A small museum moves its catalog off a computer that has been living under a desk for fifteen years. The new system shows a station bell as object 1994.18. A card in a paper drawer describes the bell as R-73. A yellow label from its last exhibition says 11. The old system had room for one number, which a new volunteer overwrote when they found the label.
After the migration, the museum's exhibition website points to the right-looking bell but links to an old loan record and a picture belonging to another display. The superficial fix is to choose what looks like the accession number and update every row. The useful repair starts by naming the object each identifier was intended to identify and the procedure that produced it.
The museum needs a chain of evidence, not just a preferred number. A temporary intake number may identify an object in a museum's care before it decides what to do with it. A published figure number may identify a place in a specific exhibition. An accession reference can identify an object or group in a particular collection system after its accessioning procedure. A public website route identifies a digital description. Keeping these distinct is the difference between changing a catalog and rewriting the institution's memory.
This article concerns collection documentation and editorial workflows, not legal ownership, conservation treatment or the resolution of a real claim about provenance. Don't change or apply a physical mark, alter accession registers or decide a transfer of title based on a training example. Use your institution's authorized practice and staff for those decisions.
Write a definition for each sort of number
Begin with a diagram of the things you might be indexing. There is the physical object, the register event, the artifact record, the media file, the accession group if one exists, and each exhibition appearance. In a small collection those can look like one entry with attachments. The entities still behave differently. One register event might concern several objects. One object can be shown many times. Several scans might represent the same card. An object card may point to two neighboring objects because someone recorded them together.
Collections Trust's Spectrum numbering guidance distinguishes situations for entry and object numbers and emphasizes a secure link between an object or group and its documentation. Its object entry procedure is different from its acquisition and accessioning procedure. Read the definitions alongside the process your institution actually used. The shape of the number alone rarely tells you what stage created it. A prefix that looks like a year might have been added later, and a gap in a run need not imply a missing piece.
Before importing anything, name the issuing organization or system, number type, exact printed string, scope of uniqueness, and evidence locator for every proposed identifier. For a temporary number, specify which intake sequence or department issued it. For an exhibition label, give the exhibition and edition, since “11” can be perfectly useful in a room guide and utterly unsafe as a global object key. For a prior object number, keep the date range if known and mark whether it was replaced, retained for discovery or still actively used in another system.
The Getty CDWA categories distinguish a work's identity and description from repository location and repository numbers. Its detailed definitions accommodate identifiers from current and previous repositories as well as number type. A former repository's identifier can remain valuable because it is what an old published caption used. Keeping it does not imply that the former repository owns the item today. It preserves a route through the literature.
Make identity an assertion you can test
For each potential match between an object and a document, record the basis. A temporary label attached to an object and a corresponding dated entry form can support an internal match. A picture showing a distinctive documented repair may substantiate continuity when the number isn't visible. Generic dimensions or a common maker will rarely decide between two matching objects by themselves. Use object particulars with an independent record when possible, and keep the responsible reviewer and date on the decision.
Write a status for weaker cases, such as possible association, stronger candidate, verified for a stated purpose, or unresolved. Avoid treating probable matches as aliases in the public site's routing system. Once an alias redirects automatically, visitors see one conclusion and may copy it into other sources. An interface that shows a clearly labeled unresolved legacy reference is better than a neat merge with no audit trail.
The absence of a number on the physical object is not proof it was never recorded. A label can detach, a marking method can change, or a number can have applied to a group. Conversely, a number written on an object could come from a different organization or workflow. Don't add an unverified identifier to the physical object or erase an old one just to improve a search. Spectrum's renumbering guidance discusses why apparent numbering irregularities are often best handled with explanation and cross-referencing rather than by reassigning numbers.
If you must change an identifier in an authorized collections process, capture the assignment event. The CIDOC Conceptual Reference Model declarations distinguish identifiers from actions that assign or deassign them. You don't need to adopt an entire ontology to apply that insight. Record the relevant entity, what number was changed, the type and source of the decision, who authorized it, the date and the relationship between the old and new forms. Locate the evidence for the action independently of the current value displayed by the software.
