Pull Request Explorer

Summary

Add new dimension type documentation

Adds documentation for the new dimension type, improving developer understanding and onboarding.

Health Assessment

• • Fast review and merge with minimal changes, indicating a straightforward documentation update.

Summary

Add custom dimensions for experiment slicing

Enables users to slice experiment results by exposure cutoff and combined dimensions, improving analytical granularity and decision‑making.

Health Assessment

• • The PR added extensive new features with over 2,500 lines of change, was reviewed quickly but required multiple iterations before merging, indicating moderate review friction and high implementation risk.

Summary

Contextual Bandit Attributes Sync

Aligns contextual bandit queries with attribute selections, preventing data drift and ensuring accurate analytics for users.

Health Assessment

• • The PR required multiple review cycles and several commits to address feedback, indicating moderate complexity. The 142‑hour cycle time and large code changes suggest a higher risk of integration issues.

Summary

Add custom values option to enum and multiselect fields

Enables users to enter values beyond predefined lists for enum and multiselect custom fields, improving flexibility in data entry and API usage.

Health Assessment

• • The PR was merged in under 12 hours with a single quick review, indicating efficient collaboration. AI-assisted comments from Cursor helped streamline the review, though several commits after the first review suggest moderate rework before final approval.

Summary

Update docs to recommend Azure DocumentDB instead of Cosmos

Docs updated to recommend Azure DocumentDB for better Mongo compatibility, improving self-hosting guidance.

Health Assessment

• • Docs change merged within minutes with no review friction, indicating a straightforward update.

Summary

Improve SQL Browser UX with Hover Actions

Enhances user experience in the SQL browser by replacing hidden double-click with visible hover actions for copying and inserting table/column paths, reducing confusion and speeding query construction.

Health Assessment

• • The PR had a moderate review friction with multiple rounds and a large code change, but the quick first review and short cycle time suggest efficient collaboration.

Summary

Add CDN Usage page tooltips and CSV export

Enables customers to view precise CDN usage metrics and export data directly, reducing ad‑hoc requests and improving transparency.

Health Assessment

• • Fast review and minimal rework; single file change with moderate line additions; quick cycle time indicates low risk and high developer efficiency.

Summary

Add ramp warning and new value action

Adds a safety warning for ramp‑ups and a new action to duplicate rules, preventing unintended traffic drops during feature rollouts.

Health Assessment

• • Fast review cycle with AI‑assisted comments, minimal rework, quick merge indicates low complexity and high confidence.

Summary

Fix flaky baseline-guard race cases

Resolved intermittent test failures caused by race conditions in baseline guard logic, ensuring consistent 409 conflict responses.

Health Assessment

• • Quick resolution of flaky tests with minimal changes, indicating efficient debugging and low impact on production.

Summary

Allow sticky bucketing to be opt in instead of opt out

Users can now opt in to sticky bucketing for new experiments, giving them control over default behavior and improving visibility of the setting.

Health Assessment

• • Rapid iteration with AI-assisted review enabled a 28‑hour cycle time, with a single review round and 10 commits to address feedback.

Summary

Add support for exclude in-progress conversions

Enables incremental pipeline mode to honor the exclude in-progress conversions setting, preventing costly full-scan queries and improving performance for large datasets.

Health Assessment

• • The PR had a long cycle time (≈12 days) and a large scope (≈1,000 lines across 17 files). Only a single review was performed, suggesting a bottleneck in the review process. The extensive changes to data pipeline logic increase risk of regressions.

Summary

Fix percent input rendering in comma-decimal locales

Ensures percent fields display correctly for users in locales using comma decimals, preventing blank inputs and improving data entry accuracy.

Health Assessment

• • The PR had a single review after almost a day, with no rework needed, indicating a clear issue and quick resolution.

Summary

Health Assessment

Summary

Add option to disable SDK feature usage events

Adds a configuration flag to the GrowthBook SDK tracking plugin that lets users suppress feature evaluation events, reducing noise while still tracking experiment views and custom events.

Health Assessment

• • Quick turnaround with minimal rework; AI assistance likely accelerated development.

Summary

Readable error and one-shot reload for auth proxy

Improves user experience for customers behind Google IAP by providing clear error messages and preventing reload loops when the proxy session expires.

Health Assessment

• • Fast turnaround with immediate merge after review, minimal rework.

Summary

Health Assessment

Summary

Updated AI models for August 2026

This PR refreshes the platform’s AI model catalog, adding the latest Anthropic, OpenAI, Mistral, and Google models while removing deprecated ones, ensuring users can select up-to-date models for text and image generation.

Health Assessment

• • The PR was merged within 10.6 hours, with rapid, small commits and no review delays, indicating a smooth and efficient process.

