diff --git a/CHANGELOG.md b/CHANGELOG.md
index a3df507..3003751 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,14 @@ This file records user-visible changes to `light-ocr`. Published artifact detail
 
 ## [Unreleased]
 
+## [0.5.6] - 2026-07-31
+
+### Changed
+
+- Reorganized the project README and replaced the minimal npm package README
+  with a complete install, image/PDF API, CLI, offline distribution, platform,
+  model-tier, resource-limit, and troubleshooting guide.
+
 ### Fixed
 
 - Bundled the checksum-pinned Noto Sans SC regional subset into every native
@@ -14,6 +22,9 @@ This file records user-visible changes to `light-ocr`. Published artifact detail
   page bytes, and end-to-end Chinese OCR without install-time or runtime
   downloads.
 
+Release preparation: [English](docs/releases/npm-0.5.6.en.md) /
+[中文](docs/releases/npm-0.5.6.md).
+
 ## [0.5.5] - 2026-07-27
 
 ### Changed
diff --git a/README.md b/README.md
index 51b8336..d1bf7a6 100644
--- a/README.md
+++ b/README.md
@@ -14,7 +14,18 @@ English | [简体中文](README.zh-CN.md)
 
 **Fast, offline OCR for Node.js and C++.**
 
-Recognize text in PDF, JPEG, PNG, or raw image data directly on your machine. `light-ocr` returns lines in reading order with confidence scores and quadrilateral coordinates. For Node.js, the npm package includes PP-OCRv6 Small, PDFium, and prebuilt components for macOS, Linux, and Windows.
+Recognize text in PDF, JPEG, PNG, or raw image data directly on your machine.
+`light-ocr` returns lines in reading order with confidence scores and
+quadrilateral coordinates. The Node.js package includes PP-OCRv6 Small,
+PDFium, a Chinese fallback font, and prebuilt components for macOS, Linux, and
+Windows—without postinstall or first-run downloads.
+
+| | |
+| --- | --- |
+| **Best for** | local OCR in Node.js apps, CLIs, desktop software, and native C++ integrations |
+| **Inputs** | JPEG, PNG, PDF, encoded bytes, or decoded pixel buffers |
+| **Outputs** | text, confidence, quadrilateral boxes, page metadata, and timing |
+| **Distribution** | one npm install; CommonJS, ESM, TypeScript, and six prebuilt platforms |
 
 ## Quick start
 
@@ -29,30 +40,36 @@ import { createEngine } from "@arcships/light-ocr";
 import { readFile } from "node:fs/promises";
 
 const engine = await createEngine();
-const result = await engine.recognizeEncoded(
-  await readFile("image.jpg"),
-);
 
-for (const line of result.lines) {
-  console.log(line.text, line.confidence, line.box);
-}
+try {
+  const result = await engine.recognizeEncoded(
+    await readFile("image.jpg"),
+  );
 
-await engine.close();
+  for (const line of result.lines) {
+    console.log(line.text, line.confidence, line.box);
+  }
+} finally {
+  await engine.close();
+}
 ```
 
-`createEngine()` automatically chooses the right execution mode for the current platform. If your application already decodes images, [`recognize()`](bindings/node/README.md#usage) also accepts `GRAY8`, `RGB8`, `BGR8`, and `RGBA8` pixel data.
+`createEngine()` automatically chooses the right execution mode for the
+current platform. If your application already decodes images,
+[`recognize()`](packages/light-ocr/README.md#recognize-decoded-pixels) also
+accepts `GRAY8`, `RGB8`, `BGR8`, and `RGBA8` pixel data.
 
-### Model tiers
+PDF pages use the same package:
 
-Small remains the stable default. N2 also provides two opt-in preview packages under the `next` tag; all three expose the same API, types, result schema, and error model, while each install contains only its selected model.
+```ts
+import { recognizeDocument } from "@arcships/light-ocr";
 
-| Tier | Package / command | Model payload | Status |
-| --- | --- | ---: | --- |
-| Small | `@arcships/light-ocr` / `light-ocr` | ~30 MB | stable default |
-| Tiny | `@arcships/light-ocr-tiny@next` / `light-ocr-tiny` | ~6.3 MB | preview; 49 languages, no Japanese |
-| Medium | `@arcships/light-ocr-medium@next` / `light-ocr-medium` | ~139 MB | preview; quality-first |
+for await (const page of recognizeDocument("report.pdf", { dpi: 200 })) {
+  console.log(page.index, page.lines.map((line) => line.text));
+}
+```
 
-Tiny and Medium stay on `next` until real use shows a clear reason to promote them; they do not change what `npm install @arcships/light-ocr` installs.
+CommonJS uses the same exports through `require("@arcships/light-ocr")`.
 
 ## CLI
 
@@ -86,9 +103,10 @@ Image commands are `recognize` (default), `detect` (boxes only), `info`
 directly to document OCR; `document` handles explicit multi-source jobs. Output
 uses a versioned `schemaVersion: 1` contract. EXIF orientation is corrected
 automatically. See the [CLI design](docs/cli-design.md) and
-[npm README](bindings/node/README.md#cli) for full reference.
+[npm package README](packages/light-ocr/README.md) for the complete
+install-and-use reference.
 
-### PDF and multi-page documents
+## PDF and multi-page documents
 
 PDF and multi-page OCR are built into `@arcships/light-ocr`. The matching
 PDFium binary and checksum-pinned Noto Sans SC fallback font are carried by
@@ -124,15 +142,6 @@ for await (const page of recognizeDocument([buf1, buf2, buf3])) {
 }
 ```
 
-## Agent Skill
-
-An [Agent Skill](.agents/skills/local-ocr/SKILL.md) is included for AI agents that can call local commands. It provides scenario-driven workflows, a decision flow for command selection, and exit code reference:
-
-- When to use OCR vs. a multimodal model
-- Detect-then-recognize two-step pattern for large images
-- ROI field extraction for receipts and forms
-- Verifying multimodal output against deterministic OCR
-
 ## What you get
 
 - **Local processing.** Images, PDFs, and OCR results stay on your machine.
@@ -158,7 +167,24 @@ The npm package provides the following six builds. The default `createEngine()`
 | Windows x64 | WebGPU through D3D12, then CPU |
 | Windows arm64 | CPU |
 
-Applications that need explicit control can choose `auto`, `cpu`, `apple`, or `webgpu` through the [`execution` option](bindings/node/README.md#with-options).
+Applications that need explicit control can choose `auto`, `cpu`, `apple`, or
+`webgpu` through the
+[`execution` option](packages/light-ocr/README.md#platform-acceleration).
+
+## Model tiers
+
+Small remains the stable default. Tiny and Medium are opt-in preview packages
+under the `next` tag. All three expose the same API, types, result schema, and
+error model, while each install contains only its selected model.
+
+| Tier | Package / command | Model payload | Status |
+| --- | --- | ---: | --- |
+| Small | `@arcships/light-ocr` / `light-ocr` | ~30 MB | stable default |
+| Tiny | `@arcships/light-ocr-tiny@next` / `light-ocr-tiny` | ~6.3 MB | preview; 49 languages, no Japanese |
+| Medium | `@arcships/light-ocr-medium@next` / `light-ocr-medium` | ~139 MB | preview; quality-first |
+
+Tiny and Medium stay on `next`; they do not change what
+`npm install @arcships/light-ocr` installs.
 