Worked example: three labels for a bell, and a false fourth match
All identifiers, objects and events in this example are fictional. They are deliberately unlike an official museum's accession register or legal documentation.
Ledger entry 1994.18 identifies a brass station bell and its iron bracket as an accessioned group in the fictional institution. The conservator's authorized inventory record distinguishes the bell as .1 and bracket as .2. One historical intake card says R-73 and lists both pieces, using dimensions and a photograph that explain why the card is associated with the group. Exhibition guide “Rail Journeys 2012” used figure 11 for the group, but a room photo shows only the bell in the case: the bracket could have been represented in the caption or removed before installation. When the old catalog moved online, its row obtained a website ID, obj-0246. The gallery also has a donated picture of a different bell from a 1979 exhibition numbered 11.
| Evidence | Entity or scope | Use in the new records |
|---|---|---|
| 1994.18 in accession ledger | Bell-and-bracket group | Preserve the stated group level |
| 1994.18.1 and 1994.18.2 in object records | Individual bell and bracket | Keep distinct records and connect them to the group |
| R-73 on intake card | Earlier intake for both pieces | Link to group with evidence and narrow scope; don't treat it as a second accession |
| Rail Journeys 2012, figure 11 | Figure or display reference inside one identified guide | Keep as exhibit-level citation; public wording may need further evidence about display contents |
| obj-0246 | Historic website record | Preserve import mapping and maintain a redirect only to a confirmed equivalent web resource |
| Standalone “11” on 1979 gallery photo | Separate exhibition reference | Don't link to the bell or group without more evidence |
Start by deciding the scope of each candidate number before attempting to connect it. The intake card's photograph, dimensions and explicit detail about the bracket add confidence that the same physical group was in care before accession. They do not establish how it passed from one owner to another. The exhibit number has narrower meaning. The guide associates it with a group; the room photograph supports a claim that the bell is visible. It does not establish that the bracket can be seen in the case. Here, the exhibit label and photograph can both be correctly linked while supporting different claims.
Suppose a reprint of the exhibit guide lists figure 11 as “Station bell and mounting.” Check whether that was an edited description or an updated record of the display before revising the case caption. The photo might have been taken at a different moment during installation. For a simple history page the safest caption could be “The station bell appears in an installation photograph from Rail Journeys 2012; the guide's figure 11 covers the bell-and-bracket group.” That phrasing can survive a future discovery about the bracket.
Now handle obj-0246. If the old route was genuinely a group record, it can redirect to the new group page once staff confirm what the old page meant. If it combined several unrelated records through a bug, silently redirecting it to the group would inherit the ambiguity. Keep an internal note and consider an explanatory landing page or other authorized treatment. Never redirect every bare path containing “11” to the same object merely because a search found one familiar bell.
There is enough evidence here for the collection to publish an editorially narrow statement: “Accessioned as a bell and bracket group; the bell is visible in the 2012 exhibit photograph.” There is not enough to say precisely when the mounting entered the gallery or to assign the other figure 11 to it. Let each unresolved part remain unresolved.
Keep provenance fields separate from identification and custody
Provenance asks about an object's history, including ownership or custody as appropriate to the record. An identifier is often one clue in that history, but it isn't an event all by itself. An object can acquire a new inventory number without changing hands. A loan can move it between sites without changing its owner. The former repository's number can help you trace a discussion of the object even when the transfer details remain unknown. Distinguish “held by,” “cataloged by,” “displayed at” and “documented in” rather than letting one number express all of them.
For a public collection page, a useful source sentence can explain its limits: “The 1964 guide used the number X; the current catalog uses Y. The date and reason for that change are not recorded in the sources inspected here.” Avoid deriving the date of an event from numeral patterns unless an institution's numbering scheme and record independently substantiate it. A value beginning 1994 might reflect catalog practice; your claim about accession date should come from the accession evidence.
