Release Summary 26.12 | June 11, 2026

ENG-32395

Find:

Checkpoint Scalability for High Cardinality QU Attributes

The checkpoint process has been enhanced to handle high-cardinality facet values in Query Understanding (QU), including scenarios where facet values reach into the millions. This improvement prevents stalling in datacenters and ensures stable processing of large volumes of unique identifiers.

The solution supports query tags for product IDs, entity IDs, and SapMaterialNumbers. During checkpoint execution, new query tags are correctly populated, legacy catalog-type query tags are cleared, and system stability is maintained throughout processing under high load conditions.

ENG-32614

Recommend:

Known Bots List Updated for Meta Crawlers

The known bots list has been updated to include meta-externalads/1.1 and other Meta web crawlers. These bots were generating recommendation clicks, which impacted the accuracy of client reporting.

By adding these crawlers to the known bots list, bot traffic is now properly filtered, ensuring more accurate measurement of recommendation performance.

ENG-32265

Recommend:

CategoryCP2 Model: Configurable Strategies vs. Legacy

Differences between the CategoryCP2 model in Configurable Strategies and the legacy Model Browser have been analyzed and clarified. Configurable Strategies apply additional runtime filters, context rules, and preview constraints, which can result in differences in returned products, ordering, and overall count.

Products may appear in one experience but not the other due to filtering, scoring adjustments, or evaluation timing. Variations in ranking and smaller result sets in Configurable Strategies are expected because of stricter runtime logic compared to the broader legacy model output.

ENG-32224

Recommend:

Cache Size Monitoring and Alerting

Metrics have been introduced to track cache size growth for proactive monitoring and early detection of performance risks. A cache size metric is now published, segmented by site and cache type, to provide visibility into overall cache usage.

Thresholds have been defined to identify abnormal cache growth, and alerts have been configured to trigger when limits are exceeded. This enhancement helps prevent performance issues caused by excessive cache expansion.

ENG-31856

Domain Alignment for Dynamic Experience and Guided Selling Templates

Guided Selling and Dynamic Experience templates have been updated to use the configured baseUrl instead of hardcoded domains, preventing Cross-Origin Resource Sharing (CORS) issues and ensuring consistent service calls. If baseUrl is not defined, the system now defaults to the appropriate domain based on the client.js version in use.

For client.js version 2.1, the fallback domain is ‘algorecs’. For versions 2.0 and 1.0, the fallback remains ‘richrelevance’. This logic is handled within each version of client.js and applies to all default templates, including Dynamic Experiences, Social Proof, Ensemble AI, Chat, and Guided Selling.

ENG-32471

Configurable Strategies:

Bulk Strategy Message Updates

The Configurable Strategies page has been enhanced to improve performance and streamline strategy message handling. The strategies API now retrieves message data only for the current strategy, reducing unnecessary data loading and memory usage. No message call is made for new strategies that do not yet have strategy messages.

Saving strategy messages now uses a single bulk update call across all page types instead of separate save and publish calls. Additionally, when setting a strategy message on a new strategy, the message is now immediately reflected in the user interface after saving in the side panel, providing clearer feedback to the user.

ENG-32414

Shopping Assistant Chatbot:

Color Support in Vector Database

The vector database schema has been updated to include a color field populated from the image tagging job. This enhancement ensures that color information is stored alongside product embeddings.

The vector job now inserts tagged color values into the database, enabling the chat experience to filter by color and return all relevant variations of the requested color.

PLAT-4356

Platform:

Streaming Sidekick: Scoped Action via item_store_index

Streaming Sidekick has been enhanced to process scoped actions using the item_store_index table. This update improves consistency and supports more efficient handling of scoped action variants.

All scoped action variants have been validated through regression testing for counts and data accuracy. Performance comparison was also conducted for the replaceAll variant between the older and updated versions to ensure improved or stable performance.

ENG-32584

Social Proof:

Coverage Analysis Report Improvements

Enhancements have been made to the Social Proof Coverage Analysis report in the portal to address inconsistencies in event and user counts across different time thresholds.

Issues where 1-day event count, user count, or views appeared higher than 2-day or 3-day values, as well as discrepancies where event count exceeded user count across thresholds, have been resolved. The report now reflects accurate and consistent metrics across all selected thresholds.

