Release Summary 26.05 | Mar 05, 2026

ENG-31976

Find:

Configurable Limit on Solr Calls in Search Service

A new configuration option has been added to control the number of Solr calls made by the search service per request. This is particularly relevant for Query Understanding features that may trigger multiple Solr queries.

The default limit is set to two Solr calls per request, with the flexibility to adjust this through search configuration. This enhancement provides better control over search performance and resource usage.

ENG-31788

Enterprise Dashboard:

Automatic Site Configuration for InSegment Strategies with Custom Segments

When a Custom Segment is added to the whitelist for InSegment strategies on the Model Options page, the required site configuration is now enabled automatically. This ensures the model builds successfully without requiring users to manually update site settings.

If the configuration is already enabled, no changes are made, and the setting is never disabled automatically. This enhancement reduces setup errors and simplifies the process of working with Custom Segments in InSegment models.    

ENG-31971

Enterprise Dashboard:

Use richrelevance.com Domain for Email Test Drive Requests

Test Drive requests for the Email channel have been updated to use the richrelevance.com domain instead of the algorecs domain. This change ensures compatibility in environments where Cloudflare is not configured for the algorecs domain.

The update applies to both Recommendations and Content Test Drives when the Email channel is selected, ensuring email previews function reliably across supported setups.

PLAT-4239

Recommend:

Instant Enrichment Data Rollup to Catalog

Enrichment data is now rolled up to catalog tables immediately after file processing, eliminating the delay previously caused by waiting for a full catalog feed. Earlier, enrichment data was stored only in staging tables and did not reflect in the product attribute values used for search and recommendations until a full feed was run.

With the new transactional rollup process, enrichment updates are applied directly to the catalog as part of the same flow. This ensures faster data availability and more timely impact on search and recommendation experiences.

PLAT-4249

Steaming/Recommend:

Improved Reliability for Streaming Itemstore Checkpoint Processing

The streaming itemstore consumer has been enhanced to handle errors more reliably during checkpoint event processing. Previously, the consumer could retry on failure and silently stop without completing the required data rollups.

With the updated handling, checkpoint events now trigger data rollups that run and persist successfully without silent failures. This ensures greater reliability and consistency in streaming data processing.

ENG-31507

Ensemble AI:

Visual AI Vector Generation Moved to Vector DB Job

Visual AI vector generation has been moved from the Ensemble AI job to the centralized vector DB job. The vector DB process now generates and stores visual AI model vectors, with a corresponding configuration available in the LLM settings to include these vectors as needed.

The Ensemble AI model build has been updated to use the pre-generated vectors instead of creating them during its own execution. This change streamlines processing, reduces duplication of work, and improves efficiency in building visual AI-powered ensembles.

ENG-31099

Science:

Automated Scheduling for LLM and Visual AI Ensemble Jobs

The process of generating ensembles using LLM or LLM combined with Visual AI has been automated and can now be scheduled as a job. The required runtime environment and dependencies have been identified to support consistent execution.

The job consumes a defined input file specifying the sites and style definitions to process, and produces a JSON output compatible with the existing Ensemble AI workflow. This ensures generated ensembles are seamlessly picked up and applied, while enabling scalable and repeatable execution.

ENG-31379

Science:

Placement-Specific Optimization Metric Configuration

The portal API now supports configuring a different optimization metric for a specific placement within a page type. While a default optimization metric can still be set at the page type level, users can now define exceptions for selected placements to use a different metric.

For example, an Item Page may use CTR as the default metric, while a specific mobile app placement on the same page type can use Attributable Conversion. Multiple placement-level exceptions can be configured, and the appropriate set of strategies is returned based on the page type and placement combination. This enhancement provides greater flexibility to optimize performance across channels and placements.

ENG-30891

Social Proof:

Social Proof TS Visualizations for Message-Level Reporting

Social Proof reporting now includes TS visualizations at the message level, allowing merchandisers to evaluate performance by message type, interval, and threshold combination. This provides clearer insight into how specific message configurations drive engagement and conversion over time.

The report includes key metrics such as visits, views, view-based add to cart visits, purchase visits, sales, orders, units, conversion rate, ATC rate, and revenue per visit. Performance can be analyzed across dimensions including date, experience or variation, category, and region, with both aggregated and de-aggregated views.

ENG-31646

Social Proof:

Expanded Page Type Support for Social Proof

Social Proof now supports all page types, removing previous limitations that restricted messaging to specific pages. This enables teams to activate Social Proof on additional experiences such as home pages, chat pages, and other custom page types without additional workarounds.

