diff --git a/.changeset/ui-helper-and-extended-vision.md b/.changeset/ui-helper-and-extended-vision.md
deleted file mode 100644
index 39523b0..0000000
--- a/.changeset/ui-helper-and-extended-vision.md
+++ /dev/null
@@ -1,15 +0,0 @@
----
-"macos-vision": minor
----
-
-feat: ui-helper (windows / displays / permissions / captureScreen) and extended Vision API
-
-- New native `ui-helper` (third prebuilt binary): `listWindows`, `listDisplays`, `checkPermissions`, `captureScreen` (returns path + geometry + sha256, never bytes). `checkPermissions` also reports `screenLocked`; capture failures consult it so a locked Mac is diagnosed outright rather than guessed at.
-- OCR tuning: `languages`, `autoDetectLanguage`, `languageCorrection`, `customWords`, `fast`, `regionOfInterest`, `minTextHeight`; opt-in content-hash `cache`; `onProgress` for PDFs.
-- `recognizeDocument` (macOS 26+): native paragraphs, tables, lists, title, barcodes and detected data with positions.
-- `extractEntities` (links, e-mails, phones, addresses, dates), `detectTextRegions`, `compareImages`, `imageInfo`, `visionCapabilities`, `supportedOcrLanguages`.
-- People & scenes: `detectFaceLandmarks`, `detectHumans`, `detectBodyPose`, `detectHandPose`, `detectAnimals`, `detectAnimalPose`, `detectHorizon`, `detectSaliency`, `detectContours`, `imageAesthetics`, `detectLensSmudge`.
-- Pixel ops that return paths: `cropImage`, `cropDocument` (perspective-corrected), `extractForeground`, `personMask`.
-- Native helpers still target macOS 12. Modern features are gated twice — at compile time on the SDK (`-DSDK_14/15/26`, set automatically by the build scripts) and at runtime on `#available` — so the `swiftc` fallback also builds on older macOS, with unavailable features reported as `false` by `visionCapabilities()` and raising `UnsupportedOnThisMacOSError`. Release builds refuse to run on an SDK too old to compile in every gated feature.
-- Helper failures now surface their own message (`Cannot open file: …`) instead of Node's `Command failed: /path/to/vision-helper …`.
-- Opt-in OCR cache is keyed on helper and macOS version, so results never survive an upgrade that changes them.
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3167277..e33e265 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,20 @@
 # Changelog
 
+## 1.6.0
+
+### Minor Changes
+
+- 2b576bf: feat: ui-helper (windows / displays / permissions / captureScreen) and extended Vision API
+  - New native `ui-helper` (third prebuilt binary): `listWindows`, `listDisplays`, `checkPermissions`, `captureScreen` (returns path + geometry + sha256, never bytes). `checkPermissions` also reports `screenLocked`; capture failures consult it so a locked Mac is diagnosed outright rather than guessed at.
+  - OCR tuning: `languages`, `autoDetectLanguage`, `languageCorrection`, `customWords`, `fast`, `regionOfInterest`, `minTextHeight`; opt-in content-hash `cache`; `onProgress` for PDFs.
+  - `recognizeDocument` (macOS 26+): native paragraphs, tables, lists, title, barcodes and detected data with positions.
+  - `extractEntities` (links, e-mails, phones, addresses, dates), `detectTextRegions`, `compareImages`, `imageInfo`, `visionCapabilities`, `supportedOcrLanguages`.
+  - People & scenes: `detectFaceLandmarks`, `detectHumans`, `detectBodyPose`, `detectHandPose`, `detectAnimals`, `detectAnimalPose`, `detectHorizon`, `detectSaliency`, `detectContours`, `imageAesthetics`, `detectLensSmudge`.
+  - Pixel ops that return paths: `cropImage`, `cropDocument` (perspective-corrected), `extractForeground`, `personMask`.
+  - Native helpers still target macOS 12. Modern features are gated twice — at compile time on the SDK (`-DSDK_14/15/26`, set automatically by the build scripts) and at runtime on `#available` — so the `swiftc` fallback also builds on older macOS, with unavailable features reported as `false` by `visionCapabilities()` and raising `UnsupportedOnThisMacOSError`. Release builds refuse to run on an SDK too old to compile in every gated feature.
+  - Helper failures now surface their own message (`Cannot open file: …`) instead of Node's `Command failed: /path/to/vision-helper …`.
+  - Opt-in OCR cache is keyed on helper and macOS version, so results never survive an upgrade that changes them.
+
 ## 1.5.0
 
 ### Minor Changes
diff --git a/package.json b/package.json
index 8ae267e..07250cc 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "macos-vision",
-  "version": "1.5.0",
+  "version": "1.6.0",
   "description": "Apple Vision OCR + image/PDF analysis for Node.js, with optional Ollama-driven Markdown pipeline — native, fast, offline",
   "author": "Adrian Wolczuk",
   "license": "MIT",