Three small blue glass bottles stand in a case. The labels were edited correctly last week. The collection records include distinctions among all three. But the case was rearranged after the labels were printed, and the web guide inherited the old sequence. A visitor naturally pairs the leftmost bottle with the text nearest it. Something in the presentation is wrong even though no sentence in the content management system is necessarily false.
That is the problem of matching, not just proofreading. A museum can have flawless descriptions and still present inaccurate information if the material evidence under a label is not the object that description concerns. Alternatively, a visitor can correctly identify the object and open a digital guide that illustrates another example from the same edition. A practical quality check needs at least three independent facts: what the record is about, what the exhibit actually contains, and which visible text directs the visitor to which thing. A completed editorial approval covers none of those relations simply by existing in a folder.
This exercise is for modest editorial and exhibition checking under the institution's approved procedures. It isn't a handling guide. Never move objects, open exhibits, expose protected storage information or carry out photography contrary to policy in order to complete a test. If an item can't be safely or appropriately examined, an authorized collections specialist's documented confirmation can carry more meaning than a volunteer's attempt to get a better angle.
Start with distinct identifiers for the object, position and caption
An object identifier, exhibit position and caption version answer different questions. The object identifier points to a managed thing or group. The position identifies where in the presentation someone can expect to see it. The caption version identifies the copy actually intended to accompany it. You can swap any one of the three without necessarily changing the others. That's how the error arises.
Use simple keys within their own scope. A collections record might identify an individually catalogued object. A gallery plan can give the cabinet and position, and an editorial proof can carry its own revision. Don't convert the objects' separate keys into the labels A, B and C until the mapping is documented. A local letter can serve visitors well without functioning as a permanent collection ID. Conversely a formal collection number can distinguish the object internally without being appropriate to print in large type in the exhibit.
The Getty CDWA categories overview distinguishes several kinds of work and repository information. It is a useful prompt to keep work-level and repository-level description apart. The Collections Trust's Location and Movement Control guidance illustrates why a record of where a managed object is located belongs to an institutional process, not to a casual assumption based on last week's layout. Follow the applicable procedures in your own institution.
A minimal check row can contain object or group key, expected placement, evidence of confirmed actual position, expected caption identifier, observed installed or rendered caption, and unresolved status. Reserve another field for the visit date and surface. A physical case and web guide might be checked by different people on different days. If replacement work has been approved in one channel, don't mark the other as changed until observed. A useful result can be “catalogue match verified; physical label unresolved” instead of resetting the whole row to none.
Keep edition, item and illustrated example apart
A print edition can include several distinct physical impressions. A manufacturing run can generate several closely comparable specimens. A catalogued work can appear in a related reproduction or photograph for study. A person describing the edition should still check that the caption's individualized facts, such as condition, marks, ownership history or the source of an illustration, apply to the exhibited object itself. A shared manufacturer and date range are not proof of item identity.
Start a photo match with a controlled view of obvious, nonidentifying features where allowed. Use broad shape and arrangement to identify candidates. Then look for a diagnostic difference known to the collection record or checked by an authorized person. It could be a visible base profile, placement of a design, a documented edition statement, or an approved non-sensitive mark. Prioritize characteristics that can be observed without altering display or handling the work. A tiny composite photo of different items without source associations can introduce another doubt rather than resolving one.
Distinguish through reliable institutional evidence, not guesswork based on similar-looking pictures. Matching a red glaze is weaker than matching a documented body shape, mark placement and verified movement confirmation in context. An object's title on a label is not independent evidence that the object matches; that's the disputed link. Two screenshots pulled from the same planned layout share the relevant assumption, even if the format differs. Look instead for a trustworthy item-level identification and an independent placement and installed-text observation.