 ## Measured performance
 
@@ -178,10 +204,22 @@ These are same-machine comparisons with the CPU path, and results vary by worklo
 
 C++ projects build the static library from source and link the `light_ocr::core` CMake target. The API accepts decoded `GRAY8`, `RGB8`, `BGR8`, or `RGBA8` pixels; start with the [C++ API guide](docs/native-api.md) and [build instructions](docs/build-and-release.md).
 
+## Agent Skill
+
+An [Agent Skill](.agents/skills/local-ocr/SKILL.md) is included for AI agents
+that can call local commands. It provides scenario-driven workflows, command
+selection guidance, and exit code reference:
+
+- When to use OCR vs. a multimodal model
+- Detect-then-recognize workflows for large images
+- ROI field extraction for receipts and forms
+- Verifying multimodal output against deterministic OCR
+
 ## Documentation
 
+- [npm package README](packages/light-ocr/README.md)
 - [CLI reference](docs/cli-design.md)
-- [Node.js API and examples](bindings/node/README.md)
+- [Node.js image engine and CLI details](bindings/node/README.md)
 - [Agent Skill](.agents/skills/local-ocr/SKILL.md)
 - [C++ API](docs/native-api.md)
 - [Apple Silicon acceleration](docs/apple-device-acceleration.md)
@@ -191,9 +229,9 @@ C++ projects build the static library from source and link the `light_ocr::core`
 - [Build and release](docs/build-and-release.md)
 - [Roadmap](docs/roadmap.md)
 - [Changelog](CHANGELOG.md)
+- [npm 0.5.6 release preparation — English](docs/releases/npm-0.5.6.en.md)
+- [npm 0.5.6 发布准备记录 — 中文](docs/releases/npm-0.5.6.md)
 - [npm 0.3.0 release report](docs/releases/npm-0.3.0.md)
-- [npm 0.4.0 N2 candidate record](docs/releases/npm-0.4.0.md)
-- [npm 0.5.0 N3 release record](docs/releases/npm-0.5.0.md)
 
 ## Community and license
 
diff --git a/README.zh-CN.md b/README.zh-CN.md
index fc0fa48..4d5ff53 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -12,7 +12,17 @@
 
 **面向 Node.js 与 C++ 的快速离线 OCR。**
 
-直接在本机识别 PDF、JPEG、PNG 或像素数据，返回按阅读顺序排列的文字、置信度和四边形坐标。Node.js 用户安装的 npm 包内置 PP-OCRv6 Small 模型、PDFium，以及 macOS、Linux 和 Windows 的预编译组件。
+直接在本机识别 PDF、JPEG、PNG 或像素数据，返回按阅读顺序排列的
+文字、置信度和四边形坐标。Node.js 用户安装的 npm 包内置 PP-OCRv6
+Small 模型、PDFium、中文 fallback 字体，以及 macOS、Linux 和 Windows
+的预编译组件；安装后或首次运行时无需二次下载。
+
+| | |
+| --- | --- |
+| **适合场景** | Node.js 应用、CLI、桌面软件与原生 C++ 集成中的本地 OCR |
+| **输入** | JPEG、PNG、PDF、编码后的字节或解码后的像素 |
+| **输出** | 文字、置信度、四边形坐标、页面元数据与耗时 |
+| **分发方式** | 一次 npm 安装；同时提供 CommonJS、ESM、TypeScript 与六个平台预编译包 |
 
 ## 快速开始
 
@@ -27,30 +37,35 @@ import { createEngine } from "@arcships/light-ocr";
 import { readFile } from "node:fs/promises";
 
 const engine = await createEngine();
-const result = await engine.recognizeEncoded(
-  await readFile("image.jpg"),
-);
 
-for (const line of result.lines) {
-  console.log(line.text, line.confidence, line.box);
-}
+try {
+  const result = await engine.recognizeEncoded(
+    await readFile("image.jpg"),
+  );
 
-await engine.close();
+  for (const line of result.lines) {
+    console.log(line.text, line.confidence, line.box);
+  }
+} finally {
+  await engine.close();
+}
 ```
 
-`createEngine()` 会根据当前平台自动选择合适的执行方式。如果应用已经完成图片解码，[`recognize()`](bindings/node/README.md#使用) 也可以直接接收 `GRAY8`、`RGB8`、`BGR8` 和 `RGBA8` 像素数据。
+`createEngine()` 会根据当前平台自动选择合适的执行方式。如果应用已经
+完成图片解码，[`recognize()`](packages/light-ocr/README.md#recognize-decoded-pixels)
+也可以直接接收 `GRAY8`、`RGB8`、`BGR8` 和 `RGBA8` 像素数据。
 
-### 模型杯型
+PDF 页面使用同一个包：
 
-Small 继续作为稳定默认。N2 在 `next` tag 下增加两个可选 preview 包；三档使用完全相同的 API、类型、结果 schema 和错误模型，每次安装只携带所选的一份模型。
+```ts
+import { recognizeDocument } from "@arcships/light-ocr";
 
-| 杯型 | Package / 命令 | 模型 payload | 状态 |
-| --- | --- | ---: | --- |
-| Small | `@arcships/light-ocr` / `light-ocr` | 约 30 MB | 稳定默认 |
-| Tiny | `@arcships/light-ocr-tiny@next` / `light-ocr-tiny` | 约 6.3 MB | Preview；49 种语言，不含日语 |
-| Medium | `@arcships/light-ocr-medium@next` / `light-ocr-medium` | 约 139 MB | Preview；精度优先 |
+for await (const page of recognizeDocument("report.pdf", { dpi: 200 })) {
+  console.log(page.index, page.lines.map((line) => line.text));
+}
+```
 
-Tiny/Medium 只有在真实使用证明存在明确受众后才会提升 stable；它们不会改变 `npm install @arcships/light-ocr` 的安装内容。
+CommonJS 可以通过 `require("@arcships/light-ocr")` 使用同一组导出。
 
 ## CLI
 
@@ -83,9 +98,9 @@ light-ocr doctor --json
 `doctor`（系统诊断）。传入 `.pdf` 路径会直接进入文档 OCR；显式
 `document` 命令用于多输入任务。输出遵循版本化的 `schemaVersion: 1`
 契约，EXIF 方向自动修正。完整参考见 [CLI 设计](docs/cli-design.md) 和
-[npm README](bindings/node/README.md#cli)。
+[npm package README](packages/light-ocr/README.md)。
 
-### PDF 和多页文档
+## PDF 和多页文档
 
 PDF 和多页 OCR 已直接内置于 `@arcships/light-ocr`。匹配当前平台的
 PDFium 二进制、校验锁定的 Noto Sans SC fallback 字体与 OCR
@@ -120,15 +135,6 @@ for await (const page of recognizeDocument([buf1, buf2, buf3])) {
 }
 ```
 
