Release Summary 26.10 | May 14, 2026

ENG-31741

Ensemble AI:

Structured Scoring Applied to Freeform Ensembles

Freeform ensemble generation now incorporates the structured ensemble scoring model to improve ranking quality. Ensembles with stronger purchase likelihood based on behavioral co-occurrence data are ranked higher, while still factoring in vector-based relevance.

The final ensemble score combines the co-occurrence score with the vector database relevance score. When no co-occurrence data exists, ranking is driven primarily by relevance. A tuning parameter (default value of 1) controls the impact of the relevance score and can be adjusted in the future. When dev=true is enabled, both the total Ensemble Score and the Relevancy Score are returned in the user interface.

ENG-32118

Ensemble AI:

Distributed Scoring for Improved Scalability

Enhancements have been made to address memory limitations when processing large volumes of product scores. The scoring architecture has been redesigned to distribute processing more efficiently across resources.

Both the structured scoring approach and the LLM-based scoring workflow have been updated to support distributed execution. This reduces memory usage per instance, improves overall scalability, and allows additional machine resources to be leveraged as needed.

ENG-31588

Enterprise Dashboard:

Tabbed Recommendations Template Added to New Dynamic Experience UI

A new Tabbed Recommendations template has been added to the New dynamic experience user interface. This expands the available dynamic experience options and enables users to configure recommendation-driven experiences in a tabbed layout directly within the updated interface.

ENG-31587

Enterprise Dashboard:

Comparison Placement Template Added to New Dynamic Experience UI

A new Comparison Placement template has been added to the New Dynamic Experience user interface. This expands the available Dynamic Experience templates and enables users to configure comparison-based placements directly within the updated interface.

PLAT-4290

Platform:

Resilient Error Handling for Cassandra Write Failures in UPS Writer

Enhancements have been implemented to improve resilience in the UPS writer service when Cassandra write failures occur. Previously, such failures could halt event consumption and leave the service in a non-functional state.

The service now logs errors and exits gracefully on failure, allowing automatic restart and safe replay from the last committed offset. Existing functionality remains unaffected.

PLAT-4268

Platform:

Synchronous Ingest Count API Updated

The synchronous ingest count API has been enhanced to return total item counts directly from the store and to remove hourly ingest counters. A new streaming.ingest_count table maintains aggregated totals per site, snapshot, and item type.

The new GET /v1/{siteId}/{snapshotId}/ingest/counts API returns low-latency, aggregate item-level metrics. Counts are incrementally maintained, cleaned up upon snapshot deletion or cancellation, and exclude ingest-specific metrics. Empty payload items are marked as invalid, and all valid items are accurately reflected in both the table and API response. Existing streaming functionality continues to operate without regression.

ENG-32433

Find:

nGram Schema Changes Applied Across Language Service

To prevent future nGramSearchable discrepancies across sites, the ngram_* dynamic field schema has been updated to use stored=true and the change has been pushed into the language service codebase. This ensures the correct configuration is applied by default for all sites.

This proactive fix prevents partial search failures caused by field omission during atomic updates and ensures consistent search behavior across environments.

ENG-32351

Enterprise Dashboard:

Test Drive Enhancements and Fixes

Several issues in Recs and Content Test Drives have been resolved. Email channel images are no longer stretched in either Test Drive. The dedupWith parameter is now included only when two or more placements are selected. In Recs Test Drive, when Force Strategy is used and the response does not match the selected strategy, the Rules table is hidden and a clear message is displayed.

In Content Test Drive, content links now correctly navigate to the content catalog by extracting the content ID from the response.

ENG-31780

Ensemble AI:

Source String Now Populated for Click Tracking

An issue was identified where click events related to similar products were missing the sourceString value when a strategyId was present. This caused reporting gaps, as it was not possible to distinguish between view, click, or swap actions for similar products.

The click tracking logic has been updated to ensure that the sourceString is populated even when a strategyId is included. This ensures accurate tracking and reporting for Ensemble AI interactions, including support for parameters such as styleId, outfit, and srcstr in future calls.