Keep labels for reproductions, reference images and originals separate. The audience can learn from a photograph of an unavailable object, a cast and an original artifact. Each presentation needs clear words describing what viewers see. An exhibit about printing can feature a plate, an impression and a modern facsimile illustrating a missing stage. Saying all three impressions belong to the historic edition because their motifs match would lose the historical point. Put “modern facsimile,” “illustration of a related item,” or a similar accurate role where a reader can find it, not only in internal design notes.
Search result importance should not replace object-level verification. A software feature that imports the primary image for a work by default may select a representative view from the collection rather than the real displayed copy. The digital guide should indicate that relationship if using such a view serves the audience. If a caption reports an inscription only on the photographed example, it cannot be silently paired with an example that lacks that feature. It may be helpful to name which example was consulted or change the chosen photo. This often matters more than matching typeface or spacing.
Decide what counts as an installed proof
An approved proof file can be the starting point, not the endpoint. To know what was installed, read the visible label in its context after the relevant installation or update, using an approved review path. To know what the web guide shows, inspect the actual rendered entry and its link to the correct object rather than relying on the CMS edit screen. For an audio guide, check the numbered stop mapping and spoken sample where applicable. A retrieval success on one device doesn't prove that an older printout, cached view, or alternate guide changed at the same time.
Before conducting checks, describe what a reader will use to pair the text and object: physical proximity, order, guide number, leader line, keyed illustration, QR destination or another marker. Then test that relationship. Reading labels in staff order doesn't capture what happens from the entrance or at normal viewing positions. Move no items and do not interfere with visitors. Where controlled observation is possible, try the same route a recipient would reasonably take. Verify the visible marker and its associated text without inventing what a viewer should infer from an unobserved angle.
The Smithsonian Guidelines for Accessible Exhibition Design provide an institutional reference for treating public labels and exhibition access as design responsibilities. Here the takeaway is that a match must be understandable to people in the public setting, not merely known by the people who installed it. This procedure does not certify accessibility or substitute for relevant professional assessment. It can identify an obvious communication fault such as two equally plausible objects beside one label or a front-of-gallery legend accessible only after a visitor has already chosen the wrong stop.
Preserve evidence with the least intrusion. A narrowly framed photograph of permitted label text and a harmless matching feature can sometimes document the pairing. In other cases a short note or sketch, checked by an authorized staff member, is enough. Avoid recording visitors, private donor paperwork, protected location information or details the institution does not permit photographing. Treat the photo as support for the specific observation, not as permission to share it publicly. If photography is disallowed, use a role-appropriate written verification rather than inventing an exception. The audit isn't a reason to widen who can see the collection's internal controls.
Synthetic case: a square base, a round base and a photograph
Every object, gallery, label, date and activity in this case is fictional. The museum has three objects in an educational display about bottles from a particular production run. Bottle I has a square base and a distinct shoulder profile. Bottle II has a round base and an otherwise similar blue upper half. Bottle III is a demonstration facsimile with a broad gray neck ring. The dry-run plans put them left to right in that order. The designer creates copy describing two originals and one facsimile, and a reviewer checks that the planned pairings are accurate.
When installation ends, the two historical examples have exchanged positions for an approved display reason. The round base is leftmost. The label closest to it still says the left object has the square base. The reviewer doesn't handle anything or decide item identity from a distance. They ask an authorized staff member to reconcile the approved identity markers and confirm which example now occupies each position. The record shows the different actual order. The facsimile remains in its planned place near the far right.
The website's record for Bottle I is already accurate. What the website gets wrong is the relation between the physical tag and the web entry: the link associated with the left position opens Bottle I instead of the newly displayed Bottle II. The image in that entry shows a cropped neck that could plausibly belong to either. Since the bottle is a closely comparable item, a casual check that simply opens a page and sees blue glass will miss the fault. The correct review traces object, physical position marker and visible copy together. A payment QR or staff-only link is not opened just to test this; only the established public guide route is in scope.
