diff --git a/docs/explorations/2026-07-05-privacy-analytics-cloudflare.md b/docs/explorations/2026-07-05-privacy-analytics-cloudflare.md
index 6e24537c..7e49b37a 100644
--- a/docs/explorations/2026-07-05-privacy-analytics-cloudflare.md
+++ b/docs/explorations/2026-07-05-privacy-analytics-cloudflare.md
@@ -39,10 +39,10 @@ beacon script 即可，最不破坏隐私定位。
 
 ## 验证（chrome-devtools + 构建产物）
 
-| 构建 | 预期 | 实测 |
-| --- | --- | --- |
-| 无 token（默认） | 不注入 | ✅ bundle 里零 `cloudflareinsights` 引用（死代码消除） |
-| 有 token · 顶层 | 注入 beacon | ✅ `data-cf-beacon={"token":...}` 正确 |
-| 有 token · `?embed=1` | 不注入 | ✅ beacon 缺席，`embed-mode` class 存在 |
+| 构建                  | 预期        | 实测                                                   |
+| --------------------- | ----------- | ------------------------------------------------------ |
+| 无 token（默认）      | 不注入      | ✅ bundle 里零 `cloudflareinsights` 引用（死代码消除） |
+| 有 token · 顶层       | 注入 beacon | ✅ `data-cf-beacon={"token":...}` 正确                 |
+| 有 token · `?embed=1` | 不注入      | ✅ beacon 缺席，`embed-mode` class 存在                |
 
 `lint:ts` ✅ / prettier ✅ / `build` ✅。
diff --git a/docs/explorations/2026-07-05-sw-stale-build-auto-reload.md b/docs/explorations/2026-07-05-sw-stale-build-auto-reload.md
index 729d0164..30132822 100644
--- a/docs/explorations/2026-07-05-sw-stale-build-auto-reload.md
+++ b/docs/explorations/2026-07-05-sw-stale-build-auto-reload.md
@@ -50,7 +50,7 @@ const hadController = !!navigator.serviceWorker.controller;
 let reloadingForUpdate = false;
 navigator.serviceWorker.addEventListener('controllerchange', () => {
   if (!hadController || reloadingForUpdate) return; // 1. 仅更新（非首装）；2. 只 reload 一次
-  if (getDocmentObj().fileName) return;             // 3. 有文档打开时不 reload，避免丢失编辑
+  if (getDocmentObj().fileName) return; // 3. 有文档打开时不 reload，避免丢失编辑
   reloadingForUpdate = true;
   window.location.reload();
 });
@@ -63,11 +63,11 @@ navigator.serviceWorker.addEventListener('controllerchange', () => {
 
 在 `vite preview`（dist）上：
 
-| 场景 | 预期 | 实测 |
-| --- | --- | --- |
-| 首次安装（无旧控制器） | 不 reload | ✅ 载入计数停在 1 |
+| 场景                                                         | 预期             | 实测                                                  |
+| ------------------------------------------------------------ | ---------------- | ----------------------------------------------------- |
+| 首次安装（无旧控制器）                                       | 不 reload        | ✅ 载入计数停在 1                                     |
 | 有旧控制器 + 未开文档（模拟新部署，合成 `controllerchange`） | 自动 reload 一次 | ✅ window marker 丢失、sessionStorage 保留 = 同页刷新 |
-| 文档已打开时 `controllerchange` | 不 reload | ✅ marker 仍在、编辑器 iframe 完好 |
+| 文档已打开时 `controllerchange`                              | 不 reload        | ✅ marker 仍在、编辑器 iframe 完好                    |
 
