v1 Prototype

Metric Catalog

8 metrics · v1.2-2026-05 · click any card to expand

How composite scores work

Each metric starts as a raw count (PRs, dollars spent, messages) in completely different units. We convert them all to a 0–100 scale so they can be combined — this is normalization. Then we take a weighted sum and rank everyone to produce one final score.

Step 1 — Normalize

Find the 95th-percentile raw value across all engineers (the "cap"). Each engineer's normalized score = min(raw / cap, 1) × 100. The cap prevents one outlier from compressing everyone else near zero.

Step 2 — Weight & sum

Multiply each normalized score by its weight (GitHub 25%, AI 25%, Jira 15% …) and add them up. An engineer scoring 60 on GitHub contributes 60 × 0.25 = 15 pts to their raw composite.

Step 3 — Rank-rescale

Sort all 936 engineers by their raw composite, then rescale so rank #1 = 100 and rank #936 = 0. A final score of 75 means you're in the top 25% of your role — it's a relative ranking, not an absolute number.

D-011 · Scores are role-scoped — Backend SWEs are ranked against Backend SWEs, not Mobile or ESYS engineers. Score model changes create a new versioned entry; no in-place edits.

Score Breakdown — Worked Example

Follow a hypothetical engineer through every calculation step, from raw signal to final score.

MetricRaw signalNormalized score / 100WeightContribution
GitHub PRs & Reviews
17 PRs
74
×25%18.5 pts
AI Tool Usage
$280/mo
62
×25%15.5 pts
Jira Engagement
140 issues
55
×15%8.3 pts
Drive Doc Creation
50 drive pts
38
×12%4.6 pts
Meeting Engagement
32 references
41
×10%4.1 pts
Slack Collaboration
286 messages
29
×8%2.3 pts
Confluence
6 comments
60
×5%3.0 pts
Σ weighted contributions56.2 pts
Step 3 — Rank-rescale
56.2 raw ptsranked #280 of 936 engineers(936 − 280) ÷ (936 − 1) × 100=

Rank #1 always = 100. Rank #936 always = 0. A score of 70 means this engineer is in the top 30% of their role.

70
Final Score

Metric Definitions

Normalized Score

concept

The intermediate 0–100 value for each metric, computed before weights are applied. This is what the bars on an engineer's profile page show — how they rank on each dimension individually.

GitHub PRs & Reviews

githubPresent

Counts the number of pull requests an engineer opened over the 90-day window. This is the single strongest signal of code output — how much working software is this person shipping? A PR is counted once when opened, regardless of size or merge status.

25%

AI Tool Usage

ai_usagePartial

Measures how much an engineer is using AI coding tools, expressed as total monthly spend (USD) across Claude, Cursor, and Copilot subscriptions attributed to their account. Spend is used as a reliable proxy for adoption intensity — engineers using these tools heavily accumulate more token cost.

25%

Jira Engagement

jiraPresent

Counts the distinct Jira issues an engineer touched over the 90-day window — created, updated, resolved, or commented on. Captures how actively someone is working the delivery pipeline, not just writing code.

15%

Drive Doc Creation

drivePartial

Measures written output in Motive's Engineering and Product Shared Drives. Creating a new doc is weighted most (×2) since it represents original work; revising an existing doc counts once (×1); leaving a comment counts half (×0.5). Captures specs, design docs, postmortems, and runbooks.

12%

Meeting Engagement

fellowPresent

Counts how many times an engineer's name appears in Fellow meeting notes over 90 days, queried via Glean search. Each reference represents that they were mentioned in an action item, decision, or meeting summary — a signal of cross-functional visibility and accountability.

10%

Slack Collaboration

slackPresent

Counts messages sent in public Slack channels over 90 days. Captures how visibly an engineer communicates — sharing updates, answering questions, and participating in team discussions. Only public channels are counted; DMs and private channels are excluded for privacy.

8%

Confluence Contributions

confluencePresent

Counts comments an engineer has left on Confluence pages over 90 days. A lightweight signal that someone is reading and reacting to internal documentation, design reviews, and RFCs — not just producing code.

5%

Glean Engagement

gleanN/A

Measures how actively an engineer uses Glean for knowledge search and AI assistance over the window. Glean Assist interactions (AI-powered Q&A) are weighted highest since they represent deeper engagement than a quick search; agent runs are highest of all.

0% (tracked, not scoring)

Model v1.2-2026-05 · Owner: Jay Rosenthal (BI Lead) · Governance approval required for any definition change