Institutions translating large IIIF collections eventually face a question that is more complicated than translating one Manifest: how do other systems discover that a record has changed? A reliable approach to IIIF Change Discovery API 1.0 localization preserves Activity Streams events, OrderedCollection paging, resource identifiers and timestamps while allowing multilingual metadata to be updated responsibly. Searches for IIIF Change Discovery API 1.0, IIIF Activity Streams harvesting, multilingual IIIF update feeds and IIIF deleted Manifest synchronization all point to the same concern: keeping distributed collections accurate over time.
Imagine a digital library whose French labels are revised while an international aggregator continues showing the older wording. Or a museum that withdraws an object from the public web but leaves its image visible in a downstream search portal. The problem is not that one translator chose a weak synonym. It is that systems need a dependable way to learn that the source resource has been added, updated or removed, and to know exactly which item the event concerns.
This guide explains the Change Discovery model in plain language: Activities, OrderedCollections, OrderedCollectionPages, resource links, publication times, deletion events and the practical meaning of updates to multilingual IIIF records. It focuses on change communication and synchronization rather than translating individual Manifest labels, conducting repository-wide OAI-PMH harvesting or moving files through preservation packages. The goal is an understandable system of changes that preserves provenance and does not confuse a translated description with a new cultural object.
Digital collections do not stand still
A museum catalogue is often treated as a finished list. In practice, it keeps changing. A newly digitized painting appears. A historic name receives a correction. A community contributor adds a language-specific label. A curator changes an attribution from ‘by’ to ‘attributed to’. A rights statement is amended. An archive withdraws an image from public access in response to an ethical concern. Each event changes what a visitor or partner system should understand.
A partner platform cannot guess these changes simply by remembering the last version it downloaded. It can repeatedly fetch everything, but that becomes wasteful across many institutions and large collections. A change discovery feed offers a more targeted signal. It tells interested systems that something relevant happened and provides the affected resource’s identity.
For multilingual institutions, the key lesson is that a language update is still a change to a resource. It may be small compared with replacing an image, but it can materially alter a reader’s understanding. The event should allow consumers to retrieve the current resource without losing its identity or confusing an editorial revision with a second object.
Change discovery is not the same as searching the collection
A search service answers a user’s question: which annotations contain this phrase, or which objects fit a query? A change discovery service answers a different question: which IIIF resources have been created, modified or removed since a consumer last checked? The consumer is usually another application, such as a federated cultural portal, research index or aggregation service.
This difference matters in translation projects because both may display language-related data. A search result can contain a translated label, while a discovery Activity identifies an object whose labels were changed. Translating a Change Discovery feed as if it were a public exhibition page misses its function. Most of the feed is machine-readable structure that tells software how to keep a collection current.
A human-facing dashboard may summarize Activities in several languages. That interface can translate ‘updated’ or ‘removed from publication’. The underlying type values and resource references still need to follow the standard. The useful distinction is the same as in a timetable: the signs can be bilingual without changing the train’s actual departure time.
What the Change Discovery API describes
The IIIF Change Discovery API 1.0 uses an Activity Streams pattern. A publishing institution provides Activities describing changes to resources. Those Activities are arranged within OrderedCollectionPages, which are associated with an OrderedCollection. Consumers can navigate the pages and learn about resource changes without retrieving the complete collection repeatedly.
An Activity has a type that describes what happened and an object that identifies the affected IIIF resource. Types used for ordinary resource lifecycle changes include Create, Update and Delete. A corresponding timestamp can help consumers determine what to process. In the resource object, the URI matters because it points to the specific resource the downstream system recognizes.
These are protocol distinctions, not English sentences awaiting translation. A JSON value of Update should remain Update where required by the specification. The French label in a monitoring dashboard may read ‘modifié’, but the feed’s fixed event type should not be rewritten. The purpose is shared interpretation by independent systems.
The OrderedCollection is the map of a changing archive
An OrderedCollection gives consumers a route through a series of change events. It does not need to contain the entire list of Activities directly. Instead, it references pages. The specification describes first and last page links, with the last page especially important for clients that begin at the newest activity and walk backward toward previously processed events.