The three states can now be described separately. Installed observation: round base at the left. Editorial proof: left describes the square-base example. Public guide behavior: first stop opens the square-base entry. There is no blanket conclusion that the catalogue is wrong. A staff fix might change direction markers, labels, guide mapping, or another element under the appropriate approvals. The job for the reviewer is to identify which relationships are failing. Leaning an appropriate new label next to the actual object without arranging the approved placement would be another unapproved editorial assumption.
After the responsible team says it updated the proof, the reviewer comes back through public access and confirms the physical label association matches the signed-off plan. They verify the guide separately and find its first stop still stale. Nothing about the successful physical correction is described as fixing the web. The note says, “Case labels and actual placement agree on follow-up observation; first guide stop still opens Bottle I where the new mapping should open Bottle II.” This is a usable report with one unresolved surface, not duplicated work described as complete.
Now consider design choices. Rather than telling visitors to count left to right, the museum can use local identifiers printed in a way people can see, paired with accurate captions and verified positions. An illustration could highlight the two base shapes if it represents the checked objects. If one digital image cannot be distinguished confidently from a related example, label it representative or choose an approved view that actually illustrates the distinguishing point. An exhibit doesn't have to expose private object IDs to solve a matching problem. It needs public cues that refer unambiguously to validated placement and copy.
Separate “text is correct” from “all the links work”
Content review verifies the statements and wording of a caption against evidence. Placement review verifies which item the reader will connect with it. Link review verifies destinations and device views. A tall exhibit can fail one of these tests without failing the others. A paper label might correctly describe a photo of a miniature intended as the neighboring object reference, but the photograph is placed above a full-sized original. A web link might open the right item page and show an out-of-date caption downloaded earlier. Giving each dimension its own check makes the narrow correction obvious.
Design a tiny matrix with item, phrase or claim tested, physical position, visible version or readable excerpt, intended destination if relevant, evidence note and result. Each row should have a specific observation. “Guide checked” is too vague if the implementation serves different links on phone and printed signage. “Stop 2 from the case label opens a page describing the same accessioned example, top image confirmed as that item” is stronger. Check only the legitimate surfaces involved. Do not follow unexpected personal or sensitive links from exhibit material in order to complete a broad guarantee.
For the physical place, test visible symbols and viewing order without assuming the curator's perspective. A group label might intentionally refer to several exhibits. It should say so plainly enough that readers won't interpret one object's individual note as applying across all cases. If a guide identifies several items with one audio stop, clarify the structure rather than marking every item as having an individually accessible description. Group level, item level and illustration level are separate audit units. Make this explicit whenever there are close variations in object size, provenance or date.
A checked site preserves navigation in the output the reader accesses. A link that opens a right chapter but wrong image has failed the relevant test. A claim that two objects share an edition is compatible with a comparison view; it is not a defense for tagging the wrong item. Verify representative alternate surfaces with volunteers or staff using allowed devices and normal roles. Keep a note about the test's scope rather than implying all readers will necessarily see the same result. There may be navigation paths, print versions or language editions not included in the check.
Use negative controls to expose copy-paste errors
Pick one nearby object that should not pass. If your verification rule identifies it as the same example, improve the rule. For example, a simple requirement that every bottle is blue cannot distinguish the matched examples. A source-identification code captured in a different area might show two records; it still may not distinguish their installed labels. An appropriately chosen negative control checks the type of evidence that could expose confusions rather than accumulating confirming adjectives. Keep both the matched and deliberately mismatched cases in your internal test notes.
Try a plausible copy-paste mistake. A label writer can replace one noun and accidentally retain the wrong date. An image field can be switched while its descriptive text remains on another item. A case map can refer to an earlier object arrangement if it was pulled from an approved but superseded export. A justified test should make it possible to tell each issue apart. A gate that only checks whether each caption string is nonempty won't detect incorrect linkage. A manual reviewer should compare the exact item-level feature or phrase with the record supplying it and then inspect the surrounding presentation.
