** Improvements**

• General Table Reasoning — Improved reasoning capabilities when processing tables in answers

** Bug Fixes**

• Fix Not-Applicable Choice Marking in Tables — Fixed post-processing to correctly handle both NA:choice_labe and choice_label formats when marking choices as not-applicable in table responses

New Features

• Knowledge Hub, New module for managing knowledge records with full CRUD operations, API endpoints, and search with pagination
• Not Applicable Detection, Screening processor now automatically identifies not-applicable questions
• Screening Guidance, Added a context field to screenings, allowing users to attach guidance and additional rules that flow into metric extraction
• MCQ Deep Analysis Mode, Refactored multi-choice question pipeline with a new deep analysis mode and ElasticSearch similarity integration

Improvements

• More Consistent Inline References for Tables, Improved inline reference post-processing for table-based answers
• General Table Reasoning, Improved reasoning capabilities when processing tables in answers
• Resource Hub Crawling, Added bonus scoring for resource hub URLs and new keywords to web crawler
• include_sub_responses Parameter, New parameter on screening responses for retrieving sub-responses
• company_id on Screenings, Screenings now carry a company identifier
• Optional file_ids Filter, get_files now supports filtering by specific file IDs

Bug Fixes

• Fix File Status Filter, Corrected recursive folder contents query when excluding by file status
• Fix Batch Context Propagation, Batch context is now properly passed when triggering executions from the file-ready path
• Graceful Error Handling on Regeneration, More graceful handling of regeneration failures

Not-Applicable Detection

• Questions, table columns, and individual cells can now be automatically identified as "not applicable" when they don't pertain to the company or document being analyzed
• Not-applicable assessments include reasoning to explain why certain fields don't apply

Table Processing Improvements

• Enhanced evidence assessment now includes referenced content for more accurate table filling
• Improved handling of fixed-row tables with better choice pre-creation
• Tables now continue processing cells even when table-level evidence is unavailable

Screening & Company Management

• Screenings are now linked to companies, enabling better organization and tracking of screening data across your company portfolio

Table Processing Enhancements

• "No evidence" tags now correctly propagate to all row-level answers in fixed tables, giving you more consistent assessment states across table responses.
• Table responses that don't align with expected structure now automatically retry, reducing incomplete or malformed results.
• You can now provide template guidance for table questions via TableQuestion.template_guidance to improve response accuracy.

Screening Improvements

• Parent items are now automatically included when executing a screening, ensuring complete hierarchical context without manual configuration.

Data Model Cleanup

• Company cohorts functionality has been removed—companies are now associated directly with folders via company_id, simplifying the data model.

Webhooks Integration You can now receive real-time notifications when events occur in your Briink workspace. The new webhook system allows you to:

• Subscribe to events and receive HTTP callbacks to your endpoints
• Benefit from automatic retry logic for failed deliveries
• Handle transient errors gracefully with built-in error handling

This enables you to build responsive integrations that react immediately to document processing completions, assessment updates, and other platform events - instead of having to poll for status changes.

Find the full documentation here! https://briink.readme.io/docs/webhooks

Enhanced AI Assessment Tracking

• Responses and response choices now include an ai_assessment_state field that indicates the assessment status
• The Briink AI automatically detects when there is insufficient evidence to answer a question, marking responses with a NO_EVIDENCE state

Improved Retrieval Quality

• Document retrieval now returns 200x relevant context by default, improving answer quality for complex questions

Bug Fixes

• Fixed an issue with questionnaire question reordering that could cause validation errors when updating questions

The Idempotency-Key header for the batch screenings endpoint is now properly documented in the API specification. You can use this optional header to prevent duplicate batch screening creation when retrying requests, ensuring safe retry behavior in your integration.

Idempotency for Batch Screenings You can now include an idempotency key when creating batch screenings to safely retry requests without creating duplicates. The API will return the original response for repeated requests with the same key.

Verification Status Management Screening responses now support verification status tracking with UNVERIFIED, VERIFIED, PENDING_REVIEW and REJECTED states.

Improved File Upload and Downloads Separated PDF OCR processing into its own container function, enabling higher concurrency, lower memory usage, effecting large files.

Signed URLs for file downloads now include proper content disposition and MIME type headers, ensuring files open correctly in browsers and download with their original filenames.

Multilingual Web Crawling

• The web crawler now supports multiple languages, allowing it to navigate non-English websites and locate relevant pages like sustainability reports and investor relations sections.
• A penalty system prevents the crawler from drifting to external domains when following links from a company's website.
• Download links are now detected and pre-fetched during crawling to confirm they serve actual documents before adding them to the collection.
• Duplicate URL detection prevents processing the same page or document more than once.

Document Download & Ingestion

• Dynamic timeout strategy adjusts based on file size, preventing timeouts on larger documents.
• Retry strategy for PDF conversion — if the initial conversion fails, the system fetches the page content directly and retries.
• File extensions are now preserved instead of defaulting to .pdf, which previously caused Excel and other non-PDF files to fail processing.
• Added support for older Excel files (.xls format).

Screening Workflows

• Mentioning a user in a comment now automatically subscribes them to that screening response for future notifications.

• Comments, mentions, subscriptions, assignees, and due dates are now preserved when regenerating a response or cloning a screening.

• Locked responses now properly block edits and regeneration with a clear error message.

• Added analytics tracking for comment actions, subscriptions, and mentions to provide better insight into team collaboration.

Screening Subscriptions

• You can now subscribe to screening execution responses to receive notifications about changes.

Due Date Tracking

• Screening execution responses now support a target date field. You can set due dates when creating or updating responses, enabling better deadline management for your screening workflows.

Threaded Comment Replies and Mentions

• Comments on screening execution responses now support nested replies. You can create threaded discussions and retrieve comments with their full reply hierarchy.
• You can now mention workspace members in comments on screening execution responses. Mentioned users will be notified, making it easier to loop in teammates and collaborate directly within your screening workflows.