-## Agent Skill
-
-内置 [Agent Skill](.agents/skills/local-ocr/SKILL.md)，供可调用本地命令的 AI Agent 使用。提供场景驱动的工作流、命令选择决策流和退出码参考：
-
-- 何时使用 OCR 而非多模态模型
-- 大图先 detect 再 recognize 的两步模式
-- 票据/表单的 ROI 字段提取
-- 用确定性 OCR 验证多模态模型输出
-
 ## 主要能力
 
 - **本地处理。**图片、PDF 和 OCR 结果始终留在本机。
@@ -154,7 +160,24 @@ npm 包提供以下六个平台版本。默认的 `createEngine()` 使用 Auto 
 | Windows x64 | 通过 D3D12 使用 WebGPU，然后使用 CPU |
 | Windows arm64 | CPU |
 
-需要明确控制时，可以通过 [`execution` 选项](bindings/node/README.md#使用)选择 `auto`、`cpu`、`apple` 或 `webgpu`。
+需要明确控制时，可以通过
+[`execution` 选项](packages/light-ocr/README.md#platform-acceleration)
+选择 `auto`、`cpu`、`apple` 或 `webgpu`。
+
+## 模型档位
+
+Small 继续作为稳定默认。Tiny 和 Medium 是 `next` tag 下的可选 preview
+包。三档使用完全相同的 API、类型、结果 schema 和错误模型，每次安装
+只携带所选的一份模型。
+
+| 档位 | Package / 命令 | 模型 payload | 状态 |
+| --- | --- | ---: | --- |
+| Small | `@arcships/light-ocr` / `light-ocr` | 约 30 MB | 稳定默认 |
+| Tiny | `@arcships/light-ocr-tiny@next` / `light-ocr-tiny` | 约 6.3 MB | Preview；49 种语言，不含日语 |
+| Medium | `@arcships/light-ocr-medium@next` / `light-ocr-medium` | 约 139 MB | Preview；精度优先 |
+
+Tiny/Medium 保持在 `next`；它们不会改变
+`npm install @arcships/light-ocr` 的安装内容。
 
 ## 实测性能
 
@@ -174,10 +197,21 @@ npm 包提供以下六个平台版本。默认的 `createEngine()` 使用 Auto 
 
 C++ 项目从源码构建静态库，并链接 `light_ocr::core` CMake target。API 可以直接识别解码后的 `GRAY8`、`RGB8`、`BGR8` 或 `RGBA8` 像素；接入方式见 [C++ API](docs/native-api.md)，各平台准备方式见 [构建说明](docs/build-and-release.md)。
 
+## Agent Skill
+
+内置 [Agent Skill](.agents/skills/local-ocr/SKILL.md)，供可调用本地命令的
+AI Agent 使用。它提供场景驱动的工作流、命令选择建议和退出码参考：
+
+- 何时使用 OCR 而非多模态模型
+- 大图先 detect 再 recognize 的工作流
+- 票据/表单的 ROI 字段提取
+- 用确定性 OCR 验证多模态模型输出
+
 ## 文档
 
+- [npm package README](packages/light-ocr/README.md)
 - [CLI 参考](docs/cli-design.md)
-- [Node.js API 与示例](bindings/node/README.md)
+- [Node.js 图片引擎与 CLI 细节](bindings/node/README.md)
 - [Agent Skill](.agents/skills/local-ocr/SKILL.md)
 - [C++ API](docs/native-api.md)
 - [Apple Silicon 加速](docs/apple-device-acceleration.md)
@@ -187,9 +221,9 @@ C++ 项目从源码构建静态库，并链接 `light_ocr::core` CMake target。
 - [构建与发布](docs/build-and-release.md)
 - [路线图](docs/roadmap.md)
 - [更新日志](CHANGELOG.md)
+- [npm 0.5.6 release preparation — English](docs/releases/npm-0.5.6.en.md)
+- [npm 0.5.6 发布准备记录 — 中文](docs/releases/npm-0.5.6.md)
 - [npm 0.3.0 发布报告](docs/releases/npm-0.3.0.md)
-- [npm 0.4.0 N2 候选记录](docs/releases/npm-0.4.0.md)
-- [npm 0.5.0 N3 发布记录](docs/releases/npm-0.5.0.md)
 
 ## 社区与协议
 
diff --git a/bindings/node/README.md b/bindings/node/README.md
index df67395..6a56169 100644
--- a/bindings/node/README.md
+++ b/bindings/node/README.md
@@ -2,6 +2,11 @@
 
 Local OCR for Node.js and Agents. Offline, no Python, no network calls. Returns text with coordinates and confidence.
 