A translator who sees ‘first’ and ‘last’ in a JSON document might think of ordinary words describing museum rooms. Within this structure, they are property names that software expects. Replacing them with translated keys would disrupt parsing. A visitor-facing explanation of the most recent change can be localized, but the mechanical links remain as defined.
The benefit of paging is scalability. A large institution can produce many changes while giving consumers a controlled way to fetch them. The pages are part of an event history, not a translated table of contents. Their identity and ordering are technical facts that must survive changes to the descriptions of the objects being referenced.
OrderedCollectionPages preserve the sequence
An OrderedCollectionPage contains a set of Activities through its orderedItems property and can link to neighboring pages. Consumers use those references to move backward or forward through the feed. Within pages, the ordering of Activities is meaningful. In the standard’s model, processing recent changes can involve starting at the last page and moving backward until an already processed interval is reached.
This means that a human-language edit should not reorder the event list. Imagine a Spanish title update occurring after an English description correction. If a feed is accidentally regenerated in alphabetical title order, downstream systems may miss the correct temporal sequence. That is a synchronization error, not a translation improvement.
Time and order also help with recovery. A consumer that was offline for a day needs to locate the events it missed. A coherent set of pages makes this possible. If links point to missing pages, the software cannot reconstruct what happened reliably. The localized public narrative should be built on top of that intact sequence.
Create, Update and Delete are different historical claims
Create communicates that a resource has been made available. Update communicates a change to an existing resource. Delete communicates that a resource was deleted or depublished according to the event semantics. The differences are important because a consumer may index a new item, refresh a known item or remove its public entry depending on the event.
If an archivist adds an Italian label to a Manifest that already exists, calling the event Create can make an aggregator treat it as a new object. If a resource is withdrawn and its event is ignored, the aggregator may continue directing users to outdated content. If a title correction is recorded as a Delete, a consumer may discard a legitimate record unnecessarily.
The human explanation of these events can be nuanced. ‘Removed from public access’ may be clearer than ‘destroyed’ when a resource has been depublished but preserved internally. A translator must not collapse deletion from a web feed into the destruction of a physical object. The event describes a change in digital availability, not necessarily an event in the object’s physical history.
Timestamps are measurements, not captions
The specification uses machine-readable time information such as endTime for Activities. Consumers may rely on timestamps to determine whether an event falls within the period they need to process. The timestamp’s format and timezone semantics are part of that technical function. Translating the date into a prose phrase inside the protocol would break comparability.
A human interface may display a time in the visitor’s local timezone or in a familiar date format. That presentation is useful when done carefully. However, the underlying event time should remain available for verification. Converting a timestamp twice or confusing local time with UTC can shift an event across a date boundary. In an hourly or daily synchronization job, such a change can cause missed or duplicated processing.
An institution should therefore distinguish the moment an editorial action happened from the time the change event was published, if its system tracks those separately. Descriptions such as ‘updated yesterday’ are context-dependent and age quickly. A precise date and time is generally better for audit and scholarly provenance.
The object identifier anchors the whole event
Every meaningful change event needs to identify the affected resource. The object URI is the connection between the change feed and the Presentation API resource that a consuming portal has indexed. When a title is translated, that URI should ordinarily stay the same. The reader may see new wording, but software should still recognize a revision to the existing object.
Consider a Manifest whose identifier is https://archive.example.org/iiif/manifests/record-42. That illustrative URL stands for a stable reference. The archive can add language maps, improve a summary and correct an attribution without generating a fresh unrelated identifier for every language. The change event then points back to the same resource.
Sometimes a legitimate repository migration does change URLs. That is a separate preservation and linking challenge. Language localization should not introduce an unnecessary migration. Keeping the same URI when the object remains the same makes it easier to preserve citations, annotations, bookmarks and federation records.
Multilingual labels are content, not event types
A Manifest may expose language maps for its title, summary and other human-facing descriptive content. When a curator publishes a new translation, the resource has changed even though no media file has been replaced. A downstream system needs to retrieve the updated Manifest and decide how to present the newly available language.