ENG-31959

Enterprise Dashboard:

Search Boost and Bury Enhancements

Search Boost and Bury rules have been enhanced to improve stability and prevent system limits from being exceeded. Product ID boosting now supports only the Conditional type and requires at least one keyword. Sitewide Product ID boosting has been disabled.

A configurable condition limit (default 100) has been introduced per rule. The user interface displays a “Too many conditions (max 100)” message when the limit is reached and prevents additional selections. Backend enforcement ensures that only rules within configured limits are processed, with excess conditions ignored and logged.

ENG-32554

Find:

Find Query Understanding: Stemmer Override Enhancement

Support has been added for a StemmerOverrideFilter in Find Query Understanding to handle irregular plural stemming cases not covered by the existing EnglishMinimalStemFilter. This ensures terms such as “dresses” correctly match “dress.”

A dictionary file (stemdict.txt) is used to define specific override mappings for edge cases. Configuration updates have been applied in both the language service and Solr config sets to enable this enhancement.

ENG-32725

Enterprise Dashboard:

Segments User Interface: “User Purchased” Constraint Not Displayed

The “User purchased” constraint was not appearing under the Product Attribute condition in the Segments user interface, even though it was supported in the backend. As a result, users were unable to create segments based on previously purchased product attributes.

This issue was caused by an incorrect constraint value in the configuration. The error has been corrected, and the “User purchased” option now appears correctly and functions as expected.

ENG-32661

Shopping Assistant Chatbot:

Color Filter Backend Field Update

The color filtering option on the LLM Configuration page was not updating the required backend field used for caching and triggering vector database updates. As a result, color filtering was not functioning as expected.

The issue has been resolved by ensuring the correct field, color_filtering_enabled, is now passed with a value of true or false. No user interface changes were required.

ENG-32428

Ensemble AI:

Incorrect Products Shown for Selected Category

An issue was identified where incorrect seed products were displayed when reviewing ensembles for selected categories. Some products shown did not belong to the chosen category.

The category filtering logic has been corrected, and seed products now accurately reflect the selected category.

ENG-32564

Recs Test Drive:

Force Strategy and Email Placement Fixes

Two issues in Recs Test Drive have been resolved. When using the Force Strategy option, the seed product was previously passed twice in the recsForPlacementsContext request, which caused conflicts in certain merchandising scenarios. The request has been updated to pass the seed product only once in the request body, ensuring consistent behavior and compatibility with the import request feature.

Additionally, for email_page placements using recsForPlacements, category-based seeds now correctly use the chi parameter instead of categoryId, aligning the request structure with expected email placement behavior.

ENG-32368

Configurable Strategies:

RichRule Actions Not Applied on End Date

An issue was identified where RichRule actions were not applied on the exact configured end date for Configurable Strategies. While legacy strategies load restriction rules from the rich_rules2 cache, Configurable Strategies use the strategy-exec-rule-action model. In this flow, rule actions became empty when the end date was reached but the end-of-day time (23:59:59) had not yet passed.

The date handling logic has been corrected to ensure restriction rules remain active through the full configured end date. Rule actions now behave consistently with expected end-of-day validity.

ENG-32627

EmbedCatalog: Preserve Zero-Padded External IDs

Zero-padded external_id values were losing leading zeros during ingestion due to automatic numeric type inference. The process has been updated to load external_id explicitly as a string, preserving the original format.

Additionally, when the -f flag is used, existing embedding_hashes for the site are cleared before inserting new ones, aligning with the full rebuild behavior of the collection.

ENG-31735

Recommend:

In-Segment Strategy Not Firing

An issue was reported where a segment-based strategy was not triggering for a user who was confirmed to be part of the configured segment. Despite the placement and strategy being published, the rule set to “Use only These,” and no restriction rules filtering results in Test Drive, the expected strategy did not return recommendations.

The issue has been identified and resolved. The in-segment strategy now fires correctly for users who meet the configured segment criteria.

ENG-32662

Engage:

Unable to Configure User Segments

An issue was identified where configuring User Segments in Engage prompted an unexpected login request. The problem occurred due to the /engage/autocomplete request not being populated with the correct user information, resulting in a 401 response.

The request flow has been updated to pass through the appropriate authentication filter and set the correct user context. User Segments can now be configured without interruption.