ORIX Scan
Document Intelligence. Beautifully Simple.
A premium, privacy-first document scanning and PDF intelligence application for iPhone and iPad.
Scan, enhance, recognize, organize, and export documents directly on your device.
Author: Simon-Pierre Boucher
Badge values are the results recorded on 2026-11-20 and listed with their evidence in Release Status and
RELEASE-REPORT.md. Nothing was run on a physical iPhone or iPad: see Known Limitations.
Table of contents
- Overview
- Product Philosophy
- Application Preview
- Core Features
- Local OCR Architecture
- PDF Processing
- ORIX Actions
- Document Library
- Privacy and Security
- Technology Stack
- System Architecture
- Project Structure
- Requirements
- Installation
- Build Instructions
- Testing
- Performance
- Accessibility
- Localization
- Development Workflow
- Release Status
- Roadmap
- Known Limitations
- Contributing / Internal Development
- License
- ORIX Ecosystem
📄 Overview
ORIX Scan turns paper into clean, searchable documents. It scans with the system document camera, cleans the page, reads the text with Apple's Vision framework, and writes PDFs that keep the scanned image and carry a selectable, searchable text layer.
Everything happens on the device. The repository contains no networking code and no third-party dependency: the app is built only on Apple frameworks.
| What it is | A native iPhone and iPad app: scanner, page editor, on-device OCR, searchable-PDF builder, document library, workflow automation |
| Who it is for | Anyone who files paper: receipts, invoices, letters, contracts, course notes, books |
| What makes it different | No account, no upload, no tracking; file sizes are measured, not estimated; every claim in this README has a test or a measurement behind it |
| Bundle ID | ai.spboucher.orix.scan |
| Version | 1.0.0 (build 1) |
🧭 Product Philosophy
| Principle | What it means in ORIX Scan |
|---|---|
| Instant Utility | The first screen is Scan Document. No onboarding, no permission prompt at launch |
| No Accounts | There is nothing to sign up for and no sign-in code in the app |
| Privacy by Design | Capture, enhancement, recognition, PDF generation and search all run locally |
| Professional Power | Perspective crop, non-destructive filters, merge/split, password PDFs, batch workflows |
| Exceptional Craftsmanship | One design system (ORIXDesign), Dynamic Type, dark mode, an iPad workspace that is not a stretched phone |
| Reliability First | Crash-safe storage, atomic writes, staged document creation, features that could not be validated are left out |
📱 Application Preview
All images below are screenshots of the running app, captured by the UI tests in the iOS Simulator (iPhone 17 Pro and iPad Pro 13-inch (M5), iOS 26.5 runtime). The documents on screen are sample pages generated by the debug build and processed by the real engines.
![]() Home Scan first, imports and recents below |
![]() Enhancement Before / after, shadows flattened |
![]() OCR Recognized lines over the page |
![]() Searchable PDF Search inside the exported file |
![]() Library Folders, favourites, text search |
![]() ORIX Actions Presets and your own workflows |
![]() Export Real file size per preset |
![]() Tables Editable preview, CSV export |

iPad workspace — sidebar, page rail and a large editor
More screens (text search, crop, action builder, run, organizer, dark mode, French, iPad)
![]() Text with search |
![]() Crop and perspective |
![]() Action builder |
![]() Action run |
![]() Organize pages |
![]() Dark mode |
![]() Français (Canada) |
![]() Settings |
No scanner screenshot. The scanner is the system document camera (VisionKit) and needs a real camera. The Simulator has none and no physical device was available, so that screen has never been displayed or captured. On a device without a camera the app says so and offers imports instead (this path is tested).
