August 2026 Release Notes
đ The Optimizer: Prove It Before You Apply It
Improving variables that are complex can be tricky. The Optimizer is now updated to make iterating on your variables more methodical.
The Optimizer analyzes the disagreement between the training labels you created manually or uploaded in Validation and Brimâs generated labels, applies Brimâs Best Practices for variable instructions, and proposes targeted changes.
Open Project Setup, click Optimize on a variable, optionally guide how youâd like it improved, and click Start Optimization. From there you can:
- Start with a hypothesis. Tell the Optimizer what you want to improve, or let it work from the disagreements it finds in your validation results.
- Easily see what changed. Current and proposed instructions sit side by side, so youâre never guessing at what the Optimizer rewrote.
- Test before you commit. If you have Training Data, you can generate on a sample set of patients to compare agreement before and after the proposed change.
- Apply or discard. If your agreement improves, apply and save the new instructions. If not, you can discard and try a different angle.
- Keep the learnings. Every optimizer run is saved in Optimization History for that variable, so you and your team can see what was tried, what improved agreement, and what didnât.
Learn more in our documentation.
đŠ Flagging & Commenting on Results
Reviewing a label is often more than a simple accept, edit, or remove. Sometimes a value needs a second opinion, two reviewers disagree, or there's context worth leaving for the rest of the team. Flags and comments let you capture that directly on a result, so triage and discussion happen alongside the data instead of in a separate thread. Both are visible to and editable by every project member.
You can find Flags and Comments where you find the Accept, Edit, and Remove buttons. To flag a value, click the flag icon to apply a color. Add a comment by clicking the comment icon, entering 500 characters or less, and clicking âSaveâ. Each label can have one flag and one comment at a time.
Both flow into the Detailed Export, persist through review status changes, and follow your choice on what to do with review status when you regenerate: cleared if you clear all values, kept if you keep human-generated ones.
Learn more in our documentation.
𧏠New Changes and Customer Requests
Several improvements based on your requests:
- Default Value filter moved into the filter popup on label generation and label review, instead of a checkbox on the label page.
- Conditional Generation indicators showing when a variable uses conditional generation, indicators when a label was âExcluded by Conditionsâ, and an additional filter option letting you show values that were skipped due to conditional generation. Learn more in our documentation.
- Filter a Label Generation by the value of a structured, one per note, or one per patient variable.
Keep them coming! Reach out to support@brimanalytics.com.
Also in this Release
- "Value" column in label tables retains the Generated Value, even on edit
- All Review Statuses now have a "Revert" option
- Filter Menu has subheaders for easily navigation, and filters persist between views.
- Manually Labelled Training Examples are retained unless specifically deleted by the user
- Cleaner "Validation is out of date" experience across screens
- Lists outputted by variables are shown as pills for clarity, and improved our list comparator in validation
- Variables can now see original Instructions while Aggregating
- Filter Label Generation by the value of a Structured Data Variable
- "minimal" effort level available for reasoning models
⨠How to Get the Latest Brim Version
This version is called:
2026.08.31
You can check your current version in Settings â About.