Role checks should stay safe and proportionate. Use synthetic preview material for rehearsing invalid links or private asset cases; don't expose sensitive material to see if a reviewer notices. Use the documented support process where a destination appears inappropriate. A failed negative control might simply mean your method would let an error go unnoticed. That is a valuable QA finding even if all current visible labels happen to agree. Correct the test while the exhibit is small rather than claiming a successful audit across many items it cannot distinguish.
A low-stakes demonstration can compare three labels shorn of their IDs and challenge reviewers to distinguish them. The intended outcome may be “no; visual cues alone underdetermine two matches.” Restore the identifiers and find a concise public label strategy with the staff. Such a result is not an admission that viewers must consult the entire catalog to enjoy the display. It can identify where a local symbol, orientation note or more distinctive context would remove ambiguity, even before a visitor sees the permanent text.
A second synthetic exercise: a reference illustration shown as if it were a substitute object
The following series, venue and publication are also invented. A small exhibition about printmaking presents an original signed impression on loan and a generic image illustrating the motif. The guide maker accidentally uses a high-resolution file of a different impression. The displayed impression has a number at the lower left. The guide crop hides its own edition markings and its accompanying phrase says “you can see the signed and numbered impression pictured here.” Archive staff have a good documented record for both items. Nothing in the phrase tells a visitor the two representations are not the same exact impression.
A first review checks whether the web page is talking about an item or an edition. Its title is the work and edition-level description. That much could properly accompany the representative image if it is identified as such and appropriately used. The next sentence claims an individualized numbered detail. Its evidence is the actual exhibited item. The comparison photo can't verify that sentence because the framing crops the relevant area and the file shows another impression. The staff choose either to supply a suitable approved photograph of the exhibited item or to label the guide image as another impression and keep the actual exhibition claim explicit. No one alters the source file's ID to make an available image match the purpose.
The physical caption is also read independently. It names the impression on view and says which marking matters. A reader who uses the online audio while standing in the gallery might still reasonably think the guide image is a photograph of the work directly before them if the text doesn't distinguish it. The wording is revised at the moment it describes the image, not only buried in documentation. The note gives both actual object reference and related image role. This change restores a true relationship while retaining the useful comparative visual.
When the loan ends in the synthetic example, the display is replaced with another impression from the same work. The motif hasn't changed, and much of the work-level label remains true. The item-level sentence and credit can no longer be inherited automatically. A quick label matrix shows which fields are shared across the work and which depend on the exact item. An authorized check confirms substitution and a fresh proof covers the affected labels and corresponding guide markers. An old print guide may still circulate; the internal status should not suggest every copy is physically corrected. It should identify the authorized updated outputs and known remaining limitations.
This case demonstrates why cataloguing a family of images or objects can be efficient while public labeling must still be specific about its claims. The right fix isn't always to discard any representative art. It is to make its actual role legible and stop it from being treated as evidence of the comparable object's unique features. A small difference in wording can preserve both usability and historical accuracy.
Choose verification evidence that respects the exhibition
A strong check does not require a dramatic close-up. The least expensive evidence may be a written confirmation from authorized staff who installed the case, paired with an independently observed label reference. Together these answer distinct questions. A wide smartphone image might show that the objects are present but omit the words or identifying cues. An approved object image might verify a place in the collection without showing the current display. Prefer the combination that actually links object, position and caption in the intended time, rather than the one that produces the most impressive file.
Document the way each observation was made. A label revision may be read on paper without having clear evidence of installation. A witness may see a public display but be unable to identify one of the bodies from normal viewing positions. A movement receipt can give the expected physical association but not the viewpoint from which a visitor will identify it. Pairing those evidence types makes the audit reliable without collecting unrelated media. If the object identity was confirmed by a custodian, acknowledge that, and do not describe it as your own direct physical identification.
Design a picture-based check around nonidentifying features and safe access. A frame can show a published object silhouette and label grouping without showing faces, private household documents or staff-only labels. Avoid capturing reflective views of people if the project does not require them. A small pointer or digital annotation on an authorized working copy can highlight the connection without altering the exhibit. Label any added pointer so readers do not mistake it for visible on-site signage. If the underlying picture can't be shared appropriately, use a concise text receipt and keep any protected support where institutional rules permit.
