November 2026 Releases

The query processor ignored the -sortall=true setting when checking tier-based sorting.

The engine enforced tier-restricted sorting instead of honoring -sortall=true.

Updated sort within tier setting to consider the -sortall value when deciding whether to disable tier-restricted sorting.

When -sortall is enabled, return the fully sorted result set per the -sort criteria and omit tier bars.

Under certain circumstances Funnelback data sources became unsearchable due to missing files or corruption. With this improvement one can determine which data source and its file is the cause.

An error message containing information to trace the missing file is now added to the log files.

On pages with a title followed by a link, PADRE could add an extra full stop in the summary, e.g. June 2020. . Download.

Summaries showed extra full stops which was not present in the source HTML.

Summaries no longer show a duplicate full stop when a title is followed by a link. Existing search indexes are corrected at query time; new crawls no longer store the extra break.

Summaries show a single full stop.

The results page switcher in the left-hand panel of the Insights dashboard displayed results pages in the order returned by the API, with no sorting applied to the displayed labels.

Consequence: Customers with many results pages found it difficult to locate a specific page in the switcher, as the list appeared in an unpredictable order.

Results pages in the switcher are now sorted alphabetically by their displayed label (the results page title, or the profile ID when no title is set). Sorting is case-insensitive.

Customers can quickly find and switch between results pages in the Insights dashboard, especially in environments with a large number of configured pages.

Data sources external metadata mapping configuration files can contain ill-formed URLS.

When indexing a Push data source, Padre could error when an il-formed URL was processed.

Now Padre is less strict in parsing external metadata mapping configuration files.

Customers will now see metadata mappings that matches their results.

The default Funnelback search results template (simple.ftl) displayed facet value counts for checkbox facets even though checkbox facet counts are disabled by default in the DXP. This mismatch could confuse users configuring checkbox facets on new results pages.

The default template has been updated so count badges are not shown for checkbox facets. Count badges continue to appear for other facet types, such as radio button and drill-down facets.

Results pages that use a customised template copied from an older simple.ftl are not affected automatically and may need the same template change applied manually.

Searching for a URL containing :// causes the query processor to strip the protocol separator without inserting a space.

Search terms are concatenated together (for example, https://example becomes httpsexample), causing searches containing full URLs to return 0 results even when the URL exists in the content or metadata.

The query processor now normalises URL-like queries before processing syntax. Bare URLs and URL metadata fields are correctly split along standard token boundaries to align with index tokenisation.

Full URL searches now break down protocol and domain terms as expected, successfully returning matching results.

When processing partial date metadata (such as "2000--") mapped via metadata mappings, the search engine’s date parser mishandled missing month or day fields, reading uninitialized values during indexing.

Partial dates indexed inconsistently across builds. Date metadata could store the raw source string (e.g. "2000--") while the document date used a different parsed value, causing unstable date displays, improper sorting, and artificial recency score boosts for incomplete dates.

Updated date extraction logic and introduced a new indexer option (-date_parse=fuzzy|strict), defaulting to fuzzy. Fuzzy mode populates missing day/month values with 1 (and missing years with 1970), while strict mode rejects incomplete dates as "No Date". Date metadata mapped to class d is now stored and indexed in canonical parsed form (e.g. 1 Jan 2000), not the raw source string.

Stabilizes date metadata extraction across HTML meta tags, XML elements, and attributes configured via metadata mappings. Collections using strict mode evaluate partial dates as "No Date" with zero recency boost, ensuring accurate search rankings.

Tag lookup assumed plain tag names without attributes. Attributed HTML tags failed mapping rules, causing snippet pointers to read beyond tag bounds into adjacent document text.

Content within attributed tags (e.g., <h1 class="…​">) was not indexed into custom metadata fields, while cross-document memory leaks created false search matches.

Replaced tag lookups with length-bounded string comparisons and anchored metadata snippet pointers strictly after the full opening tag.

Attributed HTML tags are correctly indexed, and cross-document metadata bleed is eliminated from search results.

Funnelback now uses the GeoLite2/GeoIP2 ASN MMDB in $SEARCH_HOME/share/ to resolve organisation names for predictive segmentation and Curator user-segment rules, replacing the legacy GeoIPOrg.dat file. GeoIPOrg.dat file is not supported by Maxmind anymore.

This avoids stale IP data coming from old datasets and produce correct/latest results about organisations in Search Analytics page.

Search requests can limit results by date using URL parameters. For each user-facing date parameter there is a matching system parameter that applies the same limit as part of the system search. Some results pages, plugins, and integrations rely on the system parameters rather than the user-facing ones.

When only the system date parameters were used, the date limit had no effect and results were the same as if no date had been set. The same date sent through the equivalent user-facing parameter filtered results as expected.

Date limits from system date parameters are now honoured using the same rules as the matching user-facing parameters.

