Skip to main content

Insights and Analytics

Analytics helps you understand what Buzzbin produced over a selected time range: how many reviews ran, what kinds of findings appeared, what it cost, and whether code quality signals are moving in the right direction.

Who can view it​

Organization members can view analytics for their own organization. Organization admins use the same data for configuration and cost decisions. System admins have an internal platform-wide view for product and model-quality monitoring.

Overview cards​

The overview commonly includes:

  • reviews run
  • findings surfaced
  • high and critical findings, including share of total
  • helpful percentage when enough feedback votes exist
  • IRR spend
  • total tokens used
  • estimated time saved

Estimated time saved is not a billing number. It is calculated from severity-weighted finding counts and should be shown with the assumptions used to derive it.

Breakdowns​

Findings can be grouped by:

  • category such as Security, Bug, Performance, or Testing
  • severity from Info to Critical
  • confidence
  • merge risk
  • repository
  • model

Every percentage should be read with its denominator. For example, 100% critical over one finding means something very different from 100% critical over one hundred findings.

Trends show how metrics change by day, week, or month. They help answer questions such as:

  • Are findings decreasing, or are fewer reviews running?
  • Did monthly spend increase because MR volume grew or because a different model was selected?
  • Which repository has the most high-risk findings?
  • Which category has the lowest helpful rate?

Most-common issues​

This view highlights recurring issue patterns by category, severity, and file path prefix. It does not perform free-text clustering; it surfaces actionable patterns from structured finding data.

Relationship to feedback​

Thumbs up/down reactions on findings help measure output quality. When a slice has too few votes, Buzzbin avoids showing a definitive helpful percentage and instead shows raw counts or a "not enough data" state.

For details, see Feedback and Finding Tuning.