✨ Core Features
Capture
| Feature | Notes |
|---|---|
| Document camera | VNDocumentCameraViewController: automatic edge detection, multi-page capture, flash, retake, perspective correction |
| Import photos | PhotosPicker, up to 50 at a time, no photo-library permission needed |
| Import files | Images and PDFs from Files, the share sheet, "Open in", and drag and drop on iPad |
| Protected PDFs | Asks for the password, imports an unlocked copy into the app's protected storage; the password is not stored |
| Edge detection on imports | Vision document segmentation proposes the crop for imported photos |
| Add, retake, replace | Add pages to an existing document, retake a page, replace it from Photos |
Enhancement (non-destructive)
| Feature | Notes |
|---|---|
| Filters | Original, Colour, Greyscale, Black & White |
| Shadow reduction and background cleanup | Paper estimated at low resolution, page divided by it |
| Black & White that keeps faint text | Soft threshold, tested on faint ink |
| Adjustments | Brightness, contrast, sharpening, noise reduction |
| Geometry | Four-corner crop with perspective correction, 90° rotation, automatic straightening |
| Before / after | Draggable split view against the unfiltered capture |
| Originals kept | Every capture is stored untouched; edits are a recipe replayed at render time |
Page quality advice
Blur, low contrast, dark or washed-out pages, low resolution, strong perspective, text touching the edge, and regions the recognizer was unsure about. These are heuristics, presented as advice the user can dismiss or turn off.
Text
| Feature | Notes |
|---|---|
| On-device recognition | Vision, Automatic / Fast / Accurate modes, progress, cancellation, continues briefly in the background |
| Languages | The list is read from Vision at run time; nothing is hard-coded |
| Search, copy, export | Search with match count and highlighting, copy all, .txt export |
| Review and correction | Line boxes over the page, low-confidence lines flagged, per-line editing with revert |
| Tables | Reconstructed from word positions, editable preview, CSV export, uncertain cells highlighted |
Platform integration
Files, share sheet, printing, Spotlight (title and first lines of text, on-device index), four App Intents for Shortcuts and Siri, drag and drop in and out on iPad, keyboard shortcuts (⌘N scan, ⇧⌘I import files, ⇧⌘P import photos, ⌘E export, ⌘1 / ⌘2 sections), context menus, haptics, dark mode, Dynamic Type, optional App Lock.
🔎 Local OCR Architecture
Recognition is VNRecognizeTextRequest running on the device. There is no remote OCR and no way to enable one: the code
base has no networking code at all.
flowchart LR
A["Stored original<br/>(never modified)"] --> B["PageRenderer<br/>crop · straighten · rotate · filter"]
B --> C["Vision<br/>VNRecognizeTextRequest"]
C --> D["Lines and words<br/>text · boxes · confidence"]
D --> E["ReadingOrder<br/>recursive XY-cut"]
E --> F["OCRPage<br/>ocr/<page>.json"]
F --> G["Search · copy · correction"]
F --> H["TableExtractor<br/>rows · gutters · cells"]
F --> I["Searchable PDF<br/>text layer"]
| Aspect | Implementation |
|---|---|
| Modes | Automatic = accurate with language detection; Fast; Accurate with the languages you choose |
| Languages | supportedRecognitionLanguages() per mode. On the development Mac (macOS 27.0) Vision reports 33 languages in Accurate mode and 6 in Fast mode (en-US, fr-FR, it-IT, de-DE, es-ES, pt-BR). The app shows whatever the device reports |
| Reading order | Recursive XY-cut: columns first when the vertical gutter is wide enough, then rows |
| Stale text is impossible | Each result is keyed to the page geometry; changing crop, rotation or straightening discards it |
| Black & White pages | Recognized through the greyscale render, which Vision reads better than a thresholded image |
| Handwriting | Not offered as a feature. Not validated on real handwriting (see Known Limitations) |
Measured accuracy is in Testing. Details: docs/architecture/ocr-engine.md.
