LilScript

Measured ports

Same work. Count the bytes.

These are libraries and apps we rewrote in LilScript to see what the compiler actually does. Numbers are independently measured gzip-9 and Brotli-11 bytes of the selected JavaScript, scored with the same stock zlib / official Brotli binaries. On the 15 currently comparable landing rows, the median result is 7.7% smaller under gzip and 7.5% smaller under Brotli. Zod is visible but excluded after its package surface audit invalidated the old core-only size row.

Mangling contract. On reusable-library rows, mangle: true means identifier mangling only. Oxc renames variables and private class fields; Terser property mangling is a separate option and is off here. Public ESM names and documented object keys must survive, and behavior gates run on the measured artifacts. Property mangling appears only on explicitly labeled closed-world app lanes or on LilScript-owned internals whose public ABI remains named. If Function.name, arity, or constructibility belongs to a library contract, the verifier must pin it too. See the Oxc mangling documentation and Terser API reference.

Monaco · VS Code’s editor

monaco-editor 0.56, JavaScript versus LilScript.

The JS page is npm monaco-editor. The Lil page is compiled from ports/monaco/entry.lil plus 992 catalog modules. Workers except TypeScript are LilScript; both pages still load Microsoft ts.worker so diagnostics match. The view is a functioning editor — create, type, undo, find, fold, hover, suggest — not a claim that every VS Code workbench feature is present.

Open monacolil Port source
npm ide.js 887,420 B
Lil ide.js 413,607 B
JS + workers, no tsc 1,416,410 B
Lil + workers, no tsc 415,045 B

Frameworks and apps

Runnable products, with a scope line on each card.

SolidLil · keyed app

Solid 2.0 vs SolidLil

Official js-framework-benchmark keyed table. Solid 2.0 on the left, compiler-only SolidLil on the right.

Solid 2.011,180
SolidLil3,862
Brotli−65%
 Historical Solid 1.9

solid-js Core

Archived 1.9 open-world Core row. Current published port is Solid 2.0 on the SolidLil lab.

Solid Core8,551
SolidLil8,433
Brotli−1.4%
 Historical Solid 1.9

solid-js Store

Archived 1.9 Store surface. Not the current Solid 2.0 client row.

Solid Store4,286
SolidLil4,231
Brotli−1.3%
 Lilastro app

Lastro

Parcel Market versus the accessibility-matched Astro control.

Astro2,672
Lastro2,665
Brotli−0.3%
 motion@13 surface

Motion

mix / wrap / stagger / spring. Live carousel and spring fixtures in the Motion lab. Full DOM package is still backlog.

npm Vite4,044
LilScript2,333
Brotli−42%
 Historical LSX app

Solid todolist

Archived JSX vs LSX todolist. −32% Brotli. Not the current integrated lab gate.

Solid JSX5,479
SolidLil3,722
Brotli−32%

Complete and exact entrypoints

Published roots, same input domain.

Eligible rows beat npm/Vite and public-API Closure ADVANCED on the selected codec, and stay within 1.05× on the isolated runtime gate.

Exact root

gl-matrix

Complete ESM namespace and setMatrixArrayType.

npm Vite14,330
LilScript14,116
Brotli−1.5%
 Complete package

murmurhash-js

MurmurHash 2 and 3. Reusable API, not an app-specialized kernel.

npm Vite902
LilScript409
Brotli−55%
 Complete package

js-levenshtein

Full root. Eligible on size and throughput.

npm Vite825
LilScript404
Brotli−51%
 Complete package

clamp + lerp

Both published roots compiled together.

npm Vite519
LilScript112
Brotli−78%
 Complete package

@emotion/hash

Published root versus Vite and Closure ADVANCED.

npm Vite330
LilScript231
Brotli−30%
 Complete package

robust-predicates

Large typed suite. The cut is closer to the 5–10% band.

npm Vite6,456
LilScript5,972
Brotli−7.5%
 Exact browser entry

Nano ID

Published browser entrypoint. Effectively tied.

npm Vite409
LilScript408
Brotli−0.2%
 Exact root

mitt

Complete default-export surface. Currently a few Brotli bytes up.

npm Vite300
LilScript302
Brotli+0.7%
 Exact root

clsx

Recursive class-name builder. Size gate is open on Brotli.

npm Vite493
LilScript497
Brotli+0.8%

Selected surfaces

Not the whole package. Still a real rewrite.

These rows implement a declared subset. They are useful pressure tests. They are not “we replaced Preact.”

Candidate surface

immer

produce path used by the app contract.

npm Vite4,133
LilScript3,721
Brotli−10%
 Core subset

Preact 10

Selected render surface, not the full Preact package.

npm Vite5,988
LilScript2,789
Brotli−53%
 Store subset

Redux Toolkit

Store core used by the contract. Not the full toolkit.

npm Vite6,212
LilScript3,442
Brotli−45%
 In-repo Zod 3 subset

Zod 3

Schema parse surface in the app. Not the published Zod 4 port.

npm Vite12,518
LilScript4,138
Brotli−67%
 Parse subset

Acorn 8

Parser subset required by the contract app.

npm Vite29,529
LilScript4,407
Brotli−85%
 Complete easing package

@motionone/easing

Full published easing root. Smaller than Vite and Closure; the isolated throughput gate is still just over 1.05×.

npm Vite431
LilScript266
Brotli−38%

Published · not smaller

jQuery 3.7.1 is published. It is not smaller yet.

