A donated tin reads 1948 on its lid. A 1971 photograph shows one like it on a shelf. A modern museum received the tin in 1996 and entered it into its catalog the following year. A field on the website says Date: 1996. Is the public record wrong? Not necessarily about the fact it holds. Almost certainly about what a visitor is likely to infer from the label.
An object accumulates dates as it passes through different hands and uses. A date can describe production, component assembly, alteration, documented use, depiction, movement, donation, formal acquisition, accession or the creation of a particular catalog record. These are events in the history of different things: the object, a part, a photograph, a representation of the object, or the institution's relationship with it. Treating them as values in an interchangeable year column can turn a respectable database into a bad exhibit label. The remedy is a chronology with roles, sources and uncertainty rather than a single date guessed by a sorting rule.
This is an editorial and documentation method. It does not resolve title, donor obligations, taxes, authenticity, conservation decisions or agreements. Where decisions of those kinds are relevant, leave them to the appropriate people under the institution's procedures. You can improve the historical wording without making any of those determinations.
Identify what the dated sentence actually claims
Begin with the literal wording or visual observation. “Box stamped 1948” reports an inscription. “Donated in 1996” reports a transfer if the relevant record actually warrants that description. “Produced in 1948” requires a more particular basis, which could be known markings, dated records, corroborated provenance or other appropriate evidence. A visitor might reasonably interpret “1948 tin” as a manufacture date. Ask what basis lets the caption make that leap from an observed mark. A fresh description should name the narrowest event it can establish.
Do the same for each other date. When you find 1971 in a photograph caption, does it date the image exposure, development, album inscription or a retrospective label? Did the photographer identify the same object, or is the match based solely on a common appearance? A date can be exact and still attach to the wrong subject. An older reliable photograph of the exact object can provide evidence that an object existed by the date of the image, but a similar photograph cannot simply establish that because both subjects look alike. Your chronology should not erase the difference in how images were matched.
The Getty CDWA discussion of creation provides a structured model for expressing date ranges and uncertainty about production. Its definitions of categories also distinguish functions such as provenance and credit. You do not need to adopt every field to benefit. Start by typing what kind of event each row refers to and which part or representation it concerns. Then anchor it to a source rather than copying values across event types.
When a project asks for “the date of the object,” write a question more specific than the prompt. Do they mean the manufacturing interval, the known use date, the first documented presence at an institution, or the date relevant to how a museum tells the story? The right answer may be a two-part note: “Date of manufacture unknown; in use by about 1970 according to a verified caption,” if the evidence supports both. Saying unknown in one field can coexist with saying something substantial in another. It does not represent a failure to choose a single correct year.
Keep event time and evidence time separate
A piece of paper recording an event can be created after the event. A later donor recollection can describe an earlier use. A received email can cite an earlier due date. A collection management entry may show its own creation date while reporting an earlier acquisition. Save the original statement or transcribed excerpt with its source date, then record the event date it asserts and how specifically. If the object record changed in 2015 to describe an event in 1962, those are not alternative readings of one field.
Provenance work can also distinguish location, event and knowledge. A known ownership record for a date range does not by itself represent the beginning of anyone's association with the item. A shipping log can establish that the object or a documented package was moved on a date; it may not describe a donation. An accession entry can establish an institutional record action without proving exact physical delivery. Collections Trust's introduction to Spectrum describes separate procedures, including cataloguing and acquisition and accessioning. Follow institutional definitions and evidence rather than assuming that several steps are administratively synonymous.
Write the date roles in ordinary language before deciding any technical schema. “Arrival recorded” describes a source observation. “Officially entered in register” describes a different source and event. “Ownership changed on” is a deeper claim that needs relevant evidence and authority. Don't infer the legal or financial implications just because one form exists. A chronology field can say which record the institution has and what it appears to document without asking a volunteer to adjudicate the underlying transaction.