Keep personal or private documentation in the authorized staff record. A public timeline does not need storage locations, handling details, private lender contacts or unrestricted access to acquisition paperwork. Conversely, make the public trail strong enough to let a reader see the connection between older publications and the current catalog without needing private system access. An evidence summary and permitted citation may suffice where reproducing an actual record is unnecessary.
Deal with the number that was printed wrong
Sooner or later a printed guide reverses two digits or an inscription is transcribed differently in two records. Do not fix the source by changing the archived string. Instead, record what each source shows and which form the institution uses today. If a publication made an apparent mistake, add a qualified link (“printed as 1994.81 in this edition; cataloged 1994.18”) and cite why the institution connects them. If a numeral is unreadable, use a marked transcription rather than filling in the expected number. A catalog gives readers a way to check the past; it should not make an error disappear so completely that an older reference becomes impossible to trace.
One remedy is a claim ledger with columns for assertion type, identifier string, source system, issuing date if known, author of the association, verification reason and public status. Keep the apparent error as evidence about a source, not as a new primary number, and keep an ability to retract the relation if new evidence shows the guide described a different piece. In a spreadsheet, one per-object field can remain the preferred number while the identifier table carries multiple observations. Avoid shifting the old numbers into a plain notes cell where the search engine never sees them and a future export may drop them.
To test how far a correction can go, add an example with a mislabeled photo rather than a simple typo. The written guide might be corrected confidently while the image remains uncertain; one successful review doesn't clear its neighbors. Show the different outcomes to a second reviewer. The fact that the team can accurately say which part it repaired is a better completion unit than a declaration that the entire catalog is now right.
Map custody events before comparing the labels
When a number appears in two registers, look for what each register was doing. One can identify something received for evaluation; another can identify something accessioned; another can name a batch handled for an exhibition. An older owner or collecting institution could have used the same visible syntax for a different purpose. Find the register heading, entry description, version date and indicated institution or collection. A string does not identify a role just because it begins with a familiar letter.
Trace the available records through events independently of whether the identifiers agree. For a local migration, illustrative stages might include receipt, catalog description, authorized accession, a move to another storage location, association with a later conservation report, and creation of a digital record. Not every object follows this story, and no prescribed sequence here supplies evidence that one did. Keep a locator for each actual observed record. A note in the left margin of a register may explicitly connect an old reference to a new one; a confident database field typed years later may only repeat the now-unverified assumption you set out to test.
Distinguish custody from legal or ownership claims. If a shipping record describes arrival at a building, that is evidence of a handling event. The arrival might have been a loan, long-term storage, donation awaiting action, ordinary purchased delivery or another situation. Catalog migration staff should not assign a legal ownership label from physical presence alone. If the migration reveals something needing formal review, flag it and preserve evidence under the institution's policies rather than improvising a determination. The aim is to keep the object and records identifiable while the relevant people answer their part of the question.
The Collections Trust Spectrum guidance on numbering is an institutional example of separating different number systems, such as entry and object numbers. Its implications for a migration are immediate: keep type and issuing context on every value, and do not require all observed numbers to fit the current accession format. Where the roles cannot be determined, set an explicit evidence status. An unattributed inscription can remain searchable in a staff investigation without advertising it as the accession number or silently disappearing.
Design for whole objects, sets and separately managed parts
A label can refer to an entire object, a set, or a separately managed piece. A physical lid may travel with a vessel today without a documented association at the time of initial entry. A crate can carry a group number even while its contents are individually catalogued under suffixes. Group scope matters as much as identifier value. A migration that flattens every suffix into an alternate spelling of a whole-object number can cause supporting condition records to follow the wrong physical thing.
Record the system's segmentation before converting it. Is a final letter an established part suffix, a duplicate number exception, or a collection code? Ask the repository's own documentation and inspect examples with dependable associations. Preserve both group and part identities when they are intentional, and record a documented relationship between them. An association can be provisional if the evidence is not enough. A detached tag may help identify the object or just identify an empty tray. A “found together” entry can be correct without resolving whether items were manufactured, transferred or accessioned together.
