Attrition Risk Detection

Attrition risk detection is an early warning system that identifies contributors who may be disengaging. It surfaces behavioral patterns — not conclusions — to help managers have informed conversations before it's too late.

Overview

The system monitors six behavioral signals derived from commit and PR data. Each signal is tracked independently, and a composite risk level is computed from their combination.

Route: /c/:slug/engagement

The Six Signals

1. Frequency Drop

What it measures: Declining commit frequency relative to the contributor's own baseline.

$$\text{frequency\_ratio}=\frac{\text{commits in last 14 days}}{\text{avg commits per 14-day period (last 90 days)}}$$

Threshold	Level
ratio > 0.7	Normal
0.4 < ratio <= 0.7	Warning
ratio <= 0.4	Elevated

The baseline is the contributor's own historical average, not a team benchmark. A contributor who typically ships 5 commits/week dropping to 2 is flagged, even though 2/week might be normal for someone else.

2. Score Decline

What it measures: Trending decrease in commit quality scores.

$$\text{score\_trend}=\text{avg score (last 14 days)}-\text{avg score (prior 30 days)}$$

Threshold	Level
trend > -0.5	Normal
-1.0 < trend <= -0.5	Warning
trend <= -1.0	Elevated

Minimum commits: 5 in each period (to avoid noise from small samples).

3. Activity Gaps

What it measures: Extended periods without any commits.

$$\text{gap\_days}=\text{now}-\text{last\_commit\_at}$$

Threshold	Level
gap < 7 days	Normal
7 <= gap < 14 days	Warning
gap >= 14 days	Elevated

Activity gaps overlap with the Silent Contributor alert but serve a different purpose. Alerts notify managers of current silence; attrition risk tracks gaps as one signal among many.

4. Scope Narrowing

What it measures: A shrinking set of files and domains being touched.

$$\text{scope\_ratio}=\frac{\text{unique domains (last 14 days)}}{\text{avg unique domains per 14-day period (last 90 days)}}$$

Threshold	Level
ratio > 0.6	Normal
0.3 < ratio <= 0.6	Warning
ratio <= 0.3	Elevated

A contributor whose work narrows from 5 domains to 1 may be coasting on familiar territory or mentally checking out. Alternatively, they may be deeply focused on a critical project — context matters.

5. Session Shortening

What it measures: Deep work sessions becoming shallow micro-sessions.

$$\text{deep\_session\_ratio}=\frac{\text{deep sessions (last 14 days)}}{\text{total sessions (last 14 days)}}$$$$\text{baseline\_ratio}=\frac{\text{deep sessions (last 90 days)}}{\text{total sessions (last 90 days)}}$$

Threshold	Level
current ratio > baseline * 0.7	Normal
baseline * 0.4 < current <= baseline * 0.7	Warning
current <= baseline * 0.4	Elevated

Deep sessions (sustained work periods producing multiple meaningful commits) are a strong indicator of engagement. A shift toward micro-sessions (single-commit bursts with long gaps) may signal distraction or disinterest.

6. Review Withdrawal

What it measures: Declining participation in PR reviews and discussions.

$$\text{review\_ratio}=\frac{\text{PRs reviewed (last 14 days)}}{\text{avg PRs reviewed per 14-day period (last 90 days)}}$$

Threshold	Level
ratio > 0.6	Normal
0.3 < ratio <= 0.6	Warning
ratio <= 0.3	Elevated

Review participation is a strong team engagement signal. Contributors who stop reviewing others' code are often withdrawing from the team socially before they withdraw formally.

Composite Risk Level

The six signals are combined into an overall risk assessment:

Risk Level	Condition
Low	0–1 signals at Warning, none at Elevated
Medium	2+ signals at Warning, or 1 at Elevated
High	2+ signals at Elevated

The composite level is a summary, not a verdict. Managers should examine individual signals for context.

Philosophy

Attrition risk detection walks a fine line. Done poorly, it becomes surveillance — a system that punishes people for taking vacation or having a slow week. ShipLens takes a deliberate stance:

Surface data to support conversations, not to replace them.

Every signal has legitimate explanations. A frequency drop could mean vacation. Scope narrowing could mean deep specialization on a critical project. Session shortening could mean a shift to architecture and planning work that doesn't produce commits.

The system provides the data. The manager provides the judgment. ShipLens will never:

• Automatically flag someone as "at risk" in a way visible to their peers
• Send automated alerts to skip-level managers
• Use attrition risk in scoring or performance evaluations
• Treat any single signal as conclusive

TIP

Attrition signals are visible only to company admins and direct managers. They are excluded from contributor-facing views and are never included in team-wide reports.