Track source independence as well as freshness. An older family narrative can be the only reason many later records use one manufacturing year. An apparently separate date citation might be a printed extract from the same earlier hand-written form. Preserve whether materials echo the same source or independently observed something. The public explanation could then say “described by the donating family as used around 1950” if that is accurate and appropriate to publish, instead of invoking unsupported corroboration. Personal details should not be copied into a caption merely to show that a human remembered the date.
Learn which uncertainty your date format can carry
A searchable date can be exact to day, to month, or to year; it can describe a range or one of several uncertain possibilities. A public display needs to say plainly what that uncertainty means. “Between 1930 and 1940” is not the same as “about 1935,” and neither is the same as “before 1940.” Why was the date bounded? Was it a documented established interval within which the event occurred, a broad editorial estimate or a statement that evidence ends at one known endpoint? Provide a simple role and type before formatting the text.
The Library of Congress Extended Date/Time Format specification is a documented way to represent useful forms beyond the plain numeric date, including certain uncertainty and interval expressions. A team using it should select the documented level its systems actually support and test what happens on import and export. Do not invent syntax that a target parser ignores. Nor should every project try to encode speculative detail in an advanced format. A trusted literal display value plus an appropriate verified search range can be enough if the export preserves both accurately.
Treat machine range and public wording as related but separate representations. A search index may intentionally allow a broad interval so people discover an approximate result. That does not make the endpoints exact manufacture dates. The display might say “around 1925; estimate based on identified design evidence,” with a search bound chosen through documented institutional practice. A single sorting number used to order a list is even weaker as a source of public claims. If the interface turns the sorting default into the headline, fix the presentation mapping rather than assigning all approximate artifacts fake January 1 birthdays.
An unknown endpoint is useful when it is genuinely unknown. Knowing an object was in a collection by a dated documentation event can supply a latest verified observation without proving all the years before it are plausible equally. If the system requires a start and end even where none can be responsibly supplied, distinguish a practical search convention from an actual historical statement. Prefer an explicitly unknown or unindexed state rather than storing an arbitrary date that will surface in a label. Later editors cannot reliably reverse an unmarked sentinel value into its original meaning.
Uncertainty can attach to the attribution of the date, not just its digits. “The label may refer to a repair, received before 1960” is not machine-equivalent to “repaired before 1960.” Where the underlying event remains in dispute, store the observation and the candidate interpretation separately. A correct representation of date precision is necessary and insufficient by itself; it doesn't turn a plausible event type into a verified one.
Synthetic case: the painted case dated twice
The institutions, items, labels, documents and dates below are invented. A history center receives a painted case thought to have been used to transport merchant samples. One plaque reads 1932. A paper label stuck inside its lid reads “New lining, 1954.” A collection accession printout says 2004 and shows a separate status from the receipt ledger, which records the case being brought to the site in 2003. The website merged these fields so its public label reads “Merchant case, 2004.” A curator wants to replace that with 1932 immediately.
The document-based response begins by naming what is known. The 1932 plate is on the exterior, and nothing in the example records precisely when it was attached. It may date the ownership, recognition or use of the case, not manufacture. The other label looks like a maintenance note and suggests a relining in 1954. That possible event dates part of the case's history and not necessarily the wood or finishing. The two institutional dates concern physical arrival and accessioning. Their distinct roles are encoded as such rather than both set to general date. No part of this classification requires publishing who brought the case to the site.
An archival sales photograph dated 1938 shows this kind of carrying case with a similar plate shape. The curator checks identifying evidence rather than adopting 1938 as a reliable latest production date automatically. The case in the picture has the same broad design but different visible hardware. Neither the original photo nor its description establishes that it shows this exact item. The provenance note says “a comparable case appears in a dated reference image,” and the public catalogue does not use it to bound the manufacture date of this one. The analogy may help find a design series, but the current conclusion is intentionally narrow.