Customers who use system date parameters, or features that set them (such as filtering to upcoming or future-dated content), get the intended date filtering without needing workarounds.

The formatting was incorrect in the change log.

The change log is now displaying correctly.

The previous JSONata library, JSONata4Java, was missing support for some key JSONata functions, including the object transform function. This limited what the Modify JSON data plugin could do.

The plugin now uses the jsonata-java library, which provides full JSONata support. Object transforms and the $formatInteger() and $parseInteger() functions are now available.

When two push collections with the same plugin are enabled attempt to commit documents at the same time the file connection is closed by one data source causing the other to error.

When this occurs during a commit the data source will stop further commits until the push data source is stopped.

Count the number of data sources connecting to a plugin file and only close connection when all data sources are finished.

Since the connection is open for all data sources they can finish the commit without error.

Success and error alert messages in the Funnelback admin UI were sanitising dynamic text using URL encoding (encodeURI). Collection and resource names containing spaces or other special characters were encoded before being displayed (for example, a space was shown as %20).

After creating or updating collections and other resources in the Funnelback admin UI, alert dialogs displayed URL-encoded text instead of the intended name. For example, creating a data source named courses singapore showed courses%20singapore created in the success message. This affected any alert message that interpolated user-provided names or values containing spaces.

Updated the admin UI to HTML-escape translation arguments for display in alert messages, instead of URL-encoding them. Removed redundant URL encoding when passing collection names to success messages, and aligned error message handling to use the same sanitisation approach.

Alert dialogs in the Funnelback admin UI now display collection and resource names correctly, including names with spaces and other special characters.

The IP detail view attempted to read the collection service ID from the URL route parameters. The route only includes the IP address (:ip), so the service ID was always undefined, causing a JavaScript error when constructing a ServiceId.

Clicking an IP address in the Top IP addresses table on the Search Analytics Locations page (or following an IP link from the analytics overview) threw an Invalid ID: undefined error in the browser console. The IP detail page failed to load predictive segmentation information for the selected address.

The IP detail component now obtains the current service ID from the active service context (ServiceService) and combines it with the IP address from the route before calling the predictive segmentation API.

Administrators can open an IP address from Search Analytics and view organisation and segment details without console errors or a broken detail view.

There were some issues with radio button functionality on plugin configuration screen. There were control "undefined" errors as a result of it.

When the radio buttons were toggled the screen did not behave as expected. Such as displaying greyed out drop down control instead of interactive one when option "yes" is selected.

Plugin config keys can be shown or hidden based on another key’s value (showIfKeyHasValue). This change makes that conditional logic work reliably when the associated key is referenced by schema ID (not just by key name), and improves value matching for booleans and arrays.

Conditional keys are behaving correctly now on plugin configuration page.

When switching from Preview to Live in the Funnelback template editor, the Admin UI navigated away from the file detail view and incorrectly treated Live as a permission restriction rather than an intentional read-only view. A readonly directive on the editor wrapper also blocked scrolling in the Ace editor.

Users switching to Live saw a misleading permission banner ("You don’t have permission to modify these settings"), an empty file listing ("No entries have been found"), or live template content that could not be scrolled. Switching back to Preview could also behave inconsistently.

The view switcher now stays on the file detail route when toggling between Preview and Live, uses forced read-only mode for Live (instead of permission-based read-only), removes the wrapper-level readonly directive so the Ace editor handles read-only correctly, and resets read-only state when leaving the editor.

Users can reliably open a template in Live view, read the published content in a locked editor, and scroll through the full file. The informational read-only message is shown instead of a permission error, and switching between Preview and Live works as expected.

The Edit raw data configuration editor submitted every parameter key to the API on save, even when values had not changed. Deleting keys from the editor also bypassed the standard delete flow, so keys were removed from preview only and were not unpublished from the live view first.

Saving configuration via the raw key editor was slow and could trigger intermittent server timeouts (504 errors) when many keys were present. Retrying a failed save often produced further errors. After saving, some keys could incorrectly appear as In live only. Removing a key from the editor did not prompt for confirmation and did not remove it from the live view when published.

The raw key editor now compares the editor contents against a snapshot taken when the modal opens and only submits keys that were added or changed. If no keys were modified, save completes without API calls. When keys are removed from the editor, a confirmation dialog lists the parameters to be deleted. Deletes now use the same service logic as the configuration table, unpublishing keys from the live view before removing them from preview.

Saving changes in the raw key editor is faster and more reliable, with fewer API calls and a lower risk of timeout errors. Deleted parameters are removed from both preview and live view where applicable. Keys are less likely to be left in an incorrect In live only state after save.

Without the cache header than a Cloudflare rule would need to be setup for every client

Cache header setup with 4hr expiry which will be utilised by Cloudflare to reduce calls to the DXP Search servers

Padre copied query text unchanged into XML comments in its diagnostic output. XML does not allow the sequence — inside a comment, so a query containing — (for example !nullquery—​x) produced invalid XML.

