Lock the behavior
Begin with the same project-level workload and observable result. Runtime, native, C, or package-contract gates are shown where that evidence exists.
Generated evidence catalog
Every project, lane, and byte
Search projects, isolate a fair mangling mode, and sort Brotli, gzip, or raw output. Project links open a source-and-method detail page in a new tab.
Rows are measurements, not one global race. Only artifacts inside the same project and compatibility boundary are directly comparable.
Current selection
Overall averages
These values update with the project, status, tool, mangling, and search filters above.
| Project | Category | Status | Artifact lane | Tool | Mangling | Properties | Brotli-11 · primary | Gzip-9 | Raw |
|---|
How comparison works
Hold the contract steady, then measure every lane
Each project is its own comparison boundary. Its detail page records the workload, expected output, source, compatibility status, and fairness rule; the table then reports the emitted JavaScript under several optimization and mangling policies.
Choose a legal boundary
Identifier-only rows preserve ordinary property names. Private-prefix and closed-world rows may rename properties only when their stated boundary allows it.
Emit independent lanes
Vite, Closure, LilScript, and combined pipelines are separate artifacts. Unminified output is context; it is not treated as an equivalent advanced-minification lane.
Measure transport
Brotli quality 11 is shown first, followed by gzip level 9 and raw UTF-8 bytes for every JavaScript artifact. Every codec starts from that row's exact emitted output.
compression saved = (raw − compressed) ÷ raw × 100
Brotli edge = (gzip − Brotli) ÷ gzip × 100
Smaller raw can lose after compression
Repeated long tokens compress efficiently. A shorter minified file may have less repetition, so raw, gzip, and Brotli rankings can legitimately differ.
Tiny files distort percentages
Codec framing and dictionaries are a larger share of tiny outputs. That is why the page shows both equal-row averages and byte-weighted totals.
Global averages are descriptive
Tools do not participate in identical numbers of projects or modes. Use the averages to understand this catalog, not to declare a universal compiler winner.
Compare inside one project
For a head-to-head conclusion, open a project and compare rows with the same behavior and property-mangling boundary. Cross-project byte totals are not equivalent workloads.
Fairness rule: a closed-world/property-mangled result is only directly comparable with another row allowed to rename the same properties. Public-safe and unmangled rows answer different deployment questions.