A reliable note can combine evidence requiring different levels of access. Public proof may be expressed through a case letter, visible trait and displayed wording. Internal proof can associate that case letter with the institution's managed item identifier and movement record. No need to publish a list of protected item identifiers, exact storage coordinates or full internal documents to prove the mapping. Think about what the visitor actually needs to know: these words identify the round-base item here, and the guide's image is a properly identified representation. The appropriate collection reviewer can examine the deeper proof if questioned.
Check an exhibit in the order a first-time visitor encounters it
The curator may already know which objects are contrasted, where the graphics came from, and which similar images don't matter. A visitor enters without that knowledge. First impressions can be shaped by the nearest label, the start of an audio programme, or a sign pointing to the far side of a case. Ask an authorized colleague unfamiliar with the draft to take one ordinary route, or perform a controlled review from equivalent viewing positions. Record exactly what drew their attention to each pairing. If they saw the same label from two sides but assumed it applied to two different objects, investigate the cue rather than merely telling them the answer.
A small exhibition can support choices at several levels. Entering through an orientation map identifies the display as a group. A local sign distinguishes subthemes. A separate caption explains a unique mark. Make sure combinations of cues don't conflict. A common group heading can be displayed over everything while captions below show individual examples, but on a cramped reference panel the visitor can confuse the paragraph arrangement. Adding item-level cues where needed can help, provided they are in the same small area and aren't only color differences a reader must decipher under poor conditions.
Avoid making the review feel like a recitation of the author's design. Ask what the letter, arrow, number or page title identifies before naming the intended work. If no single cue makes that clear, don't score uncertainty as reviewer failure. It is evidence about how the installation communicates. A rewritten caption that includes a distinguishing fact near the main noun, or a diagram that reproduces verified placement, can reduce ambiguity without crowding the object with more labels. An audit can note navigation and presentation problems while leaving broader accessibility assessment to the appropriate people.
During a busy open period, a reviewer may not be able to repeat every approach. Respect actual operations and note the chosen route. Another check when the exhibit can safely be reviewed might test a second direction, an alternate input method on the guide or a larger-text view. Don't promise every user has the same experience after one quick visit. Describe the accessible paths or views actually tested. That limitation supports better triage: a failed first path can be addressed immediately, while paths beyond the sample stay distinct until verified.
Inspect the medium that can return an older answer
Print, digital and audio deployments can proceed on very different clocks. An approved text might reach the CMS before the printed label is replaced. A corrected web page might be live before an older packaged guide or downloadable language version is regenerated. A physical position may stay correct while an audio stop retains last season's arrangement. Write the expected surfaces ahead of time and note who can verify each one. Then don't let the date of the latest staff change become a blanket claim about everything readers can open.
For a synthetic timing test, imagine a planned guide release is dated Tuesday and a case inspection is dated Thursday. Thursday's visit sees the new label beside the correct item. The guide link still displays the wording of the Tuesday release, even though a new source document exists for Thursday. The installed change and guide discrepancy can both be stated without inferring which mechanism served the earlier content. A visitor can encounter stale content through a downloaded resource or ordinary linked path; on-site status should reflect what was actually observed. Record the exact user-facing result and date and ask the responsible owner to follow the established delivery process. There is no need to alter device settings or defeat ordinary controls to diagnose the source of a stale view.
Scope evidence deliberately. If all the guide stops use the same local coordinate-to-object list, changing one could affect nearby entries. If text is inserted independently for one label, check that label and a plausible neighbor rather than broadening indiscriminately. Good verification is about the likely consequences of the change. You do not need a lasting media archive of each tiny edit if a small dated text or approved photo can show the relationship was repaired. But you do need a check sometimes done by someone other than the, understandably, optimistic author of the new proof.
