Servo Baseline Readiness

Data Sources

Web Features — The web-features package defines 1,129 curated feature groups that map to Browser Compat Data (BCD) keys. Each feature has a Baseline status: Widely Available (high) means the feature has been supported across all core browsers for 30+ months. This analysis focuses on the 593 Widely Available features as the target set.

WPT Feature Manifest — The Web Platform Tests project maintains WEB_FEATURES_MANIFEST.json, which maps 841 web-feature IDs to 53,549 WPT test paths. This is the critical bridge from feature definitions to measurable test outcomes.

WPT Results — Servo runs the full WPT suite daily on wpt.fyi. We use the summary_v2.json format, which records per-test status and subtest counts. Historical snapshots at 6 quarterly intervals (2023-Q3 through 2026-Q1) provide the longitudinal data.

Scoring Model

Each feature's score is the average per-test pass rate across all matched WPT tests. For tests with subtests, the score is passed_subtests / total_subtests. For simple pass/fail tests, the score is 0 or 1. A feature is considered fully supported at ≥95% score.

Features with no WPT manifest mapping (154 of 593) are categorized as “No Data.” These break down into 78 JS built-ins (supported via SpiderMonkey, tested by test262 not WPT), 23 semantic HTML elements (trivial — no special behavior to test), 20 WebGL extensions (GPU-driver-dependent), and 33 other features (basic DOM, CSS, HTTP, WebAssembly). The 78 JS built-ins and 23 semantic HTML elements are supported via SpiderMonkey and the HTML parser respectively; the 20 WebGL extensions depend on GPU driver support; the remaining 33 have unknown status.

FTE-Equivalent Contributors

The contributor count uses full-time equivalents (FTE), not raw headcount. Computed from per-author commit counts over Jul 2025 – Jan 2026 (7 months), excluding bots (dependabot, WPT Sync). Each author's FTE fraction = min(their_commits / 154, 1.0), where 154 = 22 commits/month × 7 months (~1 commit per working day). The sum across all 115 human authors yields ~13 FTE. Of these, 9 core contributors operate at ≥50% FTE, with a long tail of occasional contributors. This FTE figure is what drives the projection and cost models — it represents the effective engineering capacity, which is what matters for funding decisions.

Velocity & Projections

Per-feature velocity is computed as the improvement per quarter from 2023-Q3 to present. We use max(overall_velocity, recent_velocity) where recent velocity covers the last 4 quarters, to account for accelerating progress. Features with velocity ≤0.001 percentage points/quarter are classified as stalled.

Projections extrapolate each feature individually: quarters_to_95% = (0.95 - current_score) / (velocity × scale_factor). The scale factor models additional FTEs using a sublinear power law: scale = (new_fte / current_fte)exponent. The default exponent of 0.7 reflects Brooks's Law — communication overhead means doubling contributors doesn't double output.

Cost Model

Cost is calculated as: additional_FTE × salary × years_to_milestone, where the milestone date is determined by the Nth-percentile feature reaching 95%. This represents the marginal cost of accelerating beyond the current trajectory — not the total program cost. The model assumes FTEs can be allocated to stalled features and that the scaling exponent applies uniformly.

Salary Assumptions

The default salary of €200k/yr is composed of a €150k base (European senior software engineer median total cost to employer) × a 1.33× specialization multiplier for two factors:

• Highly specialized domain — browser engine internals (Rust, rendering pipelines, layout algorithms) draws from a very small global talent pool
• Multi-disciplinary practice — each contributor combines software development, W3C/WHATWG standards body participation, and open-source community contribution management

Reference rates used to calibrate this figure:

Limitations

• 26% of Widely Available features lack WPT test mapping, creating a coverage gap
• Servo is not tracked in BCD, so Baseline status cannot directly include Servo
• WPT test coverage varies by feature — some features have thousands of tests, others have few
• The sublinear scaling model is a simplification; actual contributor efficiency depends on the nature of the work
• Projections assume constant velocity — breakout improvements in layout, CSS, or JS engines could cause step-function changes
• Stalled features may require architectural work rather than incremental test fixes