The discovery feed does not need to carry every translated phrase to be useful. Its core responsibility is to identify the changed resource and the event. The consuming system can then fetch the current resource and inspect its language maps. This division keeps change communication distinct from the content being synchronized.
A practical consequence is that translators should not edit the feed’s machine vocabulary when adding localized descriptions. The resource’s user-facing properties are the correct destination for the new language. The event system tells others that the resource now has a revision worth retrieving. Both steps are needed if multilingual coverage is to appear consistently across a distributed ecosystem.
A corrected translation and a new edition are not always the same thing
Not every change has identical editorial significance. Correcting a spelling mistake in a translated summary is usually an update to the existing resource. Publishing a distinct scholarly edition with a different interpretive structure may require a more explicit relationship to the original object. The decision belongs to the institution’s editorial and data model, not to an automatic assumption made from language alone.
The change feed should communicate the actual resource lifecycle. If the same Manifest is revised, the event should identify that Manifest. If a genuinely new Manifest is published to represent another curated object or edition, it may have its own identity and creation event. The difference affects how aggregators group results and how scholars cite evidence.
This is especially important for collections with multiple translations created at different times. A French summary added in one year and an Arabic description revised in another are part of the record’s evolving history. They should not silently multiply one object into several unrelated records solely because language versions exist.
Provenance requires more than a changed title
A change event tells another system that something happened, but it does not automatically explain every reason behind the change. An institution may separately record editors, evidence, language reviewers, sources and approvals. Those details support provenance and accountability, particularly where a translation modifies the interpretation of contested history.
For example, an attribution might change from ‘probably made in’ to ‘made in’ after new archival evidence appears. A simple Update event can prompt a portal to fetch the revised description. The editorial record should still document why the institution strengthened the claim. Without that provenance, later readers may not be able to tell whether the change reflected new research or an accidental translation.
An effective system therefore distinguishes event discovery from the evidence trail. The feed communicates the existence of a change; the resource and linked editorial records explain its meaning. Translators should not invent editorial justifications inside technical event fields. They should ensure important qualifications survive in the human-readable record.
Deletion needs particular care
A Delete event may indicate that a resource is no longer published at its former location. For a downstream system, ignoring that change can be serious. The index may continue to show metadata that has been withdrawn, perhaps for privacy, rights, cultural sensitivity or factual error. A portal should have a defined response to deletion rather than treating it as a temporary fetch failure.
The human-facing wording must reflect what is actually known. ‘This resource is no longer available here’ is not the same as ‘the artwork was destroyed’. An institution may keep a preservation master offline or may replace a public record with a new access statement. Translation should preserve these distinctions and avoid exposing reasons that the institution has not authorized for publication.
Deletion can also affect multilingual caches. A French catalogue might remove an entry promptly while an English-language portal retains its older copy. The change feed helps federated consumers coordinate their response. A test should therefore include every published language interface, not just the default one.
Updates and caching can produce mismatched languages
A visitor may retrieve a new Manifest from the source institution while an aggregator still shows an older translated summary. A cache or search index can lag behind the updated resource. The two pages then disagree, sometimes in ways that look like mistranslation. In reality the systems are showing different revisions.
A robust synchronization process checks the change feed, retrieves the affected current resource and refreshes the relevant cached representations. It should not assume that only the language selected by the visitor needs to be updated. A change to the resource’s language maps can affect several interfaces at once, including fallbacks when one preferred language is unavailable.
When discrepancies appear, reviewers should compare resource IDs, revision information where available, and event times. They should not silently overwrite one source with an older translation just to make the pages agree. The goal is to identify which representation is current and preserve a record of why it changed.
Why reverse traversal matters
Change Discovery’s ordered pages are designed so a consumer can begin with the latest part of the activity history and work backward until it reaches events already processed. This can reduce unnecessary work for systems that poll regularly. It is a technical approach to incremental synchronization rather than a rule about how an exhibition should be displayed.
The distinction matters because translated human interfaces may list ‘newest first’ or ‘oldest first’ according to visitor needs. That UI preference should not determine the structure of the underlying feed. The consumer’s traversal logic depends on the actual page links and ordering, not on a translated heading.
