Matrix Quadrant Analysis

Matrix Quadrant Analysis (Matrix) places features/events into four quadrants along two axes (users UV × usage frequency), identifying core, star, niche, and edge features. It corresponds to the backend AnalyticsService.Matrix.

1. What Questions It Answers

Which features are "core" (many users, frequent use)? Which are "edge" (few users, rare use)?
Are there "star features" (few users but extremely high per-user frequency — hidden gems)?
How are features distributed across the matrix, to guide product resource investment priorities?

2. The Four Quadrants

Defined by thresholds (median/average) on the X axis (users UV) and Y axis (usage frequency):

Quadrant	Identifier	Characteristics
Core	`core`	Many users + high frequency
Star	`star`	Few users + high frequency (niche but intensive; potential)
Niche	`niche`	Few users + low frequency
Edge	`edge`	Many users + low frequency (broad reach but shallow use)

3. Backend API

gRPC: `AnalyticsService.Matrix`

Field	Type	Description
`time_range`	`TimeRange`	Analysis time range
`dimension`	string (optional)	Analysis dimension, default `event_name`
`app_id`	uint32 (optional)	Filter by app
`split`	string (optional)	Quadrant threshold type: `median` (default) / `avg`

Response MatrixResponse: points[] (MatrixPoint contains label dimension value / x user count / y usage frequency / quadrant), x_threshold, y_threshold, dimension.

HTTP (admin proxy)

POST /admin/v1/analytics/matrix
Content-Type: application/json

{
  "timeRange": { "startMs": 1718169600000, "endMs": 1718774399000 },
  "dimension": "event_name",
  "split": "median"
}

4. Interpreting Results

Each MatrixPoint is one dot on the scatter plot (one event/feature):

x = distinct user count (breadth of coverage).
y = total usage count (depth of use).
quadrant = the quadrant the dot falls into.

x_threshold / y_threshold are the thresholds dividing the quadrants (median or avg, per split). The frontend renders this as a four-quadrant scatter plot.

5. Typical Scenarios

Feature priority assessment

Feed each feature (dimension=event_name) into the matrix:

Core zone: maintain continuously — the product's foundation.
Star zone: few users but intensive use — possibly a hard requirement of power users, with room to grow; worth promoting.
Edge zone: broad reach but shallow use — improve the experience or consider trimming.
Niche zone: assess whether it justifies its maintenance cost.

Content/column evaluation

Switch dimension to a content-type dimension to see which columns are core/star.

6. Notes and Caveats

Two-axis semantics: x is distinct UV and y is total count; high y with low x means heavy use by a small group.
Threshold choice: median resists extremes (recommended); avg is pulled up by head features. The two splits may produce different results — choose as needed.
Dimension limits: dimension is usually event_name or a whitelisted dimension.
Empty data: events must be stored first (see Getting Started · Current State of Data Storage).