An especially helpful carryover is a label revision key represented in the staff record and a display-specific observation of an exact distinctive phrase. If anonymous screenshots don't show trustworthy timing, attach a role-appropriate check record. If a previous printed copy remains available and has meaningful errors, keep the correction with the current guide where useful but do not imply every copy is recoverable. You can say the current case and linked guide paths were checked at a known time. That is a substantive promise with evidence, and it is quite enough for most readers.
Consider whether a caption must be parked, narrowed or clarified
Not every mismatch has the same cost. An uncertain label pairing that attributes a unique maker's mark to a different artifact can misstate the history. Two broad work-level descriptions that are true of both items may be merely hard to distinguish. A missing indication that one item is a copy can undermine the point of the exhibit. Prioritize the change that would materially affect a reader's understanding, and be clear whether the evidence supports a correction or only calls for review. Don't create an entirely new label interpretation just because the first mismatch was clearly visible.
A temporary fallback can be a narrower generic panel that is true of a named group and removes an unverified individual statement, if the appropriate team approves it. Another possible outcome is leaving a placeholder caption unpublished until authority and sources catch up. The audit does not make these changes on its own. Report what the team can safely verify now and what someone responsible would need to check. A precise text change usually costs less than a hurried reinstallation based on guesses about two similar objects.
If a difference is intentional, label its intent. A digital record may retain a representative image even after actual exhibition placement changes; this can be fine where the relation is named and no individual detail is attributed to the representative view's subject. A physical display may use a substitute after an original is removed for an approved reason; a visible label can say which kind of object is presented. The verification question becomes whether the intended distinction is genuinely readable in context. It is not a defect that museums use series-level descriptions. It is a defect when the distinction disappears and audiences receive a confident claim about the exact thing on view.
Make the after-check distinguish actual correction from planned correction
A durable correction note says what the mismatch was, what change was approved, and what was observed afterward. Avoid reporting “fixed” simply because a designer exported a revised total. There is a difference between generating the revised words, installing them, observing that they refer to the right object, and checking any separate linked channels. Give remaining exceptions explicit owners or next steps under existing procedures. “Awaiting verification of panel after printing” is a result an editor can pick up.
Retest at least the failing path and a nearby case if the change can impact it. Swapping two guide stops can fix one mismatched item and make another dangling. Moving a label across a case can leave a group title harder to associate. Use the same meaningful observable cues in the comparison where possible. If a direct verification isn't available, record the appropriate receipt and distinguish it from a firsthand observation. A photograph from an authorized colleague can show the correct placement under relevant conditions; it should not be described as if you personally watched the process.
A later check can be cheap if the notes are clear. Keep a short record of the exact phrase or object trait that distinguishes the cases, which version was observed, the viewing route, the evidence type and a date. There's no need to circulate full protected inventory material just so another reviewer can see the portion relevant to the mismatch. Link it within the appropriate repository when staff need it. A well-designed short receipt should let someone recheck the link without taking over another staff member's broad access rights.
If a reference illustration becomes the centerpiece of a new display, repeat the object-identity and representation check before treating older approval as sufficient. The same image can play a different role in a new caption. The archive team can decide it is correctly representative of a production process and still find it inadequate for a statement about a particular surviving example. Label those roles explicitly. A visitor should never have to infer the function of a photograph solely from its position or its visual prominence on the page.
An effective test of these notes is to hand them to a colleague who hasn't read the proposed fix. Ask which display sentence to verify and what actual evidence would confirm it now. The correct reply should refer to a specific installed relationship, not a color or an internal render timestamp. If it doesn't, add whichever fact was lost: the exact object, visible position, actual text, or how the two corresponded. Those are small pieces of evidence, but keeping them together is what lets a museum say the label matches the thing a reader sees.
Editorial note: Drafted with AI assistance for Apiary. All museum workflow outcomes, item examples and exhibition records are synthetic.