Testing should include more than one page of events. A feed with only one page can hide errors in next or previous links. Add a realistic series of creation, update and removal events, then verify that a consuming application can recover the relevant interval without skipping or permanently repeating events.
A missed event and a repeated event are different failures
If a consumer misses an Update event, it may continue showing outdated information. If it processes the same event twice, it may fetch an unchanged resource again or generate unnecessary work. Depending on the consumer’s design, repeated processing can sometimes cause duplicate notifications or inconsistent records. The objective is reliable processing even when the network is imperfect.
Systems should keep a durable processing position or equivalent checkpoint and recover carefully from interruption. A checkpoint should describe actual processing progress, not merely the time the latest feed page was requested. If the consumer crashes after reading an event but before updating its index, it needs a strategy for retrying safely.
For translation projects, this reliability affects whether new languages appear consistently. Publishing a Spanish summary is only part of the work. If the federation misses the corresponding update, readers elsewhere may never see it. Change discovery connects editorial care to actual distribution.
Network errors should not be translated into false object states
A temporary network failure can prevent retrieval of a Manifest even when the source institution still publishes it. A downstream portal should not automatically treat that failed fetch as a Delete event. The distinction between an explicit deletion signal and a technical retrieval problem protects records from accidental disappearance.
The same principle applies to localization of status notices. ‘Unable to refresh this record’ is different from ‘the institution removed this item’. A visitor-facing service message should describe the verified state. A system log can provide technical detail for support staff, while the public message remains clear without exposing implementation internals.
This matters most in international networks where maintenance windows, connections and caching behavior differ. Distributed collections gain resilience when clients distinguish absent data, temporarily unreachable services and authoritative deletion events. Translation should make those categories understandable, not collapse them into one dramatic claim.
Restricted resources still create discovery responsibilities
Some IIIF resources can be discovered publicly even when their media require authorization. Others are restricted more fully. A change feed needs to respect the institution’s disclosure policy. It should not reveal protected descriptions or identifiers beyond what the institution has decided to publish. The discovery specification and the authorization policies work alongside one another.
A translator working on multilingual descriptions should not assume that adding a new language makes restricted information public. The decision about what metadata may be exposed is an access and governance question. If a public record is withdrawn, downstream systems must have a way to respond that does not rely on preserving the withdrawn text in every language.
For a community-sensitive archive, careful synchronization can be an ethical duty. Withdrawal from a source website is not sufficient if copies of descriptions remain in federated portals. A disciplined change feed and a responsible consumer response help reduce this risk, though they cannot by themselves recall every external copy.
Change feeds and OAI-PMH serve different purposes
Libraries often use OAI-PMH to expose metadata records for harvesting. IIIF Change Discovery offers an Activity Streams-based way to describe changes to IIIF resources. Both can participate in a repository’s wider discovery architecture, but they should not be treated as identical. Their records, requests and processing assumptions differ.
An institution may already have OAI-PMH for catalogue metadata and use Change Discovery so IIIF-aware consumers can learn about Manifest lifecycle events. The two systems need coherent resource identities and editorial policies, but one does not simply become the other through translation. A resumption token from one protocol is not a page link from the other.
For multilingual collection work, this distinction reduces confusion. Translating cataloguing terms belongs in the relevant metadata record. Translating human-readable Presentation labels belongs in the IIIF resource. Announcing that the resource changed belongs in the event stream. Each answers a different question for the people and machines using the archive.
Change feeds are not file preservation packages
A preservation package such as BagIt describes files and their integrity within a transfer. A Change Discovery Activity describes a lifecycle event concerning an online IIIF resource. The fact that both use the word ‘manifest’ in some contexts can invite confusion. A file inventory and a Presentation Manifest are not the same thing.
If an image is replaced with a newly digitized version, a preservation workflow may track file checksums and provenance, while a Change Discovery feed tells aggregators to revisit the public Manifest. Those responsibilities complement each other. Translating a BagIt metadata note is not the same as updating an Activity type, and a successful checksum check does not guarantee that downstream visitors have the current language description.