A synthetic group record could represent an assembled teaching instrument, two interchangeable parts and a fitted case. A single catalogue description may be attached to the group; a treatment report may name only a specific part and note it was previously held separately. The report can be linked to the part and made discoverable from the group through a documented hierarchy. Copying it to every part record makes future maintenance easier for a programmer and less accurate for a conservator. If the new catalog cannot express that distinction, maintain a clear structured exception record rather than filling it with duplicate copies of uncertain metadata.
The Getty CDWA overview is a useful reminder that descriptive categories, repository information and identifiers serve different purposes. A storage tray can organize the research process without becoming a stable identifier for an object. A description may aptly refer to the complete historic assemblage while an individual part is the physical object currently being examined. Schema choices should keep both references legible rather than forcing a single level of identity because only one field was initially convenient.
A second worked example: two specimen drawers, one painted number
The museum, specimens, numbers, records and results in this example are entirely synthetic. A small technical collection receives a request to identify a set of glass demonstration weights. A historic photograph shows a bench, two shallow drawers and a supplier's box. The new migration staging catalog contains Group T-0042 and a loose object tagged 42. The staging script has reported an exact match after stripping the prefix and leading zeros. A staff member wants to merge them before an internal review.
First, establish how the labels are used. The old collection register describes T-0042 as Drawer Set Two, comprising six weights. An image intake file shows the painted number 42 on a supplier's case beside two drawers, but does not show that the case contained those particular weights. A conservation report describes minor treatment to a weight under a suffix associated with T-0042 and has a drawing of a distinctive scratch. The loose object's tag says “42” without a prefix, and the best available photo does not show a corresponding scratch. No direct register evidence ties the loose tag to the group.
The team records three different observations instead of declaring a single match: the permanent-looking group label in its register, a case mark visible in one image, and the loose tag as read on intake. The case mark's role is unknown and the intake tag may be a working inventory number. No one must prove that the tag is unrelated in order to stop a destructive merge; absence of a verified same-object link is enough to hold the proposal. A useful search result can show both records as candidates, label their types, and distinguish the reviewed group from the unresolved object. That makes the limitation visible while preserving lookup.
A paper register reviewed later in the fictional case identifies “42” as an unpacking sequence assigned to a box containing several pieces for review. A hand-labelled photo in the same batch shows the loose object and the box number but no relationship to T-0042. This likely clarifies the role of the observed marker, not which collection object is in the box. The reviewer can narrow its interpretation and keep the question of identity open using the separate intake record. By contrast, if a properly identified record explicitly connected the loose object, its distinctive physical description, and the old assigned suffix, it could support a typed relationship and may route the case to an authorized reconciliation decision.
Finally, the staging script is changed in the illustrative scenario so short normalized strings produce a candidate queue, not a merge command. A reviewer can still find the case by typing 42. The catalog's durable links retain their issuing contexts; the photographs and reports continue to refer to the subjects their authors actually described. A good migration doesn't force a physical artifact to become more certain than it was simply because the software wants one row per number. It moves evidence together and provides a safe place for competing candidates.
Turn a retired number into searchable context, not a silent redirect
A system can preserve an old reference at least three ways: as a recorded former identifier under the same documented object, as a separate record held pending review, or as an explicit historical link between two records. These outcomes have distinct purposes. A redirect is appropriate only when the relationship is supported and the institution manages such redirects. It is not the default response to matching strings. A catalogue search return that can explain “older number retained for discovery; current object ID below” is better than sending a researcher to an unexplained replacement.
Keep date ranges where documented and don't invent a universal validity period. A catalogue may mark a local number retired on a particular administrative date. Previous journals, physical labels and collection notes can still cite it indefinitely. Preserve retrieval without pretending earlier citations were wrong. Keep a correction note for a past erroneous display when it affects readers who followed it, but avoid promoting a known typo to a valid associated identifier. A staff search hint and a formal former number should be distinguishable.