In a later inspection scenario, an appropriately authorized specialist compares a clear distinctive pattern of dents and a partly legible business reference with an independently dated photo of the same case taken in 1941. They record the evidence and limitations. Now one reasonable conclusion is “case documented by 1941” provided that match is appropriately reviewed. Knowing a date by which the case existed is not knowing the exact year it was made. The card might read, “Merchant sample case, documented by 1941; date of manufacture not established.” Another line can explain the printed date and its uncertain function if it matters to the exhibit. The accession date can be displayed separately and only if relevant.
The right revision therefore does not simply swap the newest year for the oldest. It changes the role of the date shown to visitors. The 1932 impression, 1941 documentation, repair clue, arrival and accessioning are all useful. They can coexist without forcing a uniform chronology of certainty. If a manufacturer record later confirms a range and can be linked to the exact case, the catalogue can add a production statement while preserving the previous evidence and correction trail. That's an improvement in what is known, not a retroactive declaration that the date on the plate always meant production.
Put component and image dates next to their own subjects
An artifact can include materials made at different times. A case lining can be replaced, a ceramic object can be joined to a later base, a window panel can be in a later frame. A collector can photograph a collection long after the pieces were made. A reproduction can be made from an original design produced years before the reproduction itself. For the public reader, a label that says “1900” without naming the role can be confusing even if the team correctly identified one of those times.
Assign separate labels to dated relationships. “Design documented in 1900,” “this reproduction made in 1970,” and “photographed in 2010” can all be simultaneously true. A depiction's date matters when studying its technology or appearance, but should not silently replace the date of an object in it. A related object might be sufficiently different to have its own catalog number and date context. A later repair is a good subject for a caption if the display discusses the artifact's progression. Indicate clearly whether its date refers to the visible repair part, its manufacture, or the known activity of installing it.
For editorial review, track each sentence's subject and date together. If a label begins “this version,” verify the relevant copy. If a date appears on a photo credit, check whether it names photography activity, publication, or the date of source artwork. A newly digitized image can seem very recent on a detail page. Modern digital creation dates aren't years of manufacture of the historical artifact. A file destination with a prominent update date may need stronger display labeling, so an otherwise correct historic date isn't overwhelmed by the wrong number. The remedy is one of information design as much as database field selection.
The Getty Vocabularies editorial descriptions of creation dates include creation-related activities and discuss expressive display dates. Scope the detail to the thing you mean. Don't translate every component activity into an invented single year. A concise note about different phases is often more faithful and more interesting: it can show how the object evolved rather than flattening the restoration, adaptation and later representation into one event.
Synthetic case: a replica that inherited an original's date
The workshop, museums, guides, item names and events in this example are hypothetical. An exhibit holds a study replica made in 1988 from a pattern originally designed sometime around 1910. A catalogue import associates it with the design series, correctly making both examples discoverable together. The public view incorrectly displays a big 1910 title above the replica picture, because the old model used one earliest date field for everything. A label nearby also says “model built in 1910” although the exact model on display was made to study the older design.
Start with the documented relation. The displayed object is a replica, not the historic item that supplied design reference. Its maker documentation identifies the later fabrication as a teaching project. A reference catalog dates the design approximately, without proving the existence of this particular physical example at that earlier time. A photograph of a lookalike has its own chronology and remains separate. The staff writer can say “1988 teaching replica of a design documented around 1910.” If the design date is not securely established, adjust that part of the sentence while keeping the replica year where documented. A display can describe complex history without forcing visitors to guess which part is real.
One authorized reviewer identifies a mislabeled photo in the web view. The image belongs to the original design sheet rather than a photograph of the replica. The correction can label the image as a drawing, add an appropriately approved depiction of the replica, or use both as clearly linked comparison views. The team doesn't edit the photo metadata so a single date can magically serve both. It keeps the represented thing and representation date distinct. The word “replica” in the caption goes near the central image, not into a hidden paragraph that most viewers never open.