Thinking in terms of purpose helps: preservation proves what files were managed; presentation explains what viewers should display; change discovery signals when a published resource’s state has changed. A trustworthy international collection benefits when each part remains precise.
Worked example: adding a new language to a manuscript collection
A university archive has one hundred digital manuscript Manifests, each with English labels. A project adds verified Arabic summaries to twenty of them. The original objects, Canvases and image services stay the same. The editorial team wants international catalogues to display the new summaries when visitors request Arabic.
The source institution should publish the updated Presentation resources and expose appropriate Update Activities for the affected Manifests. A consuming portal that processes the events can retrieve those resources and refresh its local descriptions. It should not create twenty duplicate manuscripts merely because the new language is present. The resource identifiers remain the anchor.
A meaningful test chooses a Manifest whose summary was changed, follows its Update event and checks whether the portal retrieves the current language map. Then it checks an unchanged Manifest to ensure no accidental alteration. The result is a new way to understand existing evidence, not a fictitious expansion of the physical collection.
Worked example: correcting a mistaken date
A national museum discovers that a collection object was catalogued as created in 1882, but the documentary evidence supports 1892. The change affects the source description and its translated summaries. An international portal still displays the former date, while the museum’s new website shows the corrected value.
An Update event should allow the portal to learn that its cached record is stale. The local editorial history can document the evidence behind the correction. The translator’s job is to preserve the distinction between a confirmed new date and an approximate dating range if that is what the sources establish. The change feed must not be rewritten to hide the earlier version.
Reviewers should inspect the old and new visible descriptions, the event time and the actual resource ID. They should check whether every supported language shows consistent dating after the update. A single-digit correction may be small typographically but large historically, so synchronization is part of factual quality.
Worked example: removing a sensitive photograph
A community archive publishes a photograph under an agreement that is later reconsidered. Its governing group decides to withdraw the public resource. The source website stops displaying it, but a multilingual aggregator continues listing translated captions and a cached preview.
The appropriate event communicates the resource’s deletion or depublication according to the archive’s policy and implementation. The aggregator must handle the event responsibly, removing public access as required. A translated public notice should not imply that the physical photograph was destroyed or invent private reasons for the decision. It can state that the digital item is no longer available from the source.
Testing should include links from several language-specific portals, old bookmarks and cached results. No change feed can force every unrelated copy on the web to disappear, but coordinated consumers can at least stop continuing an outdated publication. This is where technical reliability and cultural responsibility meet.
Worked example: a catalogue with two historic place names
An archive updates a multilingual Manifest after historians determine that two place names referred to the same district at different times. The new English and local-language descriptions explain the historical relationship rather than replacing one name everywhere. The external discovery portal should show that nuance after it refreshes its indexed record.
The change event only needs to identify the affected resource and the fact of an update. The explanatory note belongs in the Manifest and its linked editorial records. If a consumer tries to generate its own translation from an old cached title, it can reintroduce the very confusion that the update corrected.
Reviewers compare the source’s current description with the portal’s refreshed view and follow the same stable resource URI in both. They also confirm that a search for the historic name still leads to the right object where the index supports such discovery. A successful update preserves the distinction between naming and object identity.
Worked example: a film with new subtitles
A moving-image archive publishes a new subtitle translation for a documentary. The media file and time-based Canvas remain unchanged, but additional annotations and language descriptions are now available. A partner education platform wants to let users select the new subtitle language without rebuilding its entire catalogue.
An Update event can prompt the platform to retrieve the revised resource. The platform should preserve existing links to the film and the relevant time positions. It should distinguish a new subtitle stream from a new film object. If the platform cached prior annotation resources, it must refresh the parts relevant to the revision according to its own retrieval design.
A useful test opens the same timestamp before and after the update and verifies that the correct subtitles appear without shifting the timeline. This case illustrates why a seemingly small linguistic change can require reliable resource synchronization even when the audiovisual master has not changed.
Worked example: an aggregator recovering after an outage
An international portal loses network connectivity for several hours. During that interval, one museum creates new Manifests, another updates labels and a third withdraws two records. When connectivity returns, the portal must catch up without assuming that everything it remembers is still current.