 `pnpm run lint:ts` 通过；`sw-routing.test.ts` 31/31 通过；`pnpm run build` 成功。
 
diff --git a/docs/explorations/2026-07-05-x2t-wasm-gzip-cf-pages.md b/docs/explorations/2026-07-05-x2t-wasm-gzip-cf-pages.md
new file mode 100644
index 00000000..4f9f8151
--- /dev/null
+++ b/docs/explorations/2026-07-05-x2t-wasm-gzip-cf-pages.md
@@ -0,0 +1,55 @@
+# x2t WASM 走 gzip 客户端解压 —— 解除 Cloudflare Pages 迁移的唯一硬阻塞
+
+> 2026-07-05
+
+## 背景
+
+document 项目计划从 `ranuts.github.io/document` 迁到 `edit.chaxus.com`(Cloudflare Pages)。
+唯一的托管硬阻塞:**CF Pages 单文件上限 25 MiB**,而 `public/wasm/x2t/x2t.wasm` = **55 M**,部署时会被直接拒收(dist 文件数 336,远低于 2 万上限,无其它阻塞)。
+
+目标:让浏览器仍拿到 55M 解压后的 wasm,但**部署产物里不再有超限的裸文件**。
+
+## 方案
+
+`x2t.js`(Emscripten 胶水)默认 `fetch('x2t.wasm')` + `instantiateStreaming`。但它:
+
+1. **复用已存在的全局 `Module`**(`if (!Module) Module = ...`,并 `moduleOverrides = Object.assign({}, Module)` 回填)。
+2. 第 284 行 `if (Module['wasmBinary']) wasmBinary = Module['wasmBinary']` —— 一旦 `Module.wasmBinary` 有值,`getBinarySync` 直接返回它、`instantiateAsync` 跳过 streaming fetch,**根本不请求 `x2t.wasm`**。
+
+于是在 `@ranuts/converter` 的 `loadScript()` 里、注入 x2t.js **之前**新增 `prepareWasmBinary()`:
+
+- `fetch(x2t.wasm.gz)`(11M,under 25MiB)
+- 用浏览器原生 **`DecompressionStream('gzip')`** 解压(无新依赖)
+- `window.Module = { ...window.Module, wasmBinary }`
+
+部署产物只保留 `x2t.wasm.gz`(11M) + `x2t.js`,删掉 `x2t.wasm`(55M)与 `x2t.wasm.br`(8M,brotli 需额外解码器,不用)。
+
+## 关键坑:服务器对 `.gz` 的处理不一致
+
+一开始直接 `DecompressionStream('gzip')`,在 `vite preview` 下报 `Z_DATA_ERROR: incorrect header check`。
+
+原因:**vite 的 dev/preview 服务器识别 `.gz` 后缀,给响应加 `Content-Encoding: gzip`**(实测 `Content-Type: application/wasm` + `Content-Encoding: gzip`),浏览器/fetch 层已透明解压了一次,我再手动解压就是对已解压数据二次解压。
+
+而静态托管(CF Pages / GitHub Pages)通常发**原始 gzip 字节**(不加 Content-Encoding)。两种行为都要兼容。
+
+**解法:按 magic bytes 探测**——收到的首字节是 `1f 8b`(gzip)才手动解压;是 `00 61 73 6d`(`\0asm`,已是裸 wasm)就直接用。对任何服务器都正确。
+
+## 验证(端到端)
+
+| 环节                                | 方法                                                                            | 结果                                                                            |
+| ----------------------------------- | ------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- |
+| isGzip=false 分支(preview 自动解压) | Node fetch preview 的 .gz → 探测 → compile                                      | 54.7M,magic OK,`WebAssembly.compile` OK(28 exports)                             |
+| isGzip=true 分支(静态托管原始 gzip) | Node 直接读磁盘 .gz 原始字节 → 手动解压 → compile                               | 54.7M,compile OK                                                                |
+| 浏览器集成                          | Playwright:注入 binary → 加载真实 x2t.js → 等 `onRuntimeInitialized` + 监听网络 | **运行时初始化 OK;FS/ccall 可用;对 `x2t.wasm` 请求 0 次(只请求 `x2t.wasm.gz`)** |
+| 构建产物                            | `find dist -size +25M`                                                          | **空**(最大文件 = x2t.wasm.gz 11M)                                              |
+| 回归                                | `pnpm run lint:ts` / `pnpm run test`                                            | 通过 / 19 files · 240 tests 全过                                                |
+
+## 结论 & 影响
+
+- **迁 CF Pages 的唯一硬阻塞已解除**:dist 无 >25MiB 文件。
+- 改动集中在 `@ranuts/converter`(`prepareWasmBinary`)+ 删两个大文件,主 app 代码零改动。
+- 对现有 GitHub Pages 部署也安全(magic 探测 + 现存 .gz)。
+
+## 遗留(不阻塞 CF 迁移)
+
+- `build:single`(单 HTML,`bin/bundle_single_html.js`)靠内联裸 wasm,删裸文件后该 target 的 wasm 需改为内联 `.gz` + 同款解压逻辑。非部署链,单列跟进。
diff --git a/packages/converter/src/document-converter.ts b/packages/converter/src/document-converter.ts
index 3e192d0e..5062aed9 100644
--- a/packages/converter/src/document-converter.ts
+++ b/packages/converter/src/document-converter.ts
@@ -36,15 +36,22 @@ export class X2TConverter {
 
   private readonly WORKING_DIRS = ['/working', '/working/media', '/working/fonts', '/working/themes'];
   private readonly SCRIPT_PATH = `${BASE_PATH}wasm/x2t/x2t.js`;
+  private readonly WASM_GZ_PATH = `${BASE_PATH}wasm/x2t/x2t.wasm.gz`;
   private readonly INIT_TIMEOUT = 300000;
 
