisher, or donor supplied. This is historical evidence. Keep it, wrap it in quotation marks, and add an explanatory note. Do not rewrite it.
• Institutional voice — the scope-and-content note, subject headings, or summary a cataloguer wrote. This is the library speaking, and the library can choose different words.

Mislabelling the two is the root cause behind both over-editing (rewriting evidence) and under-editing (leaving an offensive cataloguer note in place).

Why are controlled-vocabulary terms still offensive after I edited the free text?

Because the harm often lives in the subject headings, not the prose. Editing scopecontent while leaving an outdated LCSH or AAT term attached means search and faceting still surface the slur.

Symptom	Likely root cause	Fix
Slur reappears in facets	Controlled vocabulary term unchanged	Map old heading to a current term; keep old as deprecated
Term gone from page but found via search	Full-text index not rebuilt	Reindex after the edit
New term is also dated	Vocabulary itself is the problem	Adopt a community-authored list or local thesaurus
Edit reverted on next import	Upstream feed overwrites local fix	Apply remediation in a transform layer, not the record

That last row is a frequent silent failure: a nightly harvest from a union catalogue overwrites your careful edit. Apply changes in a transformation step the import respects, or flag the record as locally maintained.

How do I keep an audit trail so changes are defensible?

Every remediation should be reversible and explainable. Keep a change log with enough columns that a future archivist can reconstruct the decision.

id | field | old_term | new_term | reason | community_consulted | reviewer | date

Store the original term in a labelled legacy element rather than destroying it. In EAD you might use a dated maintenanceevent; in many CMSes a non-public note field works. The goal is that the catalogue records its own remediation history.

What if rewriting the record breaks existing citations?

Researchers cite the words they saw. If a thesis quotes a former scope note verbatim, silently rewriting it strips the quotation of its referent. The fix is layering, not overwriting: surface the remediated description to users, but retain the original — dated and quoted — in a legacy field that persistent links still resolve to. Pair this with a public reparative-description statement so anyone confused by a changed record can find your policy.

Who should choose the replacement terminology?

The most common well-intentioned failure is a lone cataloguer picking "better" words in isolation, then discovering the chosen term is itself unwelcome to the community it describes. Where language concerns a specific group, that group should lead or co-author the new terminology. Treat published reparative-description guides and community style guides as starting points, not final answers, and record who was consulted in the audit trail.

Key Takeaways

• Three root causes explain most failures: unscoped replace, conflating title with description, and community-free terminology.
• Never auto-replace inside creator-supplied titles; they are historical evidence.
• Separate the creator's voice (keep, quote) from the institution's voice (rewriteable).
• Remediate controlled vocabulary and subject headings, then rebuild the search index.
• Apply edits in a transform layer if an upstream feed would otherwise overwrite them.
• Layer remediated and original text so citations and persistent links still resolve.
• Let affected communities lead terminology decisions and log who was consulted.

Frequently Asked Questions

Should I delete the original offensive catalogue term?

Usually no. Deleting it erases the historical record of how the item was described and can break citations. The standard practice is to move the original term into a quoted, clearly labelled legacy field and surface a remediated term to users.

Why did my find-and-replace make the catalogue worse?

Blanket replacement strips context, hits false positives, and creates anachronisms in quoted titles. Always scope replacements to description fields, exclude transcribed titles, and review matches by hand before committing.

How do I tell harmful description from a harmful original title?

A title supplied by the creator is evidence and stays as-is in quotation marks; a description written by a cataloguer is institutional voice and can be rewritten. Separating these two is the single most important diagnostic step.

What is a content warning versus remediation?

A content warning flags difficult material for users without changing the record; remediation edits the description itself. Mature programmes use both, plus a public statement explaining the policy.

How do I keep an audit trail of changes?

Version every change in a spreadsheet or repository with the old term, new term, field, date, and reviewer. Many systems also let you stamp a maintenance note in the record itself.

Who should decide on replacement terminology?

Where the language concerns a specific community, that community should lead or co-author the new terms. Lone cataloguer judgement is a common root cause of well-meant but inappropriate fixes.

Related reading

• Recognise bias in archival data
• Decolonise collections description
• Address representational bias in datasets
• Apply CARE principles for Indigenous data
• Write a scope and content note