Category Revision Log¶
Record — instantiates Implicit Bias in Knowledge Structure
A durable record of each category change — what changed, why, which tradeoffs remain, and how continuity with older data is preserved.
A Category Revision Log is the durable memory of a knowledge structure's changes. When a category is split, merged, renamed, redefined, or retired, the log records the change as an entry: what changed, the evidence and reasoning behind it, the tradeoffs that remain unresolved, and — critically — the crosswalk that keeps old records comparable to new ones. Its defining function is continuity under revision. Every other mechanism in the family finds bias or decides to fix it; the log is the one that makes a fix auditable and reversible over time and that stops a well-meant revision from silently breaking years of history. It is an artifact, not a meeting: its job is to ensure that a year from now, someone can reconstruct exactly why a category is shaped the way it is and how to read data recorded under its predecessor.
Example¶
A museum is revising the cataloging terminology for its ethnographic collection, replacing outmoded and offensive object and culture labels inherited from a century-old accession system. The revision is right, but the collection's every loan record, exhibition history, and scholarly citation references objects by their old catalog terms. A Category Revision Log governs the change so it does not orphan that history. Each entry records the specifics: the term "primitive art," retired and replaced with named cultural attributions; the reason (the label encoded a discredited hierarchy and misrepresented the works); and the residual tradeoff (the new attributions are more accurate but less searchable by researchers who only know the old term).
The load-bearing part is the crosswalk. For every retired term the log stores a mapping to its replacement(s), a note where one old term splits into several new ones, and an archival label preserving the original as historical metadata rather than deleting it. A researcher querying a decades-old exhibition catalog can now follow the crosswalk forward to the current record and back to the historical term. The log does not decide whether to rename — that judgment came from other mechanisms — and it convenes no one. It guarantees that the rename, once made, is documented, honest about what it still compresses, and continuous with everything recorded before it.
How it works¶
The log is defined by what each entry must contain and how continuity is preserved:
- One entry per change. Record the specific structural change (split, merge, rename, redefine, retire), not a vague "improved categories."
- Capture reasoning and evidence. Store why the change was made and what evidence drove it, so a future reader can evaluate the decision rather than just observe it.
- Note the residual tradeoff. Record what the new structure still compresses or worsens, so the revision does not present itself as finally neutral.
- Build the crosswalk. For every changed category, map old to new — including one-to-many and many-to-one splits — and preserve retired labels as archival metadata so historical data stays readable.
Tuning parameters¶
- Entry granularity — log every micro-edit or only material category changes. Fine granularity is fully auditable but heavy; coarse granularity is maintainable but can lose the change that later matters.
- Crosswalk fidelity — a simple one-to-one map versus full many-to-many mappings with effective dates. Higher fidelity preserves comparability through complex splits but is costly to build and maintain.
- Reasoning depth — a one-line rationale versus the full evidence chain. Deeper records survive staff turnover and audits; shallow ones are faster but decay into "because we changed it."
- Retention and access — how long entries live and who can read them. Long, open retention maximizes accountability; it also exposes the messy history some maintainers would rather bury.
When it helps, and when it misleads¶
Its strength is defeating two of the archetype's named failures at once: legacy comparability loss, by carrying crosswalks and archival labels so a revision doesn't break older reporting and search;[n1] and false-neutrality restoration, by keeping each revision's residual tradeoffs on the record instead of letting the new structure declare itself finally unbiased. It is the institution's memory of why its categories are shaped as they are.
Its failure mode is log-as-alibi: a meticulously maintained record of changes that documents everything and improves nothing, where the existence of a tidy changelog is mistaken for good governance. A subtler misuse is a crosswalk that maps labels but silently drops the cases that fit neither old nor new cleanly, quietly erasing exactly the misfits a revision was supposed to serve. The guarding discipline is to treat the log as a continuity guarantee, not a compliance trophy: every entry's crosswalk must be tested against real historical records (including the awkward ones), and the residual-tradeoff note must name what is still compressed, so the log records honest history rather than a flattering one.
How it implements the components¶
The log fills the record-and-continuity components — the ones that make a change durable and honest over time:
revision_proposal— it records each specific structural change (split, merge, rename, redefine, retire) with the reasoning and evidence behind it, as the canonical account of what was done.tradeoff_note— every entry states what the revised structure still compresses or worsens, preventing a revision from posing as finally neutral.legacy_category_history— its crosswalks and archival labels preserve comparability with older data, so a revision doesn't orphan the history recorded under the previous categories.
It does not decide changes with maintainers and affected users or capture their live disagreement (downstream_consequence_trace, dissent_record) — that adjudication is stakeholder_category_review.md — and it does not gather the misfit cases a revision should serve (edge_case_library), which is excluded_case_sampling.md.
Related¶
- Instantiates: Implicit Bias in Knowledge Structure — the log is the archetype's durable memory, keeping revisions auditable and continuous.
- Consumes: Stakeholder Category Review produces the adjudicated changes the log records and makes continuous.
- Sibling mechanisms: Bias Review Checklist · Taxonomy Bias Audit · Boundary Critique Session · Category Impact Assessment · Excluded Case Sampling · Inclusive Classification Review · Ontology Critique Workshop · Red-Team Schema Review · Stakeholder Category Review
Editorial Notes¶
Form Classification¶
Form family: Record, Log & Register
Rationale: A durable record of each category change — what changed, why, which tradeoffs remain, and how continuity with older data is preserved, making its operative form a durable record, ledger, register, or trace whose value depends on preserving actual state or history.
Independent corroboration: The frozen evidence defines Category Revision Log as 'A durable record of each category change — what changed, why, which tradeoffs remain, and how continuity with older data is preserved', so its operative form is Record, Log & Register.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Library & Information Science
Origin pattern: Convergent development
Present-day reach: Multi-domain
Rationale: Classification and authority-control practice established revision histories and crosswalks that preserve meaning and retrieval across splits, merges, and renamed categories.
Related originating lineages:
- Data Science & Analytics — Schema migration contributes machine-readable version mappings and downstream compatibility checks.
- Statistics & Experimental Design — Official statistics uses code crosswalks to preserve longitudinal comparability across revised occupational or industry classifications.
Review resolution: Library and information science is the agreed primary lineage because controlled vocabularies and cataloging practice require durable revision histories and crosswalks. Data schema evolution and longitudinal statistical comparability independently shape the continuity contract, so convergent is more accurate than one exclusive lineage.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] A crosswalk is a documented mapping between two classification schemes (or two versions of one), specifying how each category in the old translates to the new — including where one splits into several or several merge into one. It is the standard instrument for preserving longitudinal comparability across a revision, as when statistical agencies publish crosswalks between successive occupation- or industry-code editions. ↩