Summary

Fix default environment selection for linked features

This change prevents accidental enabling of disabled environments when linking feature flags to experiments by defaulting to enabled environments. It also adds warnings and clearer UI to help users understand which environments will be affected.

Health Assessment

• • The PR had a long cycle time of over 11 days and required multiple review rounds, indicating significant complexity. AI-assisted review helped surface issues, but the overall process remained slow and involved substantial rework.

Summary

Sync SDK versions with upstream releases

Updates SDK version metadata to reflect the latest releases, keeping documentation and tooling current for all supported languages.

Health Assessment

• • The PR had a long cycle time with a delayed first review, but the changes were minimal and low risk, involving only configuration updates.

Summary

Pre-download and cache the mongod binary in CI

Pre-downloads and caches the MongoDB binary in CI to prevent race conditions among test workers, improving test reliability and speed.

Health Assessment

• • Quick CI improvement with minimal changes, fast review and merge.

Summary

Add replaces field to fact metrics

Adds a field to fact metrics to link superseded metrics, enabling UI to show replacements and experiments to fall back to old metrics for snapshots, improving data continuity and migration.

Health Assessment

• • The PR had a long cycle time due to a large scope and a single review that required a follow-up commit, but the review was quick and the changes were well-documented.

Summary

Bulk fact metric API improvements

Enhances bulk import and metrics endpoints with new options, improved validation, and error handling, boosting reliability and flexibility for data migration workflows.

Health Assessment

• • PR required extensive rework and multiple review cycles, indicating complex changes and potential integration risk.

Summary

Implement Bayesian leaf selection for contextual bandits

This change dynamically determines the number of user groups (leaves) in contextual bandits using Bayesian Information Criterion, reducing overfitting and payload size.

Health Assessment

• • The PR required multiple iterations after the initial review, indicating some complexity in the algorithmic changes, but the overall cycle time remained fast.

Summary

Show CBAQ section on managed warehouse page

Users can view and edit Contextual Bandit Assignment Queries for managed warehouse connections directly from the datasource page, improving data experimentation workflow.

Health Assessment

• • PR completed in 12 minutes with a single commit and no review iterations, indicating a straightforward feature addition with minimal risk.

Summary

Fix warehouse column case-sensitivity validation

Ensures accurate column name validation across warehouses, preventing errors for BigQuery and ClickHouse, improving data integrity.

Health Assessment

• • The PR required multiple iterations and a significant amount of code changes, leading to a longer cycle time despite a quick initial review.

Summary

Docker build speedup

Optimizes Docker build by reducing node_modules churn, cutting build times from 200s to 130s locally and 450s to 270s on depot, improving deployment speed and resource efficiency.

Health Assessment

• • Fast turnaround with minimal changes; single review and quick merge indicates low complexity and high confidence.

Summary

Fix CI timeouts, memory limits, and test flake

Improves CI reliability by setting job timeouts, restoring memory limits, and stabilizing remote-eval tests, reducing build failures.

Health Assessment

• • Rapid resolution with minimal changes, indicating a straightforward fix with low risk.

Summary

docs: Correct Java cache TTL option name, document Kotlin/Swift equivalents

Updated SDK documentation to reflect correct option names and added missing Kotlin/Swift references, ensuring developers can configure cache TTL correctly and avoid compile errors.

Health Assessment

• • Fast review and merge cycle; minimal code impact; low risk to production.

Summary

Revert shared test mongod, fix revision lookup

Restores per-file MongoMemoryServer isolation to eliminate flaky tests and corrects ambiguous revision queries, improving CI stability and data integrity. This reduces false positives in test runs and ensures accurate revision handling for deployments.

Health Assessment

• • Quick fix with minimal changes, resolved flakiness and ambiguous lookup in a single PR, demonstrating efficient issue resolution.

Summary

Rewrite AI agent pages for MCP 2.x

Updated AI agent documentation to reflect MCP 2.x thin server and unified plugin install page, enhancing developer onboarding. Removed outdated references and clarified OAuth setup.

Health Assessment

• • The PR took almost two days to review and merge, indicating slow feedback loops for documentation updates. The large line changes suggest a comprehensive rewrite, but no code changes reduce risk.

Summary

Fix approval rule inheritance bug

This PR resolves a bug where disabled approval rules incorrectly inherited approver teams, preventing users from accessing settings screens, and adds a permissions summary card to improve visibility.

Health Assessment

• • Fast turnaround with minimal rework; AI-assisted review helped streamline the process.

Summary

Show marketing banner on getstarted empty state

Ensures new accounts see the marketing banner on the getstarted page, improving visibility and engagement.

Health Assessment

• • Quick, single-file change with minimal review, merged within 47 hours. No automated tests added.