+> Installing from npm? Start with the
+> [`@arcships/light-ocr` package guide](../../packages/light-ocr/README.md).
+> This document keeps the lower-level image engine, CLI contract, and
+> source-build details.
+
 ## Install
 
 ```bash
@@ -23,6 +28,9 @@ light-ocr image.png --format json
 # Just text, no coordinates
 light-ocr image.png --format text
 
+# PDF pages through the renderer bundled by the main package
+light-ocr report.pdf --pages 1-10 --format jsonl
+
 # Engine info
 light-ocr info --version
 light-ocr info --model-info
@@ -170,7 +178,8 @@ TypeScript types are included in the shared runtime [`index.d.ts`](../../package
 
 ## Capabilities
 
-- **Formats**: JPEG, PNG (memory input via `Uint8Array`)
+- **Formats**: JPEG and PNG through the image engine; PDF and multi-page image
+  jobs through the main package's document API
 - **EXIF orientation**: JPEG orientation tags 1–8 automatically corrected
 - **ROI**: pageSpace axis-aligned rectangle, coordinates offset back to full page
 - **Providers**: CPU (all platforms), Apple Core ML (macOS arm64), WebGPU (Linux x64, Windows x64)
@@ -181,7 +190,7 @@ TypeScript types are included in the shared runtime [`index.d.ts`](../../package
 
 ### Not supported
 
-- WebP, GIF, PDF, TIFF
+- WebP, GIF, TIFF
 - Character-level coordinates (recognition is line-level)
 - Multi-engine fan-out as default architecture
 - Bun (Node-API lifecycle not fully verified)
diff --git a/docs/implementation-status.md b/docs/implementation-status.md
index a4796a2..06bc760 100644
--- a/docs/implementation-status.md
+++ b/docs/implementation-status.md
@@ -1,7 +1,7 @@
 # C++ Core 与 Node-API 实施状态
 
-更新时间：2026-07-28<br>
-结论：npm `0.5.5` 已发布，但其平台包只内置 PDFium addon 与共享库，未内置 fallback 字体，因此对非嵌入中文字体的 PDF 不能视为可靠。`0.5.6` 补丁候选现已把校验锁定的官方 Noto Sans SC 区域子集及 OFL 许可证装入六个平台包，并在 patched PDFium 初始化时只使用包内字体目录；用户安装与运行仍无二次下载。当前本机 macOS arm64 已验证 `STSong-Light` 非嵌入字体映射、PNG 渲染与 `中文测试` 端到端 OCR，六平台 release smoke 尚待 CI 实跑。D116 记录修复契约，已发布 `0.5.5` 的原始证据仍见 [npm 0.5.5 发布记录](releases/npm-0.5.5.md)。
+更新时间：2026-07-30<br>
+结论：npm `0.5.5` 已发布，但其平台包只内置 PDFium addon 与共享库，未内置 fallback 字体，因此对非嵌入中文字体的 PDF 不能视为可靠。`0.5.6` 补丁候选已把校验锁定的官方 Noto Sans SC 区域子集及 OFL 许可证装入六个平台包，并在 patched PDFium 初始化时只使用包内字体目录；用户安装与运行仍无二次下载。合并后的 Core、sanitizer/fuzzer、oracle 和 WebGPU CI 已全绿，[npm 发布演练 30535822947](https://github.com/arcships/light-ocr/actions/runs/30535822947) 也已完成六平台构建、离线安装、图片 OCR 与非嵌入中文字体 PDF 端到端 smoke；该次演练明确跳过 registry 发布。D116 记录修复契约，候选闭包与后续发布步骤见 [npm 0.5.6 发布准备记录](releases/npm-0.5.6.md)（[English](releases/npm-0.5.6.en.md)）。
 
 状态含义：
 
@@ -27,7 +27,7 @@
 
 ## PDF fallback 字体修复（0.5.6 候选）
 
-状态：本机工程验证完成 / 六平台 release CI 待运行
+状态：工程验证与六平台 release CI 完成 / npm registry 发布待执行
 
 - Noto Sans SC 官方区域子集和 OFL 文本以固定 revision、bytes 与 SHA-256
   锁定；release staging 对缺失、篡改、重复或不完整 inventory
@@ -40,7 +40,10 @@
   `fontFamily = Noto Sans SC`、`isEmbedded = false`、有效 PNG，并由
   默认 OCR 模型识别出 `中文测试`。
 - 六个平台的离线 tarball smoke 同时检查字体选择、渲染结果和 OCR
-  文本；在该矩阵全部通过前，不能把本机结论写成跨平台发布完成。
+  文本；[run 30535822947](https://github.com/arcships/light-ocr/actions/runs/30535822947)
+  已全部通过，且 `publish` job 因演练参数保持 skipped。
+- 相对 `0.5.5`，每个用户实际安装的平台 native tarball 增加
+  `7,226,822–7,232,623` bytes，解包增加约 `8.34 MB`；模型包不变。
 
 ## 需求验收矩阵
 
diff --git a/docs/releases/npm-0.5.6.en.md b/docs/releases/npm-0.5.6.en.md
new file mode 100644
index 0000000..d525529
--- /dev/null
+++ b/docs/releases/npm-0.5.6.en.md
@@ -0,0 +1,145 @@
+# npm 0.5.6 Release Preparation Record
+
+[中文版](npm-0.5.6.md)
+
+Status: pre-release validation is complete, but the release has not been
+published. The npm registry and all dist-tags remain unchanged from the
+`0.5.5` release.
+
+Release identity:
+
+- PDF fallback implementation:
+  [`196ebde`](https://github.com/arcships/light-ocr/commit/196ebdee7c3adead047b485e6a558133bd0263c5)
+- Windows release builder fix:
+  [`d80a868`](https://github.com/arcships/light-ocr/commit/d80a8681485c470b48699e24fa18661d6a60b718)
+- Post-merge baselines:
+  [Core 30534418143](https://github.com/arcships/light-ocr/actions/runs/30534418143) and
+  [Native WebGPU 30534418001](https://github.com/arcships/light-ocr/actions/runs/30534418001)
+- Initial diagnostic rehearsal:
+  [30535427156](https://github.com/arcships/light-ocr/actions/runs/30535427156)
+  (macOS and Linux passed; Windows exposed the `npx.cmd` resolution issue;
+  assembly and publication did not run)
+- Complete release rehearsal:
+  [30535822947](https://github.com/arcships/light-ocr/actions/runs/30535822947)
+  (all six platforms passed with `publish_to_registry=false`)
+
+## User-visible changes
+
+- Fix PDF rasterization for common documents that reference non-embedded
+  Chinese fonts. PDFium now renders those pages with the Noto Sans SC fallback
+  font carried by the current native npm package before passing the resulting
+  pixels to OCR.
+- Ship the font, its OFL license, the PDFium addon, and the matching shared
+  library in each platform package. The supported installation works with
+  `npm install --ignore-scripts`; customer installation and runtime perform no
+  font download or native compilation and do not require a system Chinese
+  font.
+- Pin `NotoSansSC-Regular.otf` at `8,331,336` bytes with SHA-256
+  `faa6c9df652116dde789d351359f3d7e5d2285a2b2a1f04a2d7244df706d5ea9`.
+- Preserve the existing public API and schema. The image, PDF, and multi-page
+  entry points in `@arcships/light-ocr` are unchanged.
+- Keep the immutable `0.5.5` packages available, but do not recommend them for
+  workloads that require reliable rendering of PDFs with non-embedded Chinese
+  fonts.
+
+## Candidate version closure
+
+| Maturity | Package | Candidate version | Planned tags after release |
+| --- | --- | ---: | --- |
+| stable | `@arcships/light-ocr` | `0.5.6` | publish to `next`, then promote to `latest` |