📑 PDF Processing
flowchart TB
subgraph Pass1["Pass 1 — Core Graphics"]
J["Rendered page JPEG<br/>(embedded as is)"] --> K["PDF page<br/>A4 · Letter · own proportions"]
L["OCRPage"] --> M["Invisible text<br/>per word, scaled to its box"]
M --> K
end
subgraph Pass2["Pass 2 — PDFKit"]
K --> N["Assemble in page order"]
O["Imported PDF pages<br/>(copied, vector and text kept)"] --> N
N --> P["Metadata · optional password"]
end
P --> Q["Atomic replace at destination"]
| Feature | Notes |
|---|---|
| Searchable PDF | Image kept, invisible text placed word by word; tests read the file back with PDFKit and check that search hits sit on the recognized words |
| Create, import, view | Scans, images and PDFs can be mixed in one document |
| Merge, split, extract | Merge several documents; split after a page or every N pages; extract selected pages. Sources are never modified |
| Pages | Reorder, rotate, duplicate, delete, insert scans / images / PDFs |
| Search and copy text | In the app's text view and inside the exported PDF preview |
| Compression | Maximum Quality, Balanced, Small File Size — the app builds the PDF and shows its real size; it never promises a reduction |
| Password | PDFKit user/owner password on export; protected PDFs can be opened and imported |
| Metadata | Title, author, subject, keywords |
| Share, print, save | Share sheet, AirPrint, Save to Files |
Imported pages that already contain text are copied untouched by every preset: rasterizing them would destroy their text.
Image-only imported pages are re-rasterized by Balanced and Small. Details:
docs/architecture/pdf-engine.md.
⚡ ORIX Actions
A workflow is an input (Scan, Photos, Files or library documents) followed by ordered steps. Every step runs real engine code; the builder edits the same model the engine executes.
flowchart LR
I["Input<br/>Scan · Photos · Files · Library"] --> S1["Step 1"] --> S2["Step 2"] --> S3["Step n"]
S3 --> R["Result<br/>document · exported files"]
S1 -. "progress · cancel" .-> U["Run screen"]
S2 -. "failure stops the run,<br/>later steps skipped" .-> U
Steps: Detect Edges · Straighten · Enhance (filter) · Recognize Text (mode) · Rename (pattern with {title} {date}
{time} {pages}) · Move to Folder · Mark as Favourite · Create PDF (quality, searchable) · Export Text · Export Tables.
| Preset | Steps |
|---|---|
| Scan to PDF | Enhance (Colour) → Create PDF (Balanced, image only) |
| Scan to Searchable PDF | Enhance (Colour) → Recognize Text → Create PDF (Balanced, searchable) |
| Receipt Archive | Enhance (Greyscale) → Recognize Text → Rename Receipt {date} {time} → Move to "Receipts" |
| Document Cleanup | Detect Edges → Straighten → Enhance (Black & White) |
| Book Scanner | Straighten → Enhance (Greyscale) → Recognize Text (Accurate) → Create PDF (searchable) |
| Extract Text | Detect Edges → Recognize Text → Export Text |
| Extract Tables | Detect Edges → Straighten → Recognize Text (Accurate) → Export Tables |
| Compress PDF | Create PDF (Small File Size, searchable) |
Presets are read-only; duplicating one makes an editable copy. Workflows can be renamed, duplicated, reordered, deleted, and
run on several library documents in one go. Details: docs/architecture/workflows.md.
🗂 Document Library
Recent documents, all documents, favourites, folders, Recently Deleted (30 days), search across titles and recognized text (case- and accent-insensitive), sorting (modified, created, name, size, pages), filters (favourites, with / without text, scans, imports), grid or list, rename, duplicate, move, batch selection with export / move / favourite / merge / delete, thumbnails with a disk and memory cache.
flowchart TB
subgraph Disk["Application Support / ORIXScan"]
M["Documents/<id>/manifest.json<br/>source of truth, replaced atomically"]
P["pages/ — original captures"]
F["pdf/ — imported PDFs"]
O["ocr/ — recognized text per page"]
C["cache/ — previews, safe to delete"]
S["Staging/ — documents being created"]
end
X["Library.store<br/>SwiftData index"] -- "rebuilt from manifests<br/>if missing" --> M
S -- "one directory move<br/>on commit" --> M
A document only appears once it is complete: it is assembled in Staging/ and moved into place in a single step. Leftovers of
an interrupted import are discarded at the next launch. Details:
docs/architecture/storage.md.