Summary

Speed up local test runs

Optimizes test suite runtime by stubbing heavy modules, reusing MongoDB, and tuning worker settings, boosting developer productivity and CI efficiency.

Health Assessment

• • Fast review and merge cycle (1.1h total) with minimal rework; performance gains achieved with low risk and moderate code changes.

Summary

Improve approval flow clarity and early notification

Users now see which environments and teams trigger approvals before the first review, reducing surprise and streamlining the publishing process.

Health Assessment

• • Fast cycle time (9.3h) and single review round indicate efficient collaboration; high line count suggests substantial UI refactor but limited scope to frontend, keeping risk moderate.

Summary

Add error handling for unique index creation

Ensures the API remains operational during startup failures, improving reliability for Cosmos DB deployments and preventing login outages.

Health Assessment

• • Quick, single-commit fix with minimal code changes and fast review turnaround, indicating low complexity and low risk to production.

Summary

Speed up CI test suite by parallelizing and sharding

Reduces CI pipeline duration from ~15 minutes to ~5 minutes, accelerating feedback loops and cutting compute costs.

Health Assessment

• • The PR introduced parallelism and sharding in CI, significantly reducing test times with minimal rework and quick approval.

Summary

Health Assessment

Summary

Add project-scoped attribute targeting and bugfixes

Ensures attributes from all targeted projects are available for flags and experiments, reducing validation errors and improving configuration accuracy for users.

Health Assessment

• • Fast review and merge (under 3 hours) despite a large number of changes across 20 files, indicating efficient collaboration and AI-assisted review comments.

Summary

Add cross-fact-table funnel metrics to experiments

Enables funnel metrics to read from multiple fact tables, improving flexibility and performance for multi-table funnels.

Health Assessment

• • The PR had a quick initial review but required many subsequent commits to resolve conflicts and bugs, indicating complex changes. The large scope and multiple commits after the first review suggest higher maintenance risk for future releases.

Summary

Add effect standard error to API results

Adds posterior standard deviation of effect to API responses, enabling richer statistical insight for users.

Health Assessment

• • Quick turnaround with minimal changes indicates a straightforward enhancement.

Summary

fix(holdouts): Include Holdouts targeting in payload

Ensures holdout targeting is included in the payload, enabling accurate targeting and traffic control for experiments and improving data accuracy for stakeholders.

Health Assessment

• • The PR required multiple iterations to address holdout targeting logic, indicating moderate complexity and a need for careful testing.

Summary

Rename OpenAPI union variants for clearer Mintlify tabs

Ensures Mintlify displays descriptive titles for union types, improving documentation clarity and developer experience.

Health Assessment

• • Long review cycle and multiple merges indicate significant integration effort; the change touches core OpenAPI spec, affecting downstream tooling.

Summary

Prevent false concurrent‑edit warnings for migration fields

This change stops migration‑added fields from triggering false edit conflict warnings and displays project names in conflict rows, improving user experience.

Health Assessment

• • Fast cycle time, minimal rework, single review, small scope, low risk

Summary

Fix duplicate rows in feature attention table

Ensures accurate display of feature revisions, preventing duplicate rows and improving navigation to specific revisions.

Health Assessment

• • Quick fix with minimal changes, reviewed and approved within 15 hours, indicating low complexity and high confidence.

Summary

Move Reo docs tracker to Mintlify custom script

Enables Reo docs tracker to load automatically via Mintlify, improving analytics tracking without manual script inclusion.

Health Assessment

• • Long review cycle (~4.8 days) but minimal code changes; low risk to production.

Summary

Fix OAuth login failures across multiple tabs

Reduces authentication errors when users have several GrowthBook tabs open, improving login reliability and decreasing error logs.

Health Assessment

• • The PR was reviewed quickly (within 6 minutes) and required a single round of changes, indicating a focused fix. The change spans 308 lines across nine files, a large scope for a bugfix, but the low review friction suggests manageable complexity.

Summary

Silence pm2 getconf errors on hardened image

This change suppresses non‑critical getconf errors during container startup, improving log clarity and reducing noise for operators.

Health Assessment

• • Quick fix with minimal changes, fast review and merge, low risk.

Summary

Update Open in Explorer button size for consistency

Standardizes button size across product analytics pages, improving UI consistency and user navigation.

Health Assessment

• • Fast cycle time but required multiple commits after review, indicating iterative refinement of the UI change.

Summary

Restore review controls for approved revisions

Ensures that approvals from non-required teams do not hide review controls, improving workflow visibility and preventing gate failures.

Health Assessment

• • Resolved quickly with minimal rework; AI review identified the issue early, leading to a fast, low-friction fix.