+| stable | `@arcships/light-ocr-runtime` | `0.1.6` | publish to `next`, then promote to `latest` |
+| stable | six native platform packages | `0.5.6` | publish to `next`, then promote to `latest` |
+| compatibility | `@arcships/light-ocr-document` | `0.1.2` | remain on `next` |
+| preview | `@arcships/light-ocr-tiny` | `0.1.5` | remain on `next` |
+| preview | `@arcships/light-ocr-medium` | `0.1.5` | remain on `next` |
+
+No model receives a new version. Small continues to use `0.3.4`, while Tiny
+and Medium continue to use `0.1.0`. The rehearsal rebuilds or retrieves model
+tarballs for offline installation testing. During publication, an immutable
+model package is skipped only when the same version already exists in the
+registry with identical integrity.
+
+## Six-platform native candidates
+
+The following values come from the complete rehearsal's
+`release-manifest.json`. Each delta uses the published `0.5.5` tarball as its
+baseline. A customer installs only the one native package selected for the
+current platform.
+
+| Platform package | Compressed bytes | Delta from 0.5.5 | Unpacked bytes | SHA-256 |
+| --- | ---: | ---: | ---: | --- |
+| `darwin-arm64` | 22,889,542 | +7,227,406 | 56,356,978 | `ae325aaf9a3ff79cd184a862ea36687fa32227bf5a048136f02591626ce16342` |
+| `darwin-x64` | 24,853,681 | +7,229,542 | 62,109,821 | `4f3d2f9cb6fba776181e327e7321787d91b516c0a39497d8ebf2ad56bf6112cc` |
+| `linux-arm64-gnu` | 20,453,485 | +7,227,462 | 40,261,084 | `6dfbedc1637ff17aa13117925ca08f80b07367d62216f90a05b7614bff038078` |
+| `linux-x64-gnu` | 27,033,591 | +7,226,822 | 59,390,897 | `eb67edb08db86998c8b232c42df90eda47c0b1a30311f78e1f7f5023f0c22a15` |
+| `win32-arm64` | 16,663,753 | +7,232,623 | 30,985,476 | `0ad4e90cadd230405c244a309f9dbaac33b371ae9e6a4868d7eaf31c5962723d` |
+| `win32-x64` | 30,387,143 | +7,232,239 | 63,070,260 | `bcbd5b92526f0a640db53154f6482ee0c58a4a9d3c559c5ecac0611f2b21a2dd` |
+
+## Release gates
+
+- [x] Post-merge six-platform Core, sanitizers, fuzzer, and oracle
+- [x] Linux and Windows Native WebGPU contracts
+- [x] npm registry availability for all 11 new package identities
+- [x] PDFium addon rebuilt from the pinned source on all six platforms
+- [x] Fallback font and OFL license present in all six native packages
+- [x] Complete closure installed on all six platforms in offline mode with
+      installation scripts disabled
+- [x] Real image OCR on all six platforms
+- [x] Non-embedded `STSong-Light` mapped to `Noto Sans SC` on all six platforms
+- [x] Non-empty PDF raster and end-to-end `中文测试` OCR on all six platforms
+- [x] Candidate tarball manifest, byte counts, SHA-256 values, and npm integrity
+- [ ] Publish the immutable candidate to `next` with
+      `publish_to_registry=true`
+- [ ] Reinstall from the npm registry and verify integrity
+- [ ] Promote the stable Small/runtime/native closure to `latest`
+- [ ] Create the `v0.5.6` GitHub Release
+
+## Publication sequence
+
+1. Run `npm release` again from `main` with version `0.5.6` and
+   `publish_to_registry=true`. The workflow repeats the same build and smoke
+   tests before publishing only to `next`.
+2. Verify registry integrity and reinstall the complete closure.
+3. Run `npm promote` to move only Small `0.5.6`, runtime `0.1.6`, and the six
+   native `0.5.6` packages to `latest`.
+4. Create the `v0.5.6` tag and GitHub Release, then update this record with the
+   final status and workflow IDs.
+
+Tiny, Medium, and Document are not part of stable promotion and remain on
+`next`. No external user validation or adoption evidence blocks this release.
+Every blocking gate is satisfied by the source, candidate tarballs, real
+six-platform runners, and npm registry identity.
+
+## GitHub Release draft
+
+Release name:
+
+`墨字归真，六境同明 · Restore Chinese PDF rendering in 0.5.6`
+
+Highlights:
+
+- Fix PDF rasterization when documents reference common non-embedded Chinese
+  fonts.
+- Bundle a checksum-pinned Noto Sans SC fallback font and its OFL license in
+  every supported native package.
+- Keep installation and runtime fully self-contained: no postinstall download,
+  system-font requirement, GitHub access, or local compilation.
+- Preserve the existing image, PDF, and multi-page APIs without schema changes.
+- Add about 7.23 MB compressed and 8.34 MB unpacked to the one native package
+  selected for the customer's platform.
+
+Versions:
+
+- `@arcships/light-ocr@0.5.6`
+- `@arcships/light-ocr-runtime@0.1.6`
+- six native platform packages at `0.5.6`
+- Document compatibility facade `0.1.2` under `next`
+- Tiny/Medium preview facades `0.1.5` under `next`
+- unchanged Small `0.3.4` and Tiny/Medium `0.1.0` model packages
+
+Verification:
+
+- pre-release dry-run:
+  <https://github.com/arcships/light-ocr/actions/runs/30535822947>
+- release workflow: add after publication
+- stable promotion: add after promotion
diff --git a/docs/releases/npm-0.5.6.md b/docs/releases/npm-0.5.6.md
new file mode 100644
index 0000000..96ab542
--- /dev/null
+++ b/docs/releases/npm-0.5.6.md
@@ -0,0 +1,132 @@
+# npm 0.5.6 发布准备记录
+
+[English version](npm-0.5.6.en.md)
+
+状态：发布前验证完成，尚未发布。npm registry 与 dist-tag 均保持
+`0.5.5` 发布后的状态。
+
+发布身份：
+
+- PDF fallback 实现提交：
+  [`196ebde`](https://github.com/arcships/light-ocr/commit/196ebdee7c3adead047b485e6a558133bd0263c5)
+- Windows release builder 修复：
+  [`d80a868`](https://github.com/arcships/light-ocr/commit/d80a8681485c470b48699e24fa18661d6a60b718)
+- 合并后基线：
+  [Core 30534418143](https://github.com/arcships/light-ocr/actions/runs/30534418143)、
+  [Native WebGPU 30534418001](https://github.com/arcships/light-ocr/actions/runs/30534418001)
+- 首次诊断演练：
+  [30535427156](https://github.com/arcships/light-ocr/actions/runs/30535427156)
+  （macOS/Linux 通过，Windows 暴露 `npx.cmd` 寻址问题，未组装、未发布）
+- 完整发布演练：
+  [30535822947](https://github.com/arcships/light-ocr/actions/runs/30535822947)