After the code fix, the ordinary text page displays both date roles appropriately. A timeline export still puts the replica in the early design year. That's a legitimate taxonomy choice only if the timeline is explicitly about when designs originated; a chronology of dated physical exhibit objects should use the replica date or explain how records are grouped. Verify what the visualization actually purports to show. Do not automatically insist that every list use the same date role. A mixed timeline can be useful with clearly identified types and a way for readers to distinguish them.
Write a source note that doesn't expose the donor
Private or internal documentation may contain more detail than a public label needs. Use an appropriate internal locator to the records and preserve the exact observation supporting the date, but do not reproduce personal identifiers, contact information, financial details or unrelated donor correspondence. A role-appropriate caption can say “recorded in the collection since 2004” or include the institution's authorized credit. A credibility gap in a caption should not be solved by displaying private research data. Evidence can remain verifiable to authorized stewards while the public language remains appropriately scoped.
Keep source title, section or page, wording transcribed where permitted, date type, date of review, and who is responsible for checking the reasoning. A document may be cited internally even if it shouldn't be published. If the public read is important, a nonidentifying note can name the record class and state the circumstance so the claim remains understandable. Someone auditing needs to know whether the statement came from a dated register, contemporaneous note, retained image or retrospective memory. They needn't be told a home address to understand a dated event.
Do not misrepresent a retained form as authority for an event it does not cover. If an entry form was created at receipt, label it accordingly. If memory is the only available lead to a date, say so and seek independent context where appropriate and permitted. Absence of public donor names should not force artificially strong object claims. The narratives can be cautious and informative. This is especially useful where a museum intends to distinguish documentary proof from a later oral account without making either person's identity the focus of the exhibition.
When publicly correcting a record, present the relevant difference. “The earlier page used the accession date as the object date. It now distinguishes manufacture not established from accession year.” This is a precise correction describing roles. It doesn't need to frame the individual who entered the data as responsible, nor disclose the materials the reader isn't meant to see. Keep private workflow fields available to appropriately authorized staff for later reconstruction rather than circulating them freely for editorial convenience.
Check display, filtering and sorting separately
A date field can look right in the main object view but be wrong in search, highlights or timeline exports. Verify the published date phrase and nearby qualifiers. If a date comes from a heading such as “gift dated,” test whether the short search preview includes that role or accidentally drops the qualifier. An export may flatten uncertainties, sum ranges incorrectly or enforce a numeric map with unintended consequences. Test a few carefully chosen cases with appropriate data rather than assuming an internal spreadsheet guarantees correct public presentation.
Start with an exact known creation year, a bounded possible range, an unknown manufacturing date with known acquisition and a dated reproduction alongside a documented earlier design. For each, write the expected result for a public label, chronological sort and query filter. The correct behavior could differ. A search for work made in 1910 might exclude an explicitly 1988 replica, while a search for designs around that date could include it with an appropriate cue. This is not a trick: it defines what the UI says. A generic dates label cannot explain a system that mixes several of these roles without warning.
For a collection migration, preserve the older literal statement temporarily in a governed version history where useful, rather than deleting it before staff determine why it was stored. The current public view can distinguish an obsolete default from a verified date. If the main caption becomes more cautious while a mobile tile or print list still shows a raw year, the work remains incomplete on that surface. A content system timestamp does not verify a gallery label, an image credit or a tablet display. Read each meaningful surface, with the appropriate role and approved access, and note what was actually observed.
A small negative control can catch dangerous defaults. In a fictional fixture, set a known acquisition year while the creation year remains unknown. Test whether any outward-facing script fills it with the known value. Set contrasting date ranges for design and reproduction and examine whether the site merges them sensibly or should keep them distinct. Use sandbox or approved training materials rather than manipulating real objects or private records to produce a useful failure. If there is no safe way to verify an external use, identify the unknown rather than inferring it from internal testing.