The public-library row is a pressure test: host-heavy, string keys, a plugin facade. Behavior lanes for core, deferred, data, events, selectors, ajax, and effects are in the repo. The current Brotli-oriented compiler output is larger than npm jquery.min.js. That is stated here so the other cards stay honest.

Full surface · not eligible

jQuery

Dual configs: stable $ for a script tag, full mangle for a LilScript-only app. Official min is still 1.12× smaller.

npm min27,445
LilScript30,741
Brotli+12%

Parse API

marked 18.0.13 package, scored without the extension system.

The 18.0.13 package exposes the parse API, live option methods, GFM, and CommonMark HTML, with 660/660 against marked@18.0.10. This port has no use(). Official JavaScript is the parse-only 18.0.10 sources, then Oxc and Terser with identifier mangling enabled or disabled; property mangling is off. The published LilScript artifact is the compiler-selected ESM — not bundled, not post-minified, and still smaller on raw, gzip-9, and Brotli-11 than either minifier, while parsing the same corpus faster in Chromium.

Parse API only

marked

Fair size lane is parse-only official Oxc with identifier mangling, not the npm file that still includes Hooks and use(). All measured lanes run the same 660-case browser corpus.

Oxc identifiers10,092
LilScript9,515
Brotli−5.7%

Zod 4.4.3 · size audit open

The behavior result remains; the size win does not.

zod@4.4.3 rewritten in LilScript and published as @itslil/zod. The adapted official suite is 1,353/1,353, but the old 34,152 B Brotli figure measured only zod.core.js while the official lane bundled the full zod/v4 entry. Linking the actual package entry produces 62,763 B gzip and 52,583 B Brotli, before any claim about parity. That package exposes 192 ESM names versus 240 on the official entry, so no compression rate is published until the boundary matches.

Classic-compatible API · not size-eligible

Zod

The old core-only row is withdrawn. The linked package is a useful diagnostic, but 192 and 240 exports are not the same reusable library surface.

Official exports240
Port exports192
Size claimWithdrawn

Six scoped lanes · posthog-js 1.418.10

A capture kernel plus five independently measured packs.

UUID, flag utils, cookie identity, request routing, token and bucketed rate limits, queue batching, bot detection, and portable helpers — not the published PostHog IIFE. Official JavaScript is that kernel, then Vite 8 Oxc, Terser, and esbuild. The current Brotli-scored compiler artifact is larger than Oxc on this compiler. Autocapture utilities, replay core, surveys, error tracking, and OTLP now have separate public surfaces and tabbed comparison lanes. The two browser packs are 24.4% and 18.6% smaller than Oxc on Brotli; the lab reports every pack's wins and losses without folding them into the kernel result.

Kernel result · 5 packs on the lab

posthog-js

Fair size lane is official Oxc with identifier mangling only, not the npm client bundle. Property mangling is off; the source kernel and port share the same behavior contract. 5,622 B → 5,985 B Brotli.

Oxc identifiers5,622
LilScript5,985
Brotli+6.5%

Open the pairs

Left is the original. Right is the compile.

These are the hosted comparisons. When this site is on GitHub Pages, they are the github.io URLs.

SolidLil

yeargun.github.io/solidlil

Solid 2.0 client lab and the official keyed table (11,180 B → 3,862 B).

Open SolidLil Motion

yeargun.github.io/motionlil

Full Motion 13 DOM runtime exports and 16 paired browser fixtures.

Open Motionlil MobX

yeargun.github.io/mobxlil

MobX 7 public API, official tests, size vs Vite 8 Oxc.

Open MobXlil jQuery

yeargun.github.io/jquerylil

jQuery 3.7.1 drop-in. Official min is still 1.12× smaller on Brotli.

Open jQuerylil marked

yeargun.github.io/markedlil

marked 18.0.13 package with the 18.0.10-compatible parse API. 9,515 B vs parse-only official Oxc 10,092 B, not the full use() package.

Open Markedlil Zod

yeargun.github.io/zodlil

Zod 4.4.3 classic-compatible API. Size claim withdrawn: the linked package exposes 192 ESM names versus the official entry's 240.

Open Zodlil posthog-js

yeargun.github.io/posthoglil

Capture kernel plus five tabbed lanes. Autocapture is 4,215 → 3,186 B and replay core is 4,258 → 3,465 B Brotli versus Oxc.

Open Posthoglil Monaco

yeargun.github.io/monacolil

Submodule catalog and the paired editors.

Open Monacolil This site

In-repo demos

Lastro, hashes, algorithms, Closure pairs — Solid keyed, Motion carousel.

Browse demos

More tables. Filterable corpus: library lab, benchmarks, explorer. Delivery architecture: Lilpack. Language: syntax and behavior.