+  （六平台通过，`publish_to_registry=false`）
+
+## 用户可见变化
+
+- 修复常见非嵌入中文字体 PDF 的渲染前置缺陷。PDFium 现在从当前
+  native npm 包内的 Noto Sans SC fallback 字体绘制页面，再把正确的
+  像素交给 OCR。
+- 字体、OFL 许可证、PDFium addon 与共享库全部随匹配平台的 npm 包
+  交付。支持 `npm install --ignore-scripts`；客户机安装和运行均不下载
+  字体、不编译 native addon，也不要求系统预装中文字体。
+- 锁定的 `NotoSansSC-Regular.otf` 为 `8,331,336` bytes，SHA-256 为
+  `faa6c9df652116dde789d351359f3d7e5d2285a2b2a1f04a2d7244df706d5ea9`。
+- 没有 public API 或 schema 破坏。`@arcships/light-ocr` 的图片、PDF
+  与多页图片入口保持不变。
+- `0.5.5` 的不可变包继续保留，但不应再用于需要可靠处理非嵌入中文
+  字体 PDF 的场景。
+
+## 待发布版本闭包
+
+| 成熟度 | 包 | 候选版本 | 发布后标签计划 |
+| --- | --- | ---: | --- |
+| stable | `@arcships/light-ocr` | `0.5.6` | `next` 验证后晋升 `latest` |
+| stable | `@arcships/light-ocr-runtime` | `0.1.6` | `next` 验证后晋升 `latest` |
+| stable | 六个平台 native | `0.5.6` | `next` 验证后晋升 `latest` |
+| compatibility | `@arcships/light-ocr-document` | `0.1.2` | 保持 `next` |
+| preview | `@arcships/light-ocr-tiny` | `0.1.5` | 保持 `next` |
+| preview | `@arcships/light-ocr-medium` | `0.1.5` | 保持 `next` |
+
+模型包不重发新版本：Small 继续使用 `0.3.4`，Tiny/Medium 继续使用
+`0.1.0`。演练会重建或取得模型 tarball 来完成离线安装验证；发布器只在
+registry 中的同版本 integrity 完全一致时跳过这些不可变包。
+
+## 六平台 native 候选
+
+下表来自完整演练的 `release-manifest.json`。增量以已发布 `0.5.5`
+tarball 为基线；用户只安装当前平台对应的一个 native 包。
+
+| 平台包 | 压缩 bytes | 相对 0.5.5 | 解包 bytes | SHA-256 |
+| --- | ---: | ---: | ---: | --- |
+| `darwin-arm64` | 22,889,542 | +7,227,406 | 56,356,978 | `ae325aaf9a3ff79cd184a862ea36687fa32227bf5a048136f02591626ce16342` |
+| `darwin-x64` | 24,853,681 | +7,229,542 | 62,109,821 | `4f3d2f9cb6fba776181e327e7321787d91b516c0a39497d8ebf2ad56bf6112cc` |
+| `linux-arm64-gnu` | 20,453,485 | +7,227,462 | 40,261,084 | `6dfbedc1637ff17aa13117925ca08f80b07367d62216f90a05b7614bff038078` |
+| `linux-x64-gnu` | 27,033,591 | +7,226,822 | 59,390,897 | `eb67edb08db86998c8b232c42df90eda47c0b1a30311f78e1f7f5023f0c22a15` |
+| `win32-arm64` | 16,663,753 | +7,232,623 | 30,985,476 | `0ad4e90cadd230405c244a309f9dbaac33b371ae9e6a4868d7eaf31c5962723d` |
+| `win32-x64` | 30,387,143 | +7,232,239 | 63,070,260 | `bcbd5b92526f0a640db53154f6482ee0c58a4a9d3c559c5ecac0611f2b21a2dd` |
+
+## 已通过的发布门
+
+- [x] 合并后的六平台 Core、sanitizer、fuzzer 与 oracle
+- [x] Linux/Windows Native WebGPU contract
+- [x] 11 个新 package identity 的 npm registry 空位检查
+- [x] 六个平台从锁定源码重建 PDFium addon
+- [x] 六个平台 native 包包含 fallback 字体与 OFL 许可证
+- [x] 六个平台在离线模式且禁用安装脚本的条件下安装完整闭包
+- [x] 六个平台运行真实图片 OCR
+- [x] 六个平台验证非嵌入 `STSong-Light` 映射到 `Noto Sans SC`
+- [x] 六个平台验证非空 PDF raster 与端到端 `中文测试` OCR
+- [x] 候选 tarball manifest、bytes、SHA-256 与 npm integrity
+- [ ] 以 `publish_to_registry=true` 发布不可变候选到 `next`
+- [ ] 从 npm registry 回装并核对 integrity
+- [ ] 将 stable Small/runtime/native 闭包晋升到 `latest`
+- [ ] 创建 `v0.5.6` GitHub Release
+
+## 实际发布顺序
+
+1. 从 `main` 重新运行 `npm release`，版本 `0.5.6`，
+   `publish_to_registry=true`；流水线先重复相同构建与 smoke，再只向
+   `next` 写入候选。
+2. 核对 registry integrity 与完整闭包回装结果。
+3. 运行 `npm promote`，只把 Small `0.5.6`、runtime `0.1.6` 与六个
+   native `0.5.6` 晋升到 `latest`。
+4. 创建 `v0.5.6` tag/GitHub Release，并把本记录状态和最终 run ID
+   更新为已发布。
+
+Tiny、Medium 与 Document 不属于 stable promotion；它们继续停留在
+`next`。整个发布不依赖外部用户验证或采用证明，所有阻断门均由源码、
+候选 tarball、真实六平台 runner 和 npm registry identity 自证完成。
+
+## GitHub Release 文案草案
+
+Release name：
+
+`墨字归真，六境同明 · Restore Chinese PDF rendering in 0.5.6`
+
+Highlights：
+
+- Fix PDF rasterization when documents reference common non-embedded Chinese
+  fonts.
+- Bundle a checksum-pinned Noto Sans SC fallback font and its OFL license in
+  every supported native package.
+- Keep installation and runtime fully self-contained: no postinstall download,
+  system-font requirement, GitHub access, or local compilation.
+- Preserve the existing image, PDF, and multi-page APIs without schema changes.
+- Add about 7.23 MB compressed and 8.34 MB unpacked to the one native package
+  selected for the customer's platform.
+
+Versions：
+
+- `@arcships/light-ocr@0.5.6`
+- `@arcships/light-ocr-runtime@0.1.6`
+- six native platform packages at `0.5.6`
+- Document compatibility facade `0.1.2` under `next`
+- Tiny/Medium preview facades `0.1.5` under `next`
+- unchanged Small `0.3.4` and Tiny/Medium `0.1.0` model packages
+
+Verification：
+
+- pre-release dry-run:
+  <https://github.com/arcships/light-ocr/actions/runs/30535822947>
+- release workflow: 待实际发布后补入
+- stable promotion: 待晋升后补入
diff --git a/packages/light-ocr/README.md b/packages/light-ocr/README.md
index 4ca65fc..890f139 100644
--- a/packages/light-ocr/README.md
+++ b/packages/light-ocr/README.md
@@ -1,32 +1,299 @@
 # @arcships/light-ocr
 
-The stable PP-OCRv6 Small entry for local image and PDF OCR on Node.js 22 and 24.
+[![npm version](https://img.shields.io/npm/v/%40arcships%2Flight-ocr?color=CB3837)](https://www.npmjs.com/package/@arcships/light-ocr)
+[![Node.js 22 and 24](https://img.shields.io/badge/Node.js-22%20%7C%2024-339933)](https://nodejs.org/)
+[![Apache-2.0](https://img.shields.io/npm/l/%40arcships%2Flight-ocr)](https://github.com/arcships/light-ocr/blob/main/LICENSE)
+
+![light-ocr](https://raw.githubusercontent.com/arcships/light-ocr/main/docs/assets/light-ocr-banner.png)
+
+**Offline image and PDF OCR for Node.js — prebuilt, typed, and
+self-contained.**
+
+`@arcships/light-ocr` is the stable PP-OCRv6 Small package. It recognizes
+JPEG, PNG, PDF, decoded pixels, and multi-page image jobs without Python,
+cloud APIs, postinstall downloads, or local native compilation.
+
+| | |
+| --- | --- |
+| **Input** | JPEG, PNG, PDF, `Uint8Array`, or decoded pixel buffers |
+| **Output** | text lines in reading order, confidence, quadrilateral boxes, and timing |
+| **Runtime** | CommonJS + ESM + bundled TypeScript declarations |
+| **Platforms** | macOS, Linux glibc, and Windows on x64 and ARM64 |
+| **Node.js** | 22 and 24 |
+
+## Install
 