ENG-32456

Social Proof:

Social Proof Coverage Analysis -  Production Memory Issue Resolved

An issue was identified where the Social Proof Coverage Analysis rollup was failing in Production due to memory constraints caused by high data volumes and cross joins.

The rollup logic has been optimized to handle large datasets more efficiently, and it now completes successfully in Production without memory-related failures.

ENG-32399

Recommend:

Category Affinity Calculation Fixes

Fixed inconsistencies in category affinity calculations when the Category weight was set to ‘0’ in the default User Affinity configuration. Category browse events now correctly factor in the Category weight, and attributes with a weight of ‘0’ are excluded from score calculations and removed from the affinityScores response.

For Configurable Strategies using Category Affinity as a seed, a fallback logic has been added. If the default configuration has a Category weight of ‘0’, category affinities are evaluated with a score of ‘1’ to ensure results are still generated.

ENG-32303

Engage:

Engage Content View Events Deduplication Fix

An issue was identified where Engage content requests were logging multiple Avro view events with the same viewGuid, siteId, and eventTime. These were treated as duplicates by the data pipeline and discarded by the log compactor, resulting in permanent data loss and incomplete reporting .

The logging logic has been corrected to ensure that multiple Engage placements within a single request are properly consolidated and no longer generate duplicate view events. This prevents unintended deduplication and preserves accurate reporting downstream.

ENG-32330

Social Proof:

Incorrect “-multiple messages” Logging Resolved

An issue was identified where Social Proof messages on cart pages were being logged with the suffix “-multiple messages” even when only a single message variation was configured. This suffix is intended only for scenarios where multiple messages are enabled and displayed in rotation.

The logging behavior has been corrected to ensure that “-multiple messages” is appended only when multiple message variations are explicitly enabled.

ENG-32474

Enterprise Dashboard:

View Results Tab: User ID Display and Affinity Configuration Fix

An issue was identified where the View Results tab displayed Product ID instead of User ID for strategies using user affinity. In some cases, the User ID field did not appear in preview, and selected user affinity configurations were reverting to “default” after saving.

This occurred when opening a strategy in edit mode and navigating directly to the Results View tab. The issue has been resolved, and User ID and selected affinity configurations now display and persist correctly.

ENG-32474

Enterprise Dashboard:

View Results Tab: User ID Display Fix

An issue was identified where the View Results tab displayed Product ID instead of User ID for strategies that require user-based input. This caused confusion when previewing strategies that rely on user context.

The issue has been resolved, and the View Results tab now correctly displays the User ID where applicable.

ENG-32496

Enterprise Dashboard:

Top Products -  Incorrect Attributable Sales Display Resolved

An issue was identified in the dashboard where the Top 10 Products view displayed incorrect values for attributable sales. Instead of showing total attributable sales on the x-axis, the chart was displaying the external product ID.

The issue has been corrected, and the Top Products dashboard now accurately displays attributable sales values for the selected time ranges.

ENG-32476

Helper Chatbot:

Prevent Empty Chat History Entries

An issue was identified where some historic chat sessions appeared with empty chat history. This led to confusion when users attempted to access past conversations.

The logic has been updated to ensure that only chats with valid history are returned. Empty or invalid chat sessions are no longer surfaced in the chat history list. This enhancement is working as expected.

ENG-32509

Ensemble AI:

Excluded Categories Displayed in Review Screen

An issue was identified where excluded subcategories were appearing in the category dropdown on the Ensemble review screen. These categories were not meant to be available for selection or ensemble generation.

The dropdown logic has been updated to include only allowed categories. Excluded categories are now properly filtered out and no longer appear in the review screen.

ENG-31255

Enterprise Dashboard:

Reports - Extra Day Displayed Beyond Selected End Date

An issue was identified in Sales and Sales by Channel reports where data for one additional day beyond the selected end date was being displayed. This occurred across different time buckets, including Monthly and Quarterly views, resulting in reports showing data for dates outside the user-selected range.

The date handling logic has been corrected to ensure reports strictly respect the selected end date, and no additional day is included in the results.