Determine what kind of temporal boundary a document can supply
Sometimes a record can't tell you when anything happened but can put a credible bound on the event. If an independently dated drawing demonstrably represents the exact object already in existence, it may establish that the object existed by that time. It doesn't establish when construction began. If a supplier ledger independently documents the first availability of a component necessary to a particular configuration, it might provide evidence about when that configuration could first exist. That does not by itself date every earlier component or the whole object's original structure. Before inferring a bound, write the chain of premises and mark the one most likely to be uncertain.
Two boundaries do not both have to come from the same source. For a small study, imagine a printed repair notice distinguishes the old and new configuration and a securely dated later image shows the new one. The bounds might bracket the presence of those two known configurations rather than when a particular hand performed the work. Only if the objects and states are genuinely linked should they support “change probably occurred in this interval.” If an earlier photograph merely fails to show the relevant corner, it is not necessarily negative evidence of the feature. A date range whose weakest premise concerns what could actually be seen should say that explicitly.
An open interval can be the honest result of many hours of research. A finding might say “catalogued in this form by 1935; no earlier verified occurrence located in these records.” That's more useful than pretending “before 1935” is a complete production range. Give the maximum effect to observed evidence and the minimum necessary effect to a missing observation. A church inventory may not include small objects; a trade brochure might advertise options selected for market interest rather than describe every type already sold. One source's absence cannot establish a universal earliest year.
Pay particular attention to the word “first.” A first reported use in the sources checked is not necessarily a first historic use. A first catalogued date in the surviving system is not an object entry if the institution previously maintained a separate register. If a public label uses first as part of the story, explain first under what scope and according to which record. A less dramatic statement near the exhibit can be better supported and give readers something they can understand without access to every closed archive. The next investigator then knows which earlier sources could matter.
Design an uncertainty note independent of the sort index
Imagine a system requires a date sort key for every public object. Someone selects the earliest year on the record, which often comes from a design source or an illustration, not the artifact itself. Another person chooses the accession day because it is the only exact triple of year, month and day. Either solution makes editorial output brittle if the sort field is rendered as a date or its role forgotten. Decide what the list is for: ordering acquisition by institutional year, arranging work by estimated year of creation, or tracing a design theme across periods. A single unknowable “object date” can't power all of them correctly.
A practical record can have a public display phrase, a typed historical date or range, a sortable presentation key where justified, and a note about precision. Where the event is uncertain, a broad index may help discoverability without giving the public a false sharp endpoint. Verify what your actual implementation does. A numeric conversion might silently trim a qualifier or parse a valid expression as missing. Conversely, a human-readable field might not make useful filter results if the numeric index lacks a linked range. Treat label display and search behavior as separate tested outputs rather than expecting format conformity to help every reader.
Suppose a store card says “from an 1880s pattern,” and component evidence suggests later assembly. Indexing the pattern reference for discovery may be appropriate. A filter advertised as “objects made in the 1880s” should not include the item solely because it finds that label in a description. If a broad search includes it, distinguish why it matched. Perhaps it appears under “associated designs,” not under confidently dated physical objects. Keep attribution fields where a reader with limited time won't mistake the pattern label for an assertion about manufacture.
When range and description do not match, find out which is authoritative for each purpose instead of normalizing until they agree. A bounded search might include multiple possible years while the public label reports “probably late nineteenth century.” A verified month and year might be enough for a precise institutional timeline without a day. If a stand-alone export cannot preserve role and qualifier, consider a plain accompanying note and verify that the exported timeline is not mistaken for evidentiary text. An honest handoff can still serve search and interpretation without inventing information merely to satisfy a data type.
Retain conflicting observations without averaging their years
Two sources can name adjacent or widely separated periods. Before selecting an average, ask whether either source claims a particular kind of date and whether the sources concern the same physical thing. An association that the early date relates to a handwritten design and the later one to object assembly can remove an apparent dispute without correcting either source. Perhaps one date belongs to a collection label and the other to a contemporary shop record. The point is not that contemporary sources deserve automatic victories. It is that the observed event, the evidence relationship, and the resulting claim need to be separately understood.