🔒 Privacy and Security
| Rule | How it is met |
|---|---|
| No account, no registration | No sign-in code exists |
| No remote OCR, no external AI | Recognition is Apple Vision on device |
| No upload, no cloud requirement | No networking code: grep -rnE "URLSession|URLRequest|NWConnection|http://|https://" App Packages --include='*.swift' returns nothing |
| No analytics, no ads, no tracking | No SDK of any kind; PrivacyInfo.xcprivacy declares no tracking and no collected data |
| Permissions only when needed | Camera is requested when the scanner opens; the photo picker needs no permission |
| Files encrypted at rest | iOS data protection, class complete until first user authentication, so recognition started in the foreground can finish if the device locks |
| Optional App Lock | LocalAuthentication (Face ID, Touch ID or passcode); a cover hides content in the app switcher |
| Temporary files | Exports go to a dedicated directory, emptied at launch |
| Logs | The app writes no logs containing document content; error messages never include it |
App Store privacy answers ("Data Not Collected") and the required-reason API declarations are in
docs/release/app-store.md.
🧰 Technology Stack
| Layer | Technology | Role |
|---|---|---|
| Language | Swift 6 language mode, strict concurrency | App and package |
| UI | SwiftUI (NavigationSplitView, TabView, Observation) |
All screens; no web view |
| Scanner | VisionKit | System document camera |
| Recognition | Vision | Text recognition, document segmentation |
| Imaging | Core Image, Core Graphics, ImageIO | Rendering pipeline, quality analysis, encoding |
| PDFKit, Core Graphics, Core Text | Assembly, text layer, encryption, preview | |
| Index | SwiftData | Library records, folders, saved workflows |
| Storage | FileManager, atomic writes | Originals, manifests, recognized text |
| Automation | App Intents | Shortcuts and Siri |
| Search | Core Spotlight | On-device index |
| Security | LocalAuthentication, iOS data protection | App Lock, encryption at rest |
| Project | XcodeGen, Swift Package Manager (local package) | Build definition |
| Third-party code | None | — |
Core ML is not used directly: recognition goes through Vision, which is the supported interface to Apple's on-device text models.
🏗 System Architecture
flowchart TB
subgraph App["App target — SwiftUI, main actor"]
UI["Screens<br/>Home · Library · Document · Text · Export · Actions · Settings"]
AM["AppModel<br/>capture, edits, recognition, export"]
RT["Router<br/>tabs on iPhone, sidebar on iPad"]
PL["Platform<br/>Intents · Spotlight · App Lock · Share · Print"]
end
subgraph Kit["ORIXKit — engines, no UI"]
DP["DocumentProcessor<br/>facade"]
DS["DocumentStore<br/>actor, files"]
PR["PageRenderer<br/>Core Image"]
OC["OCREngine<br/>Vision"]
PB["SearchablePDFBuilder<br/>PDFKit + Core Graphics"]
WE["WorkflowEngine"]
LI["LibraryIndex<br/>SwiftData"]
end
DSY["ORIXDesign<br/>tokens and components"]
UI --> AM --> DP
UI --> DSY
AM --> LI
AM --> WE --> DP
DP --> DS
DP --> PR
DP --> OC
DP --> PB
PL --> AM
flowchart LR
subgraph In["Import"]
A1["Document camera"] --> B1
A2["Photos"] --> B1
A3["Files · drag and drop · Open in"] --> B1
B1["AppModel.receive"] --> C1["Draft in Staging/"]
C1 --> D1["commit → Documents/"]
end
D1 --> E1["Recognize (optional, automatic)"]
E1 --> F1["Index + Spotlight"]
subgraph Out["Export"]
G1["PDF"]
G2["Text"]
G3["CSV"]
G4["Images"]
end
F1 --> G1 & G2 & G3 & G4
G1 --> H1["Share · Print · Save to Files · Preview"]
Full description: docs/architecture/overview.md.
