Archively Research Note 01
Figure 4 of 4, from When Does an AI Proposal Become Archival Description?, placed in section 7, What proof looks like.
Figure 4
Evidence ledger for one accepted descriptive statement
What a committee can inspect: the wording as proposed, the phrase a reviewer changed, the evidence available at the time, and what happened after publication.
Constructed worked example. The repository, record, people, dates, place names and correction event are fictional: what it illustrates is the evidence a decision leaves behind, not an institutional case. The categories are not new. Sources used, rules applied, the agent responsible and the revision date are description control under DACS 8.1, which names generative artificial intelligence systems, alongside people, as creators or revisers of a description. The description, the describing activity and the agent responsible are entities in their own right in Records in Contexts. Review by a person before acceptance, with machine output treated as a non-authoritative suggestion, is common to the AI Preparedness Guidelines for Archivists by Colavizza and Jaillant, published by the Archives and Records Association, and to the Program for Cooperative Cataloging task group report on artificial intelligence and machine learning. Documented roles and responsibilities, post-deployment monitoring and a route to appeal and override come from the NIST risk management framework, which is cross-sector rather than archival. What changes is that the describing agent is a machine proposal and an institutional acceptance rather than a single act of authorship.
The repository, record, people, dates and places are invented
Author’s constructed example.
Last verified
Figure files:SVGPNG 16:9PNG 4:5
Text version:Read Figure 4 as text
Constructed worked example showing a fictional archival description, its original machine proposal, institutional acceptance record and later correction. A record shown first as a change of wording and then as the trail that change left. The plate opens on a line reading: constructed worked example, no real repository or holding is represented. Under it a line names the record, CEA slash PH slash 1962 slash 0114. Below that, the largest type on the plate, in two lines that open on the same words: on a tinted ground labelled a machine proposal, not yet asserted, the statement as the machine proposed it, “Photographs of the harbour dedication ceremony, 1962”, with the words “harbour dedication ceremony” struck through and the rest of the sentence in ordinary ink; and under it, on the same alignment and at the same size, the accepted statement, the same opening words and the same year with “East Harbor reading room dedication” in archival blue standing in place of the struck phrase. Between the two, hanging off a short vertical rule, the decision: modified, then accepted by an authorised reviewer, and the reason recorded at the time, that the programme in the same folder names the reading room and that no harbour works appear in the 1962 Council minutes. Beneath all of it, under a line reading “the record kept at each stage”, three numbered stages follow one another in the order they happened, each opening on its own ruled line with the name of the stage, the date it covers, and the entries the record held at that point. One, evidence available, before 11 March 2026: the source object, the collection context, the sources consulted and the rule applied. Two, decision, 11 March 2026 at 09:42: the reviewer’s role and delegated authority, the accepted version, and a quality review passed on 19 March 2026. Three, after publication, from 2 April 2026: the publication date, rights and access cleared with no restriction, and a correction reported by a reader on 17 June 2026 and carried as a second version. A rule closes the plate, and one line of type under it states that evidence a decision occurred is not evidence that the decision was correct.
Constructed worked example. No real repository or holding is represented.Record CEA/PH/1962/0114
Machine proposal, not yet asserted
Photographs of the harbour dedication ceremony, 1962
Modified, then accepted by an authorised reviewer
Photographs of the East Harbor reading room dedication, 1962
Change: the event name was replaced. The rest of the statement stands as proposed. Reason recorded: The programme in the same folder names the reading room. No harbour works appear in the 1962 Council minutes.
The record kept at each stage
1 Evidence available
Before 11 March 2026
- Source object
- Six gelatin silver prints, folder 3
- Collection context
- Constructed Example Archive, Public Works photographs
- Sources consulted
- Dedication programme, same folder; Council minutes, 14 May 1962
- Rule applied
- DACS 2.3 title; local rule LR-04, event names
2 Decision
11 March 2026, 09:42
- Reviewer role
- Descriptive archivist, delegated authority for series PH
- Accepted version
- v1, internal working description
- Quality review
- Passed 19 March 2026
3 After publication
From 2 April 2026
- Published
- 2 April 2026
- Rights and access
- Cleared, no restriction
- Correction
- Reported by a reader, 17 June 2026; place name corrected as v2
Evidence that a decision occurred is not evidence that the decision was correct. It makes the decision inspectable, attributable and corrigible.