Where two systems use the same identifier value independently, show the namespace and type in result views. Internal titles might display “2020.11 in Workshop Teaching Collection” and “2020.11 in Traveling Exhibit Ledger” with the distinct use described. A result without context asks a user to infer which collection generated the number. If you run a public lookup, expose only the useful public metadata and appropriate discoverability. A former identifier might encode private intake or location details that shouldn't automatically show in an unrestricted view. Preserve access policy separately from search convenience.
Maintain a clear update trail for reconciliations. The CIDOC CRM declarations provide an example of modeling identifier assignment as an event rather than treating a new value as an invisible overwrite. A local project need not adopt the whole model to benefit. It can keep the observed or assigned value, issuer, time where known, source, associated object or candidate, and review outcome. This is enough to separate modern catalog work from the object's earlier history and retain the actual record of how the relationship was formed.
Verify three classes of lookup after moving records
The first class is historical retrieval. Can a researcher with an older published reference locate the relevant object or an explanatory match? Test with several dated numbers, including one corrected typographical reference and one merely similar label. Confirm whether the resulting object has documented continuity or is one candidate among several. A correct outcome for the latter might be an honest ambiguity message rather than a strong destination link.
The second class is internal operations. Can authorized staff retrieve the exact current identifier while distinguishing the whole from its parts and a physical object from supporting material? A returned image that looks right is insufficient. Open the condition or intake record and inspect the scope it actually names. Check that minor punctuation or export formatting still preserves meaningful differences. If an internal identifier must not leave a particular audience, verify that constraint with approved test fixtures or established access procedures rather than publishing restricted values to see whether users notice.
The third class is reconciliation. Can a reviewer reconstruct why records were paired, what remained unresolved, and the recorded sources for each promoted relationship? An excellent migration log specifies which checks the automated rules do and which assertions were left to humans. For ambiguous cases, let someone work from both original records without seeing the proposed final answer first. Then compare observations. If neither record ties to the physical object, report a documentary gap. Do not create a clearer record of a relationship that no evidence has yet supported.
Prepare the rollback and the next handoff as different things
A project may retain a protected export or an otherwise governed record that makes restoration possible if the move was wrong. That is not the same thing as a fully usable handoff. Another staff member needs to know the mapping decisions and why they were made, the proposed actions that did not occur, and which unresolved associations require further work. Separating technical restoration, audit evidence and ongoing research helps the archive avoid distributing sensitive full backups merely so colleagues can read an exception note.
Give each unresolved exception a specific next question. “Check whether the old lab register assigns this suffix to the detached part” is a question worth carrying. “Fix old numbers” is an invitation to indiscriminate cleanup. When the reported effect is merely inconvenience in search, consider an added search alias or a clearer label under appropriate review rather than rewriting record identity. When the error risks misidentifying an object or moving an attachment to another record, stop before making the change and preserve the evidence for the responsible reviewer. Progress should be measured in more dependable relationships, not merely a smaller exception table.
Keep attachment links from voting on main identity
Supporting files may have old numbers in their filenames because a photographer exported several sessions in the same folder. Filenames are useful search hints. They do not establish which object a condition report describes or which exact version a published caption refers to. A batch of ten files with the same prefix can all reflect the same one incorrect selection made years earlier. Treat them as dependent observations. Verify what the document says about the object, then record the file-level association distinctly from the proposed object reconciliation.
When a report explicitly names its subject and location, check whether those fields and the accompanying image describe the same thing. A printed inventory value can be crossed out; an annotated replacement may itself be an uncertain guess. A published photograph might depict a matching design but not the same physical item. Record where a clue appeared and how the reviewer matched it. If a replacement association remains uncertain, retain a candidate relationship rather than moving an entire folder for aesthetic consistency. If the migration software cannot show private uncertainty alongside public catalog data, retain the material under the appropriate internal review instead of attaching it to an unrelated object to ensure it is visible somewhere.