 ```bash
 npm install @arcships/light-ocr
+```
+
+That single command installs the Small model, shared JavaScript runtime, and
+the native OCR/PDF package matching the current platform. The package has no
+install script and remains installable with scripts disabled.
+
+## Quick start
+
+### Recognize an image
+
+```js
+import { createEngine } from "@arcships/light-ocr";
+import { readFile } from "node:fs/promises";
+
+const engine = await createEngine();
+
+try {
+  const result = await engine.recognizeEncoded(
+    await readFile("receipt.png"),
+  );
+
+  for (const line of result.lines) {
+    console.log(line.text, line.confidence, line.box);
+  }
+} finally {
+  await engine.close();
+}
+```
+
+CommonJS uses the same API:
+
+```js
+const { createEngine } = require("@arcships/light-ocr");
+```
+
+### Stream pages from a PDF
+
+```js
+import { recognizeDocument } from "@arcships/light-ocr";
+
+for await (const page of recognizeDocument("report.pdf", {
+  dpi: 200,
+  pageRange: { start: 1, end: 10 },
+})) {
+  console.log(`page ${page.index + 1}`);
+  for (const line of page.lines) console.log(line.text);
+}
+```
+
+`recognizeDocument()` accepts a file path, PDF/image bytes, or an array of
+image paths and byte buffers. Pages are yielded as they finish, so callers do
+not need to retain the complete document result.
+
+### Use the CLI
+
+The `light-ocr` command is included:
+
+```bash
+# Image OCR
 light-ocr image.png --format text
+
+# PDF OCR; PDFium and the fallback font are already installed
 light-ocr report.pdf --pages 1-10 --format jsonl
+
+# Multiple images as one document
+light-ocr document scan-1.png scan-2.png scan-3.png --format text
+
+# Detection only
+light-ocr detect screenshot.png
+
+# Voluntary hardware/provider diagnostics
 light-ocr doctor --json
 ```
 
-The package exact-pins one model-free runtime, the Small model, and the native
-component, PDFium renderer, and Noto Sans SC fallback font for the current
-platform. It has no install script and its complete release closure is tested
-with npm offline and scripts disabled.
+## Everything needed for PDF OCR is included
+
+The package does not defer essential files to an installer or first-run
+download:
+
+| Component | Distribution |
+| --- | --- |
+| PP-OCRv6 Small model | exact-pinned required npm dependency |
+| JavaScript runtime and types | exact-pinned required npm dependency |
+| Native OCR addon and runtime libraries | matching platform npm dependency |
+| PDFium addon and shared library | inside the matching platform package |
+| Noto Sans SC fallback font and OFL license | inside the matching platform package |
+
+The bundled, checksum-pinned fallback font is used when a PDF references a
+common Chinese font without embedding its glyphs. PDFium therefore renders the
+page correctly before OCR instead of handing missing-glyph boxes to the model.
+
+There is no postinstall fetch, runtime font/model/PDFium download, GitHub
+access, compiler requirement, or system Chinese-font requirement on customer
+machines.
+
+## Node.js API
+
+### Main exports
+
+| Export | Purpose |
+| --- | --- |
+| `createEngine(options?)` | Create a reusable image OCR engine |
+| `recognizeDocument(source, options?)` | Stream PDF or image pages with automatic cleanup |
+| `createDocumentEngine(options?)` | Reuse one engine across multiple document jobs |
+| `hasPdfSupport()` | Check that the bundled PDF renderer can be loaded |
+| `modelProfile` | Inspect the selected model tier and language metadata |
+| `OcrError` | Stable typed error with a machine-readable `code` |
+
+### Recognize decoded pixels
+
+Applications that already decode images can avoid re-encoding:
+
+```js
+const result = await engine.recognize({
+  data: rgbaBytes,
+  width,
+  height,
+  stride: width * 4,
+  pixelFormat: "rgba8",
+});
+```
+
+Supported pixel formats are `gray8`, `rgb8`, `bgr8`, and `rgba8`.
+
+### Region, cancellation, and execution provider
 