   /**
-   * Load X2T script file (using ranuts scriptOnLoad utility)
+   * Load X2T script file (using ranuts scriptOnLoad utility).
+   *
+   * We first decompress the gzipped WASM and hand the bytes to Emscripten via
+   * `Module.wasmBinary` (see prepareWasmBinary), so x2t.js never fetches the raw
+   * 55 MB `x2t.wasm` itself. This lets us ship only the ~11 MB `x2t.wasm.gz`,
+   * staying under Cloudflare Pages' 25 MiB-per-file deploy limit.
    */
   async loadScript(): Promise<void> {
     if (this.hasScriptLoaded) return;
 
     try {
+      await this.prepareWasmBinary();
       // scriptOnLoad accepts an array of URLs
       await scriptOnLoad([this.SCRIPT_PATH]);
       this.hasScriptLoaded = true;
@@ -56,6 +63,50 @@ export class X2TConverter {
     }
   }
 
+  /**
+   * Fetch the gzipped x2t WASM, decompress it in the browser, and stash the raw
+   * bytes on `window.Module.wasmBinary` *before* x2t.js runs. Emscripten checks
+   * `if (Module['wasmBinary']) wasmBinary = Module['wasmBinary']` and then skips
+   * its own fetch/instantiateStreaming of `x2t.wasm` entirely.
+   *
+   * Uses the native `DecompressionStream('gzip')` — no extra dependency.
+   *
+   * Servers disagree on how they serve a `.gz` file: some (e.g. Vite's dev /
+   * preview server) send it with `Content-Encoding: gzip`, so the browser has
+   * already transparently decompressed it by the time we read the body; others
+   * (static hosts like Cloudflare Pages / GitHub Pages) serve the raw gzip
+   * bytes. We detect which by the leading magic bytes and only decompress when
+   * the payload is still gzip (`1f 8b`) rather than an already-raw wasm module
+   * (`00 61 73 6d`). This keeps it correct on every host.
+   */
+  private async prepareWasmBinary(): Promise<void> {
+    const globalScope = window as unknown as {
+      Module?: Record<string, unknown> & { wasmBinary?: ArrayBuffer };
+    };
+    if (globalScope.Module?.wasmBinary) return; // already prepared
+
+    const response = await fetch(this.WASM_GZ_PATH);
+    if (!response.ok) {
+      throw new Error(`Failed to fetch x2t WASM at '${this.WASM_GZ_PATH}' (${response.status})`);
+    }
+    const raw = await response.arrayBuffer();
+    const head = new Uint8Array(raw, 0, Math.min(2, raw.byteLength));
+    const isGzip = head[0] === 0x1f && head[1] === 0x8b;
+
+    let wasmBinary: ArrayBuffer;
+    if (isGzip) {
+      const stream = new Response(raw).body!.pipeThrough(new DecompressionStream('gzip'));
+      wasmBinary = await new Response(stream).arrayBuffer();
+    } else {
+      // Already decompressed by the browser via Content-Encoding.
+      wasmBinary = raw;
+    }
+
+    // Pre-seed the global Module so x2t.js (which reuses an existing global
+    // Module) picks up the binary. Preserve any properties already set.
+    globalScope.Module = { ...globalScope.Module, wasmBinary };
+  }
+
   /**
    * Initialize X2T module
    */
diff --git a/public/wasm/x2t/x2t.wasm b/public/wasm/x2t/x2t.wasm
deleted file mode 100644
index afe6a24f..00000000
Binary files a/public/wasm/x2t/x2t.wasm and /dev/null differ
diff --git a/public/wasm/x2t/x2t.wasm.br b/public/wasm/x2t/x2t.wasm.br
deleted file mode 100644
index c94ad7ec..00000000
Binary files a/public/wasm/x2t/x2t.wasm.br and /dev/null differ