Give transformed derivatives their own names, even if issued from a verified source record. A published detail crop and its complete view can belong to different media records supporting different purposes. A conservator should be able to tell which photographed surface is available without consulting the person who made the crop. Keep file provenance, differences and use scope together. These names do not override the object identifier or become alternate accession numbers. Conversely a note about removing a scratch from a derivative is not, by itself, evidence of treatment to the object. This is one reason to place media editing and object lifecycle events in separate fields.
One restrained review pattern is to choose one candidate group, verify the set of related files under the intended role, and fill a short relationship register: identifier as printed, person or department responsible for the file where relevant, evidence tying it to the physical subject, relationship status and publication restrictions. If one file refers to a different version or another object entirely, flag that without rewriting its original information. Missing associations should be counted as missing, not repaired by the broad assumption that a path or a filename was always truthful.
Compare two exports without crossing their scope
An apparent number loss after a migration may arise because one export includes only accessioned objects and another includes objects awaiting decisions. Before reconciling counts, write down the selection rules and revision dates used in each view. If the larger export incorporates group and part records and the smaller export uses one line per group, identical raw totals would themselves be suspicious. Compare known samples under both rules rather than expecting a single count to prove continuity. Each selection should illustrate whether an old number remains findable and whether its current relationship is explained.
An acceptable outcome can include preserved unresolved differences. Suppose the older working list has four associated numbers for a set and the verified catalog has three; it is better to show the fourth as a documented exception than to erase it or assume the set must be expanded. The next custodian can test what sort of number it was and decide whether it names a missing part, a container or something not tied to the set at all. The migration retains meaning when it makes that question easier, even if it does not answer every question inherited from the old register.
Test the migration against deliberate failure cases
Create a small validation set up front. Include one old number that should still return the current group, one same-looking value from a different institution that should not merge, one numeral that identifies an exhibit figure only, and one legacy record whose identity remains unresolved. Add a test for a group with two independently cataloged components. A migration with zero obvious errors can still turn component pages into clones of a group, and only a specific test for differing relationships will expose it.
Then test from both sides. In the staff view, search the exact old identifier and verify issuer, scope and supporting documents; inspect what a reviewer can correct without changing the object itself. In the public view, open a known archived hyperlink and ensure the page is actually the same entity, not merely a related result. Check the credits for any displayed image; a valid redirect cannot make a mistaken photo match acceptable or grant reuse rights. A change to a URL should not accidentally alter an image's rights statement.
Count merge and split decisions separately. If two old catalog rows clearly describe one object, keep their source identifiers with a reason for consolidation and preserve distinct historical statements. If one row described two physical objects, create a deliberate one-to-many mapping. A migration report that says “100 percent mapped” can conceal a batch of incorrect one-to-one guesses. It is more useful to report counts by evidence strength and exception type, then hand back a list of the cases a person needs to see.
Save a reversible record of the old system's values, subject to access and retention policy, along with the mappings you applied. Record which normalized field was derived during import, which records were explicitly reviewed and which remain parked. A screen can show one preferred number without the system losing the rest. A staff person should be able to reconstruct all the original paths after a website redesign or a later cross-institution handoff.
A museum that preserves old identifiers treats its catalog as a record of changing knowledge rather than a collection of final answers. The point isn't to display every obsolete label on every page. It's to make the object findable from each trustworthy trail without pretending each trail describes the same event. Keep the old number because it still tells a researcher where somebody once looked.
AI disclosure: Drafted with AI assistance for Apiary. The bell, identifiers and institutional decisions in the example are fictional. References describe documentation standards, not legal title or an actual object.
Source notes
- Collections Trust: Numbering, Object entry, Acquisition and accessioning, and Renumbering collections.
- Getty CDWA: Overview of categories and definitions.
- CIDOC CRM: Classes and property declarations, version 7.3.2.