📁 Project Structure
ORIXScan/
├── App/
│ ├── Sources/
│ │ ├── Core/ app entry, AppModel, Router, settings, naming, shared views, root
│ │ ├── Home/ first screen
│ │ ├── Library/ library, folders, batch actions
│ │ ├── Document/ workspace, page editor, crop, organizer, text, tables, export
│ │ ├── Capture/ document camera wrapper
│ │ ├── Actions/ ORIX Actions list, builder, run
│ │ ├── Settings/ settings and language picker
│ │ ├── Platform/ intents, Spotlight, App Lock, share, print, thumbnails
│ │ └── Debug/ sample documents and isolated store (Debug builds only)
│ └── Resources/ asset catalog, string catalogs, Info.plist, privacy manifest
├── Packages/ORIXKit/
│ ├── Sources/ORIXDesign/ design tokens, components, haptics
│ ├── Sources/ORIXKit/ Model · Imaging · OCR · PDF · Storage · Actions
│ └── Tests/ORIXKitTests/ engine, pipeline, library and benchmark tests
├── UITests/ XCUITest suite and import fixture
├── Scripts/ build, benchmark, release, localization, hygiene
├── design/icon/ icon concepts, masters and generator
├── docs/ architecture · design · development · release · screenshots · assets
├── project.yml XcodeGen definition
└── RELEASE-REPORT.md| Code | Files | Lines |
|---|---|---|
| App (Swift) | 22 | 4 852 |
| ORIXKit engines (Swift) | 17 | 2 923 |
| ORIXDesign (Swift) | 3 | 499 |
| Package tests | 5 | 1 233 |
| UI tests | 1 | 13 tests |
| Localized strings | 464 keys | English, French (Canada) |
📋 Requirements
| Run | iPhone or iPad with iOS / iPadOS 18.0 or later. Scanning needs a camera |
| Build | Xcode 27.2 (Swift 6.4 toolchain), XcodeGen, macOS with the iOS 27.2 SDK |
| Accounts | None to run. An Apple Developer team to sign for devices |
| Dependencies | None beyond the Apple SDK |
📥 Installation
ORIX Scan is not on the App Store. Build 1.0.0 (1) is on TestFlight for the account's internal testers only. To run it otherwise, build from source:
git clone https://git.spboucher.ai/orix-scan.git
cd orix-scan
brew install xcodegen # once
Scripts/build.sh generate # creates ORIXScan.xcodeproj
open ORIXScan.xcodeproj # run the ORIXScan scheme on a simulator or a deviceTo run on a device, set your own team in project.yml (DEVELOPMENT_TEAM) and bundle identifier if needed.
🔨 Build Instructions
Scripts/build.sh generate # XcodeGen
Scripts/build.sh kit # engines and design system on the Mac: all package tests
Scripts/build.sh debug # Debug build for the simulator
Scripts/build.sh release # Release build for the simulator
Scripts/build.sh uitest # UI tests on the "ORIX iPhone" simulator
ORIX_SIM="ORIX iPad" Scripts/build.sh uitest
Scripts/benchmark.sh # OCR accuracy and timings → docs/development/benchmarks/
Scripts/check.sh # author headers and secret scan
Scripts/localize.py # refresh the string catalog and merge FrenchThe dedicated simulators are created once with
xcrun simctl create "ORIX iPhone" "iPhone 17 Pro" and xcrun simctl create "ORIX iPad" "iPad Pro 13-inch (M5)".
Debug builds accept -ORIXUITest YES (isolated store), -ORIXReset YES and -ORIXDemo YES (sample documents).
Setup details: docs/development/setup.md.