Non-item and non-quick view pages can be handled appropriately as list-style views where required, ensuring consistent message behavior across different layouts. This enhancement provides greater flexibility in deploying Social Proof across the entire site experience.

PLAT-4199

Platform:

SFTP Support Added to BuildFTP Server

BuildFTP has been enhanced to support automated file retrieval from client-hosted SFTP servers for configured sites. Uploaded ZIP files containing UPS batch, user linking, or Hive data, along with metadata, are now processed through the existing pipeline at defined intervals.

The SFTP watcher runs only for configured sites and has no impact when disabled. Files are archived based on processing outcome, and existing UPS and Hive workflows continue to function as before, ensuring seamless integration with the new capability..

PLAT-4255

Platform:

Country-Specific Product Compatibility Enablement

Phone-to-accessory compatibility matching has been enabled across multiple Nordic countries. The pipeline uses Azure OpenAI to extract phone model names from product descriptions and automatically match phones with compatible accessories such as cases and screen protectors.

This enhancement improves product discovery and relevance by ensuring shoppers see accessories that are accurately compatible with their selected devices across supported regions.

ENG-31830

Find:

Refined Elevation and Boosting Logic for Search

Boost & Bury product boosting now applies only to products already present in the search result set, preventing unintended sitewide elevation behavior.

To ensure a product always appears in results, clients should use the existing Search Boost Rule or Visual Merchandising features. Boost & Bury will now impact ranking only within matched results, while elevation remains handled separately through dedicated mechanisms.

ENG-31708

Recommend:

Reduced Excessive Logging in rrserver

We have fixed an issue where rrserver was generating a high volume of error and warning log messages related to missing inventory attribute keys and request parameter handling. These messages were being logged repeatedly, creating unnecessary noise in server logs.

The logging behavior has been corrected to prevent these messages from appearing at inappropriate log levels. This reduces log clutter, improves system observability, and ensures that only relevant issues are surfaced for monitoring and troubleshooting.

ENG-31931

Recommend:

Resolved Slot Reduction Issue in Email Recommendations with dedupeWith

We have fixed an issue where email placements configured for eight slots were returning only four recommendations when the dedupeWith parameter was included. Removing one placement from the dedupe configuration temporarily restored all slots, indicating an issue in deduplication handling across placements.

The deduplication logic has been corrected to ensure that the configured number of slots is honored while still preventing duplicate products across placements. Email placements now return the expected number of recommendations, even when dedupeWith is applied.

ENG-31993

Recommend:

Corrected HTTPS URLs in Category Recommendation Responses

We have fixed an issue where category recommendation links were returned with http instead of https in the API response, even when secure requests were expected.

The API now correctly returns HTTPS URLs for category recommendations, preventing client-side errors and ensuring secure link handling in production environments.

ENG-31919

Recommend:

Restored UI Control for Feed Processing in Feedherder

We have fixed an issue where changes to the “watching” status in the Feedherder UI were not being saved or applied correctly, requiring manual database updates.

The UI now correctly supports enabling or disabling the watching status for feed profiles, allowing teams to pause and resume feed processing directly from the interface as intended.

ENG-31887

Recommend:

Resolved Incomplete Results for Browse History Configurable Strategy

We have fixed an issue where a configurable strategy based on user browse history returned only one product, even when multiple valid items existed in the user’s history.

The strategy now correctly retrieves and displays all eligible products from the user’s browse history, ensuring accurate and complete recommendation results during preview and runtime.

ENG-31948

Find:

Resolved Random Behavior in Find Filter Flag

We have fixed an issue where the Find filter flag was defined as a shared class-level variable, leading to inconsistent behavior in multi-threaded scenarios.

The flag has been moved to the method scope to ensure it is handled independently per request. This prevents cross-request interference and ensures consistent filter behavior during search processing.

ENG-31913

Engage:

Resolved Delay in Scheduled Banner Publishing

We have fixed an issue that caused a delay in publishing scheduled banner content. In this case, the banner did not go live at the configured time and appeared in production later than expected.

The underlying processing has been corrected to ensure scheduled content updates and expirations are executed in alignment with the configured site timezone. This ensures banners go live and expire at the intended time, providing more reliable scheduling and a smoother content publishing experience.

ENG-32021

Resolved Intermittent Missing Category Names in Category Recommendations

We have fixed an issue where category recommendations intermittently returned blank or missing category names in both production and integration environments.

Category data is now consistently populated in API responses, ensuring that category names display correctly across placements. This prevents incomplete category tiles from appearing and ensures a reliable experience on high-visibility pages.