 ```js
-const { createEngine } = require('@arcships/light-ocr');
+const controller = new AbortController();
+
+const engine = await createEngine({
+  execution: { provider: "auto" },
+  queueCapacity: 4,
+});
 
-const engine = await createEngine();
 try {
-  const result = await engine.recognizeEncoded(imageBytes);
+  const result = await engine.recognizeEncoded(imageBytes, {
+    region: { x: 100, y: 80, width: 640, height: 320 },
+    applyExif: true,
+    signal: controller.signal,
+  });
+
   console.log(result.lines);
+  console.log(engine.info.execution.selectionTrace);
 } finally {
   await engine.close();
 }
 ```
 
-The main package exports `recognizeDocument()` and `createDocumentEngine()`.
-PDFium's native files and the checksum-pinned fallback font are inside the
-platform npm package, so neither install nor runtime performs a secondary
-download or requires a system Chinese font.
+Recognition runs on a dedicated worker instead of blocking the JavaScript main
+thread. Queues are bounded, `AbortSignal` is supported, and engines must be
+closed explicitly when the application is finished with them.
+
+### Document resource limits
+
+PDF and multi-page calls apply conservative defaults before or during
+rendering:
+
+| Option | Default |
+| --- | ---: |
+| `dpi` | 150 |
+| `maxFileBytes` | 100 MiB per input |
+| `maxPages` | 100 |
+| `maxPagePixels` | 4096 × 4096 |
+| `maxTotalPixels` | 100 Mi pixels |
+
+Limits are configurable per call. Violations reject with
+`OcrError.code === "resource_limit_exceeded"`.
+
+### Result shape
+
+Image OCR returns one line entry per recognized line:
+
+```json
+{
+  "text": "TOTAL 42.00",
+  "confidence": 0.98,
+  "box": [
+    { "x": 24, "y": 80 },
+    { "x": 190, "y": 80 },
+    { "x": 190, "y": 108 },
+    { "x": 24, "y": 108 }
+  ]
+}
+```
+
+Coordinates use `pageSpace`: top-left origin, positive x to the right, and
+positive y downward after EXIF correction. Document pages add a stable page
+index, source metadata, dimensions, applied PDF transforms, and timing.
+
+## CLI reference
+
+| Command | Purpose |
+| --- | --- |
+| `light-ocr [recognize] <image>` | Full image OCR; `recognize` is optional |
+| `light-ocr <document.pdf>` | Stream a PDF through document OCR |
+| `light-ocr document <source...>` | Process a PDF or multiple page images |
+| `light-ocr detect <image>` | Return text-region boxes without recognition |
+| `light-ocr info --version` | Print package/Core version information |
+| `light-ocr info --model-info` | Print model and execution information |
+| `light-ocr doctor --json` | Print voluntary system/provider diagnostics |
+
+Image OCR supports `--format json|jsonl|text`, `--region x,y,w,h`,
+`--provider auto|cpu|apple|webgpu`, `--stdin`, and automatic EXIF correction.
+Document OCR adds `--pages N-M`, `--dpi`, and the page/file/pixel limit flags.
+
+`detect` always emits structured JSON and does not accept `--format`.
+Diagnostics do not include the hostname or a stable device identifier.
+
+## Platform acceleration
+
+The default `provider: "auto"` chooses a qualified accelerator when available
+and ends with CPU as the stable fallback:
+
+| Platform | Auto path |
+| --- | --- |
+| macOS 15+ on Apple Silicon | Core ML, then CPU |
+| macOS on Intel | CPU |
+| Linux x64 with glibc | WebGPU through Vulkan, then CPU |
+| Linux arm64 with glibc | CPU |
+| Windows x64 | WebGPU through D3D12, then CPU |
+| Windows arm64 | CPU |
+
+Explicit `cpu`, `apple`, and `webgpu` requests fail closed when the requested
+provider is unavailable; they do not silently switch providers.
+
+## Model tiers
+
+This package is the stable Small tier. Tiny and Medium are opt-in previews with
+the same API, types, result schema, and error model:
+
+| Tier | Install | Model payload | Status |
+| --- | --- | ---: | --- |
+| Small | `npm install @arcships/light-ocr` | about 30 MB | stable default |
+| Tiny | `npm install @arcships/light-ocr-tiny@next` | about 6.3 MB | preview; 49 languages, no Japanese |
+| Medium | `npm install @arcships/light-ocr-medium@next` | about 139 MB | preview; quality-first |
+
+Each facade installs only its selected model.
+
+## Errors and diagnostics
+
+Expected failures reject with `OcrError` and a stable `code`, including
+`invalid_argument`, `invalid_image`, `resource_limit_exceeded`,
+`package_load_failed`, and `inference_failed`.
+
+For environment reports:
+
+```bash
+light-ocr doctor --json
+```
+
+For package/PDF capability checks:
+
+```js
+import { hasPdfSupport, modelProfile } from "@arcships/light-ocr";
+
+console.log(hasPdfSupport());
+console.log(modelProfile);
+```
+
+## Documentation
+
+- [Project overview](https://github.com/arcships/light-ocr)
+- [Node.js and CLI reference](https://github.com/arcships/light-ocr/blob/main/bindings/node/README.md)
+- [PDF 0.5.6 release notes](https://github.com/arcships/light-ocr/blob/main/docs/releases/npm-0.5.6.en.md)
+- [Agent Skill](https://github.com/arcships/light-ocr/blob/main/.agents/skills/local-ocr/SKILL.md)
+- [Changelog](https://github.com/arcships/light-ocr/blob/main/CHANGELOG.md)
+- [Issues](https://github.com/arcships/light-ocr/issues)
+
+## License
+
+Apache-2.0. The package also carries the applicable licenses and notices for
+the bundled PP-OCRv6 model, native runtimes, PDFium, and Noto Sans SC font.