The OrderedCollectionPages and their ordering give the consumer a route through the changes it missed. A sound checkpointing process helps it identify the relevant interval. It should apply creation, update and deletion according to each event’s meaning. If it simply searches for new titles, it may miss withdrawals and corrections.
Reviewers can recreate a short outage in a test environment and compare the resulting index with the publishers’ current state. They should include translated labels in the comparison. A portal that restores English titles but forgets updated multilingual descriptions has not fully recovered its service.
Worked example: an update close to midnight
An archive in one timezone publishes a translation correction close to midnight local time. The Activity’s machine-readable time is expressed in a defined format, while the dashboard shows dates in the visitor’s preferred locality. One viewer sees the change on Tuesday and another sees it on Wednesday.
This is not inherently an inconsistency if the instant is the same. The danger is converting the timestamp incorrectly or treating a localized date string as the authoritative processing key. A feed consumer should use the underlying machine-readable time, not a translated dashboard label, to determine its processing window.
Test with events around date boundaries and daylight-saving transitions where relevant. Compare the original instant with localized displays and verify that no event disappears from an hourly or daily synchronization window. Translation can change presentation; it must not alter chronology.
Worked example: many small translations in one batch
A major gallery adds translations of object captions in stages. Several language teams finish revisions on the same day. The repository emits multiple updates. A portal consumes them gradually while visitors use its search and collection pages.
The portal should avoid treating each language update as a new artwork. It can refresh the current resource and any derived language-specific views affected by the change. Repeated events may cause additional retrieval work, but they should not generate duplicate catalogue identities. Reviewers should inspect both the machine index and the displayed exhibition.
A good acceptance test includes two consecutive revisions to the same Manifest. After processing, the portal should show the latest approved labels, not a mixture where the newer French text appears alongside an older Spanish description that was also revised. Chronology and identity together support accurate multilingual publication.
Worked example: an item that moves repositories
A research collection migrates to a new institutional platform. Its technical team must manage resource links, redirects and discovery events. Translators are also updating descriptions, creating a temptation to combine every change into new language-specific URLs.
That approach can multiply the migration problem. Resource identity changes need a deliberate technical and preservation decision. Language edits should be managed separately, with clear relationships and redirects where necessary. The change feed can communicate relevant lifecycle changes, but it is not a magic replacement for maintaining links or documenting the new location.
Reviewers should start from an old citation and follow the intended route to the current resource. They should also check whether translated labels remain attached to the right object after migration. A successful move preserves intelligible provenance and minimizes broken scholarly references.
A useful synchronization checklist
Start by verifying the identity of the publisher’s OrderedCollection and its last page. Follow page links and inspect their order. Confirm that Activities have meaningful types and identify the affected resources with stable URIs. Check that timestamps can be interpreted consistently and that a consumer can traverse more than one page without losing position.
Next, create representative change events: a new resource, a corrected title, a newly added translation, an access-related withdrawal and a temporary fetch failure that is not a deletion. Confirm that a consumer responds differently to each case. An update should refresh a record; a deletion should trigger the defined removal behavior; an ordinary network error should not create a fictitious Delete.
Finally, compare multilingual views after synchronization. The source institution and partner portals should agree on the current resource identity and material description. Differences in interface phrasing may be legitimate, but outdated facts, missing language versions and stale withdrawn entries deserve investigation. A test passes when the systems represent the current record accurately.
What to record when investigating a discrepancy
When one portal shows an old translation, the first question is which resource URI it has indexed. The next is whether the publisher emitted a relevant Activity and whether the consumer processed it. Then check whether the updated Manifest or annotation content was actually retrieved. Only after those facts are established should the team focus on a possible language rendering problem.
Useful evidence includes the event type, resource ID, event time, collection page, consumer checkpoint and the current source record. Keep these machine facts separate from human-language display strings. If a dashboard says ‘updated’ but the underlying event is Create, the interface may mislead support staff; if the event is correct but the dashboard translation is weak, the underlying synchronization may still work.
