Release Summary 26.04 | Feb 19, 2026

ENG-31760

Find:

Region-Aware Query Tag Handling in Search

Query tag handling has been improved to prevent zero-result scenarios caused by region mismatches. Query tags are now stored with associated regions as a list, ensuring region-specific accuracy during search processing.

The Search Service now selects query tags based on the requested region at query time. This enhancement improves search relevance and consistency across regions, especially in multi-region setups.

ENG-31710

Find:

Configurable Tagged Filters for Query Understanding

New configuration options have been introduced to provide greater control over Query Understanding filters. The enableTaggedFilters flag allows teams to control tagging and exclusion behavior for filters. By default, this flag is set to false. When enabled, tagged filter clauses are applied as part of the query, aligning with existing tagged filtering behavior. When disabled, filters are applied without tagging, ensuring that only filtered values are shown in the facet. Query Understanding filters also apply only when a facet contains a single tagged value.

In addition, a new configuration setting, queryUnderstanding.tagging.enableFilters=true/false, has been introduced to completely enable or disable Query Understanding filters in Find requests. Setting queryUnderstanding.tagging.enableFilters=false disables Query Understanding filters in subsequent calls. This provides clients with more flexibility to manage filtering behavior based on their specific requirements.

ENG-31154

Find:

Migration of Publisher Jobs to Airflow

The WOC Aggregator, Global Ranking Publisher, and Related Searches publisher jobs have been migrated to Airflow for centralized orchestration and management. Required configuration updates were applied in Consul and Airflow variables, including container mappings, memory allocation, scheduling, and job chaining to ensure seamless execution.

With this migration, these jobs are now managed through standardized Airflow workflows, improving operational consistency, visibility, and maintainability across environments.

ENG-27507

Find:

Query Tag Index Updated to Respect Sitewide Filters

Query tag generation has been enhanced to respect sitewide filters defined in the JSON configuration. When generating query tags, the system now applies these fixed filters to ensure that the resulting collections accurately reflect the intended catalog scope.

In addition, the query tag collection is automatically recreated when catalog updates occur. This ensures the index stays aligned with the latest catalog state and maintains accurate search behavior over time.

ENG-31764

Enterprise Dashboard:

Improvements to the New Configurable Strategies Page

Several usability updates have been made to the new Configurable Strategies page. Tooltips now work on both click and hover, model type labels have been clarified, and the “Filters” panel has been renamed to “Advanced Settings” for better clarity.

In addition, the “No Filters Applied” option is now selectable for all models, including User’s Activity models. These updates improve consistency and make strategy configuration more intuitive.

ENG-31462

Recommend:

Automatic Resolution of Previously Unknown Products in Rules

Rules that were created with products not yet present in the catalog are now automatically updated when those products become available. When Rec Restriction, Boosting, or Manual Recommendation rules are retrieved through the Dashboard API, any products that previously had null names are resolved against the current catalog.

If the product now exists, its name is populated in the response instead of being shown as unknown. This ensures rules display accurate product information and behave as expected at runtime, without requiring manual updates from merchandisers.

ENG-31836

Chatbot:

Configurable Number of Chatbot Response Carousels

The chatbot UI now supports configuring the number of product carousels returned in a response. This allows teams, especially for mobile app experiences, to limit responses to a single carousel to keep conversations focused and reduce the likelihood of duplicate products appearing across multiple carousels.

The system dynamically adjusts based on the number of examples defined in the LLM configuration. If only one example is configured, the chatbot generates a single carousel instead of multiple responses. This provides greater control over the conversational experience and improves clarity for shoppers.

ENG-31395

Enterprise Dashboard:

Placement-Level Optimization Metric Configuration in Experience Optimizer

Experience Optimizer now allows optimization metrics to be configured at the placement level within a page type. In addition to setting a metric for an entire page type, users can add placement-specific configurations and select a different optimization metric where needed. For example, a mobile app placement on the item page can use attributable conversion, while other item page placements continue to use click-through rate.

Users can add and manage placement-level configurations directly under each page type, with validation to ensure the placement metric differs from the page type metric. This provides greater flexibility to fine-tune optimization strategies across channels and placements.

ENG-31925

Social Proof:

Experience ID Included in Social Proof API Requests

The Social Proof UI has been updated to append the Experience ID in spMessages API requests. This ensures that the correct experience context is passed as part of message retrieval and tracking.

PLAT-4231

Platform:

Datacenter-Based Configuration for Kafka Repeater

Datacenter-Based Configuration for Kafka Repeater

Kafka repeater configuration has been updated to use a datacenter-based approach instead of static region matching. Previously, regional filtering required manual configuration updates when datacenter-to-region mappings changed.

The repeater now resolves its regional scope dynamically using centralized configuration, allowing updates to be applied in real time without service restarts. This improves operational efficiency and reduces configuration errors.

PLAT-4230

Platform:

Region and Datacenter Deletion Support in RapidConfig

RapidConfig now supports deleting regions and datacenters through both the UI and API. Administrators can review impact details before deletion, including affected sites or assignments, and must confirm the action to proceed.

All deletions include automatic cleanup of related references and are executed within database transactions to ensure data consistency. This provides safer and more controlled management of regions and datacenters.

ENG-31859

Enterprise Dashboard:

Lowercase Normalization for Catalog Embedding Image Jobs

Catalog embedding processing has been updated to normalize gender attribute values to lowercase during image job execution. For example, values such as “Male” are now converted to “male” before being stored or used in vector searches.

This ensures consistent matching in the vector database and prevents mismatches caused by case differences, improving the accuracy of image-based search and recommendation results.

PLAT-4200

Platform:

Airflow Rollup for Streaming Enrichment Data

An automated rollup process has been implemented to sync enrichment data from Postgres tables to the Enrichment API for streaming clients. This ensures that enrichment data created through the Portal or client feeds is consistently available for real-time consumption.

Each rollup execution replaces the active dataset with the latest data from Postgres, handling dataset management, calculations, subscriptions, and ingestion as part of the process. This ensures streaming clients always receive up-to-date enrichment data without disruption.

ENG-30949

Data Engineering:

Sanitization of Special Characters in Social Proof API Features

We have fixed an issue where special characters such as spaces, colons, equals signs, or pipes within string features caused incorrect parsing in the Social Proof Optimization model and impacted downstream logging.

String features in both context and action inputs are now sanitized before calling the Inference API and logging. Special characters within feature names or values are replaced with a standard alternative to ensure consistent parsing and prevent model corruption.

ENG-31738

Enterprise Dashboard:

Placement Profile Chart Display for Single Currency Sites

We have fixed an issue where the chart on the Placement Profile page was not displaying for single currency sites, even though data was available.

The chart now loads correctly when the View Chart option is selected, ensuring accurate visualization of placement performance.

ENG-31897

Enterprise Dashboard:

Error Handling for Relative Click and Product URLs

We have fixed an issue where relative product URLs caused the UI to fail while constructing click URLs, resulting in an error message even when the chatbot returned a valid response.

Defensive error handling has been added around the URL construction logic to safely handle relative paths. This prevents UI failures and ensures chatbot responses are displayed correctly.

ENG-31909

Science:

Improved Handling of Empty Results in Chatbot Searches

We have fixed an issue where the chatbot indicated that products were found but did not display any results to the user. The final response logic has been corrected to properly evaluate intermediate search results and clearly indicate when no products are found.

Hard filtering has also been refined to prevent overly strict filters from excluding valid results. The search query now applies only approved hard-filter values, reducing false negatives and improving the overall search experience.