A small conflicts log needs two literal assertions, what each one refers to, how the association was checked, and current editorial outcome. If the conditions needed for one date to be reliable are not yet met, note those conditions rather than stripping the claim out of the record. Another investigator might later find a reference confirming the correct identification or show lack of independence. If neither source can support more than a general estimate, use a phrasing consistent with the actual evidence or set the public date to not established. Preserve the specific dates in appropriate source notes, not as two equally verified options on a display without explanation.
Defensibility matters more than narrative neatness. A curator could introduce a current label saying “Formerly catalogued as 1920; newly reviewed documentation concerns a later repair, not original manufacture.” That correction distinguishes evidence and context without alleging who made the previous error. A simplistic correction saying “actually 1958” could be worse if it reflects only a newly found receipt. The best update may narrow the visible claim even as the underlying chronology gains information. If the public story is about longstanding use, repair and donation rather than a manufacturing date, the few verified intervals will be relevant and genuinely interesting.
Use a rehearsal to show where meaning gets lost
Take a small set of approved examples or invented records, not sensitive materials. Include a date from an object inscription of uncertain role, a verified dated image of the exact example, and a verified acquisition with no reliable manufacture date. Ask one editor to draft captions and a second to read them literally. Their task is to identify the event they think each date describes. If the reader repeatedly infers manufacturing where the editor meant accession, move clarifying nouns into the main label. It is better to find this failure on synthetic data than in a printed photo book that may circulate for years.
Next compare the text in three places: the full page, the short printed label and any list view or timeline. Keep the expected role and actual observed display for each. Do not treat a test showing three corrected captions as proof that every other item in the catalogue has been checked. Consider which common import or formatting rule might affect other entries and give the responsible team a way to sample those. Check that uncertainty remains legible when a layout shortens the text before a qualifier. Your method succeeds when a reader who was not part of the research can tell whether a date signals manufacture, documented presence, later work, acquisition or something unresolved.
Hand the next writer a chronology with labeled limits
A useful packet has one row per supported dated statement. Give it subject, event role, exact date as observed or cited, precision, whether it is inferred or directly recorded, evidence locator, and public wording. An observation can remain a source clue if the role is unknown. A separate line can summarize the preferred display statement, and link to related records that cover repair, acquisition or reuse. This is smaller and more meaningful than a folder full of old scans with no explanation of which date the editor wanted.
Ask the next writer to explain aloud what the public sentence dates. If their answer is “the donation” but the label headline implies creation, move the clarifying noun into the headline. If it isn't possible to say it briefly, choose the narrower fact for public display and keep a link to the detailed note. The finding could be “the exhibited copy was probably made in the 1930s; exact year unresolved,” alongside an assured later receipt date. The uncertainty belongs with the relevant claim instead of under an unrelated small-print disclaimer.
For timeline projects involving both original and reproduced items, include an explicit “represented work” relation beside the date role. A readable label might say “created in 1988; reproduces a design documented around 1910.” If an exported historical overview puts both dates into a single event order, show which source attaches to which of the two claims. A reader should be able to sort by the manufacturing chronology without losing the displayed history of the design. This also lets a team explain why a new finding affected the original design estimate but left the later date of fabrication unchanged. The better representation doesn't force every reader to use both dates. It keeps the choice intelligible.
Maintain a record of change when an object date materially shifts. Name the prior interpretation, the evidence now available, the fact it actually changes and where the revised language was verified. Adding a production estimate does not alter the known arrival year. Discovering a photograph source error does not automatically invalidate all other dates. A correctly typed timeline allows improvements one relationship at a time. This is why the simple habit of asking “date of what?” matters. It lets the museum keep a reliable story without demanding pretense whenever a field hasn't been filled.
Editorial note: Drafted with AI assistance for Apiary. Collection records, cases, dates, replicas and demonstration outcomes are synthetic.