This disciplined sequence prevents a common mistake: asking translators to fix words when the real problem is stale data. It also prevents developers from dismissing genuine editorial errors as caching issues. Good diagnosis respects both the technological and linguistic dimensions of the record.
Common mistakes to avoid
Do not translate fixed event type values such as Create, Update or Delete within the protocol. Do not change OrderedCollectionPage property names to match interface language. Do not reorder events alphabetically by translated titles. Do not treat a date displayed in local time as a different moment from the underlying timestamp. Do not assign new Manifest identities to every additional translation.
Do not equate a Delete event with destruction of a physical object. Do not treat every failed network fetch as deletion. Do not assume a multilingual portal is current merely because its default-language page has refreshed. Do not overwrite documented editorial history in order to make a cached description appear consistent.
Most importantly, do not forget that a change feed is a public relationship between institutions and their consumers. Its value depends on stable contracts and dependable interpretation. Translation can improve the understanding of changes, but should never silently rewrite the machine facts that make those changes discoverable.
Frequently asked questions
Does Change Discovery translate IIIF labels itself?
No. It communicates changes to IIIF resources. Multilingual labels and descriptions belong in the appropriate source resources. Consumers can retrieve the revised content after an Update event.
What is an OrderedCollection?
It is the container that organizes a feed of change activities through linked pages. Clients can use its page references to find and process events rather than download the entire repository repeatedly.
Why does a client start near the last page?
The model supports processing recent Activities by beginning at the newest part of the feed and moving backward until the client’s previously processed interval is reached. Correct order and page links are essential.
Should the activity type be translated?
Not in the machine-readable event. The protocol values have defined meanings. A human dashboard may display a localized equivalent while preserving the original event type in the data.
Is a translated title correction a new object?
Usually not. If the original Manifest still represents the same object, the correction is a revision to that resource. Its URI can remain stable while descriptive text changes.
Does Delete mean the physical object was destroyed?
No. A Delete Activity concerns the availability or publication of a digital IIIF resource under the model. It should not be presented as evidence of physical destruction.
Is this the same as OAI-PMH?
No. Both can support discovery-related workflows, but their structures and purposes differ. OAI-PMH harvests metadata records; Change Discovery communicates lifecycle changes to IIIF resources through an Activity Streams pattern.
How can a library test multilingual synchronization?
Publish controlled examples of creation, translation updates and withdrawal, then process the feed in a test consumer. Verify that resource IDs remain consistent and that all supported language views reflect the intended current state.
The principle worth remembering
A translated record becomes truly useful in a distributed collection only when other systems can learn that it changed. IIIF Change Discovery provides an interoperable signal: which resource was created, updated or removed, and how a consumer can follow that history. The feed’s technical values are not prose to rewrite; they are the contract that keeps institutions synchronized.
Good localization preserves that contract while helping readers understand the current descriptions and why those descriptions may differ from older copies. The result is not only a better translation. It is a more dependable cultural record across languages, institutions and time.
Sources and further reading
International Image Interoperability Framework, Change Discovery API 1.0: https://iiif.io/api/discovery/1.0/
International Image Interoperability Framework, Presentation API 3.0: https://iiif.io/api/presentation/3.0/
International Image Interoperability Framework, Content Search API 2.0: https://iiif.io/api/search/2.0/
Open Archives Initiative, Protocol for Metadata Harvesting: https://www.openarchives.org/OAI/openarchivesprotocol.html
Related eduKateSG guide, OAI-PMH 2.0: https://edukatesg.com/2026/10/30/translate-like-a-pro-3430-oai-pmh-20-metadata-harvesting-sets-datestamps-resumption-tokens/
Related eduKateSG guide, BagIt RFC 8493: https://edukatesg.com/2026/10/30/translate-like-a-pro-3440-bagit-rfc8493-archive-packages-manifests-checksums-metadata/
Related eduKateSG guide, IIIF Presentation API 3.0: https://edukatesg.com/2026/10/30/translate-like-a-pro-3490-iiif-presentation-api-30-manifests-canvases-ranges-language-maps/