When contextual navigation was enabled, a search containing — returned a 500 error or "Unable to parse PADRE response" instead of search results.

Padre now converts query and document text into a comment-safe form before writing it into a comment. The second hyphen of each — is changed to _, a trailing hyphen is changed to _, and control characters are changed to spaces. This applies to all diagnostic comments that include query or document text.

Queries containing — now return search results normally.

When generating a broken links report, checking each external URL sequentially in a single request could cause the process to exceed server time limits if an asset contained a large number of external links.

This resulted in a 504 Gateway Time-out error, preventing the report from completing and causing the background process to become unresponsive.

The broken links reporting process now actively monitors time limits during execution. If validating links on an asset takes too long, the task saves its current progress and pauses, allowing subsequent background requests to seamlessly resume and complete the report.

Broken link reports now complete successfully without encountering timeout errors, even when scanning assets containing a large volume of external links.

A recent change stopped passing through the HTTP status from the upstream API when visitors open a REST Resource page directly on the public site. Error responses were often sent as a successful page load instead of the real error status.

Edge caches and monitoring tools could not tell a failed API call from a successful one. Caching rules that depend on error status did not behave as expected. Integrations that rely on standard HTTP status codes could misread failures.

Direct requests to a REST Resource page again use the upstream HTTP status when the API returns an error or the connection fails. A new Expose HTTP status on frontend option on the REST Resource lets you control this: it should be enabled when passing the status directly is desired, and disabled when your Javascript will process any errors.

Caching and observability can now use real error status codes where you enable the option, while existing Javascript-based error handling can stay on the previous behaviour when you leave the option disabled.

Mirror export always tried to persist new local asset to mirror ID mappings in sq_mirror_map. When the database connection was read only, that INSERT failed and the export stopped.

Customers running large asset sync exports saw progress freeze part way through the tree, with an incomplete export package that could not be used for inbound sync.

Matrix now keeps the mapping in memory for the export process when the INSERT hits a read only transaction error, rolls back to a save point so the database transaction can continue, and reuses the same MID for that local asset ID during the export.

Asset sync export can finish on connections that cannot write mirror map rows, so customers can complete the export and run inbound sync after upgrading.

Meta-collections were capped at 64 components. Any definition with more than 64 returned an error.

Meta-collections with more than 64 components could not be used; indexing or search reported an error when that limit was exceeded.

The component limit is raised to 1024. Meta-collections with up to 1024 components are supported, and a clear error is returned if the limit is exceeded.

Meta-collections with up to 1024 components are supported. To make room for this, the maximum number of documents in a single collection is now about 16.7 million, down from about 268 million. An error is reported when a collection reaches this limit.

Previously, the output format of a metadata date field matched the input format. A release changed that behavior so values were always formatted as d MMM yyyy (for example, 2 Nov 2026), regardless of how the date was indexed. Templates that expected ISO 8601 input — such as <h1>Metadata Date "${s.result.listMetadata.d?first?datetime.iso?string("dd MMMM yyyy")}"</h1> — then failed because datetime.iso requires an ISO 8601 date string, not the new display format.

As a result, Search pages started showing incorrect date values or sometimes no values. It also caused artificial score inflation.

The date parsing logic is reverted now. In strict mode, invalid dates now correctly evaluate to a null date without artificial score inflation. In fuzzy mode, incomplete dates (such as "2000--") evaluate to January 2000. Additionally, metadata and faceted navigation now correctly return the raw input string instead of an incorrectly parsed date.

All the active outages are resolved now, and search results are no longer affected by artificial score inflation caused by invalid dates and are displaying correct dates.

By default, the summary field has double-encoded HTML characters, requiring template updates to use noesc.

The ability to control how the summary field is encoded is required so that admin users do not have to edit their templates.

A new results page option, ui.modern.enable_summary_encoding, was added to control the encoding of the summary field.

Introduced the ability for designated users to remove scheduled tasks from the task queue, controlled per client via a new feature flag and dedicated permission.

Previously, users could not cancel or delete queued tasks generated by the scheduling process. This created operational challenges for administrators who needed to abort long-running tasks to prevent them from impacting performance during business hours.

Administrators assigned the new permission (MANAGE_CLIENT_QUEUE_PRIORITY) can now delete scheduled items from the queue. The feature is controlled per client via a global configuration setting (task.task-picker-management-client-list) and automatically updates the relevant client roles upon logon.

Added visual status indicators for Push data sources directly on the list and details pages in the Admin UI, bringing parity with Web sources.

Previously, users could not easily view the operational status of Push data sources alongside other sources, making it harder to verify at a glance whether push commits were succeeding.

Push collections now display clear status badges in the UI: Success when the latest commit completes, Commit failed if an issue occurs, or N/A when status tracking is not applicable (such as for secondary/QP collections).