Understand where each developer excels and where they can grow. The Skill Matrix maps every team member’s proficiency across four specialist domains — Security, Performance, Correctness, and Patterns — using a color-coded heatmap and a polar radar chart that make strengths and gaps immediately visible.
Scores are computed from real code review findings over a rolling 90-day window, so the matrix always reflects recent work rather than historical baggage. For a complementary view that combines quality with productivity, see the Leaderboard.
What you get
- Color-coded heatmap showing all developers’ scores at a glance
- Per-developer radar chart showing proficiency shape across four domains
- Team-level overlay comparing top performers against the team average
- Auto-derived strengths and development areas for every developer
- Weekly findings trend and resolution tracking
- Recurring pattern identification through an AI-generated force graph
The four domains
Each domain corresponds to a specialist agent that reviews your code:
| Domain | Specialist | What it measures |
|---|
| Security | Security specialist | Vulnerabilities, secret exposure, injection risks, auth issues |
| Performance | Performance specialist | Inefficient queries, memory leaks, N+1 problems, unnecessary allocations |
| Correctness | Correctness specialist | Logic errors, edge cases, type mismatches, missing validations |
| Patterns | Patterns specialist | Anti-patterns, code smells, style violations, maintainability concerns |
How scores are calculated
Scores use Bayesian shrinkage — a statistical method that accounts for how much data we have about each developer. Developers with fewer MRs have their scores pulled toward the team average, preventing unreliable extreme scores from small samples.
- Gather findings — All findings from Advanced code review specialist agents over the past 90 days are grouped by developer and domain.
- Normalize — For each developer in each domain,
findings per MR is calculated.
- Shrink toward team average — Each developer’s rate is blended with the team-wide average. Developers with many MRs keep a rate close to their observed value; developers with few MRs are pulled closer to the team average.
- Score against threshold — The adjusted rate is compared against a domain-specific threshold:
Score = max(0%, 1 - adjusted_rate / threshold).
- Average — The four domain scores are averaged into a single Overall Score.
Developers need at least 3 MRs reviewed in Advanced mode to appear on the Skill Matrix. Fast mode reviews are not included because they do not use domain-specialist agents.
Why Bayesian shrinkage?
A developer who submitted 7 MRs with zero findings is not necessarily better than one who submitted 67 MRs with some findings — we simply have less data about them. Bayesian shrinkage solves this by treating small samples with appropriate uncertainty:
- High-volume developers — their score closely reflects their actual finding rate
- Low-volume developers — their score is pulled toward the team average until they have enough data to stand on its own
This prevents the common pitfall where new team members or infrequent contributors appear with misleadingly perfect (or terrible) scores.
Understanding your score
| Overall Score | Label | Interpretation |
|---|
| 80% or above | Excellent | Consistently clean code across all domains |
| 60 – 79% | Good | Above average with minor areas to improve |
| 40 – 59% | Fair | Room for growth in several domains |
| 20 – 39% | Needs Work | Significant findings in some domains |
| Below 20% | Poor | High finding rates across multiple domains |
Focus on your lowest-scoring domain first. Moving from 40% to 60% in one area has a bigger impact on your overall score than going from 85% to 95% in another.
Advanced mode only
The Skill Matrix exclusively uses findings from Advanced code review, where four specialist agents (Security, Performance, Correctness, Patterns) independently analyze your code. MRs reviewed in Fast mode use a single general-purpose agent whose findings are not classified by domain, so they cannot contribute to domain-specific scoring.
If your workspace uses Fast mode, the Skill Matrix will show fewer qualifying developers or may be empty. Switch to Advanced mode in your code review settings to get full Skill Matrix coverage.
Where Skill Matrix appears
Team Skill Matrix (heatmap)
The Skill Matrix tab on the Analytics page shows a heatmap table of all qualified developers. Each cell is color-coded from green (excellent) to red (poor), with an overall score and MR count per developer. A Team Average row at the bottom provides the baseline. Click any developer row to open their profile.
Skill Radar (polar chart)
Below the heatmap, a polar radar chart overlays the top 3 developers with a dashed Team Average line. This gives a quick visual snapshot of who is leading in each domain and where the team shape differs.
Developer Profile dialog
Click any developer to open their full profile, which includes:
| Section | Description |
|---|
| Profile header | Avatar, MR count, total findings, average quality score, overall skill percentage |
| Skill Radar | Polar chart of all four domains for this developer |
| Resolution Rate | Percentage of findings resolved before merge, shown as a progress bar |
| Focus Areas | Auto-derived strengths (top 2 domains) and development areas (bottom 2 domains) |
| Weekly Findings Trend | Stacked column chart of findings by severity over the past 90 days |
| Pattern Force Graph | AI-identified recurring patterns visualized as a force-directed network |
Use cases
1. Targeted skill development
Use the auto-derived Focus Areas to guide 1:1 conversations. If a developer scores 90% in Correctness but 45% in Security, pair them with a security-focused mentor or assign security-related tasks to build that skill.
2. Balanced code review assignment
Route critical PRs to reviewers who score highest in the relevant domain. A developer with 95% Security is the right person to review authentication changes.
3. Tracking improvement over time
The Weekly Findings Trend shows whether a developer’s finding count is decreasing. A downward trend in a previously weak domain validates that coaching or training is working.
4. Team composition planning
The team overlay radar reveals collective blind spots. If the entire team dips in Performance, consider investing in performance-focused training or tooling.
Scores are based on absolute thresholds, so they are stable across team changes. However, the team average row will shift as members join or leave. Use individual scores for personal growth tracking and the team average for organizational benchmarking.