🧪 Testing
Results recorded on 2026-11-20 on the final code. Full strategy and what is not covered:
docs/development/testing.md.
| Suite | Where it runs | Result |
|---|---|---|
| Package tests (engines, storage, library, workflows) | Mac host, real Vision / PDFKit / Core Image / SwiftData | 79 passed, 0 failed (plus 2 opt-in benchmark tests, also passing) |
| UI tests | iPhone 17 Pro simulator, iOS 26.5 | 13 passed, 0 failed |
| UI tests | iPad Pro 13-inch (M5) simulator, iOS 26.5, landscape | 13 passed, 0 failed (one harness flake in a third run, see testing.md) |
| Physical device | — | Not run. No iPhone or iPad was connected |
| Code coverage | — | Not measured |
What the tests actually prove, beyond "it compiles":
- Generated PDFs are read back with PDFKit: text extraction,
findString, and the position of each search hit against the recognized word box. - After every compression preset the exported page is rasterized and recognized again to check it is still readable.
- A simulated crash during import leaves nothing behind; a committed document survives a relaunch.
- Corrupted, truncated and password-protected PDFs, unreadable images, an empty document and low storage each produce a specific error.
- In the UI: first launch with no prompt, camera unavailable, empty library, import → recognition → search inside the exported PDF, page editing, split, an action run end to end, deletion and restoration, persistence after relaunch, dark mode, the largest accessibility text size, French.
OCR benchmark
A synthetic corpus of 17 pages with known ground truth, drawn with Core Text and degraded on purpose (skew, blur, shadow, low resolution). These are not photographs of paper: they measure the pipeline, not real-world capture. Character error rate (CER) and word error rate (WER), lower is better; measured on the development Mac with the same engine code the app ships.
| Page | Automatic CER / WER | Accurate CER / WER | Fast CER / WER |
|---|---|---|---|
| English, clean print | 0.3 % / 2.2 % | 0.3 % / 2.2 % | 0.0 % / 0.0 % |
| English, dense small text | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| English, serif | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| French, clean print | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.3 % / 2.4 % |
| French, serif | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.3 % / 2.4 % |
| Mixed English / French | 1.1 % / 6.8 % | 1.1 % / 6.8 % | 0.4 % / 2.3 % |
| Receipt, monospaced | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 5.2 % / 61.9 % |
| Low resolution (75 dpi-equivalent) | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 5.4 % / 20.0 % |
| Skewed 3° | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| Skewed 8° | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| Two columns | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| Strong side shadow | 0.3 % / 2.4 % | 0.3 % / 2.4 % | 0.3 % / 2.4 % |
| Dark page | 0.3 % / 2.2 % | 0.3 % / 2.2 % | 0.0 % / 0.0 % |
| Faint ink | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 0.0 % / 0.0 % |
| Out of focus | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 53.4 % / 77.8 % |
| Script font (Snell Roundhand) | 0.0 % / 0.0 % | 0.0 % / 0.0 % | 100 % / 100 % |
| Casual script (Bradley Hand) | 0.3 % / 2.4 % | 0.3 % / 2.4 % | 41.9 % / 90.2 % |
Reading: Accurate mode is strong on everything in this corpus; Fast mode is unreliable on receipts, low resolution, blur and
script fonts, which is why Automatic uses the accurate recognizer. The two script rows are computer fonts, not handwriting.
Raw data: docs/development/benchmarks/ocr-accuracy.json.
🚀 Performance
Measured, not estimated. Engine figures come from Scripts/benchmark.sh on the development Mac (Apple M5 Max, 18 cores,
48 GB, macOS 27.0), which was running other workloads; they are not iPhone or iPad figures. Launch time comes from
XCTApplicationLaunchMetric in the Simulator. Method and caveats:
docs/development/performance.md.
| Measurement | Result |
|---|---|
| Cold launch to first responsive frame, iPhone simulator | 1.13 s average of 5 (1.08–1.17 s); a later run under load: 1.41 s (1.12–1.60 s) |
| Cold launch, iPad simulator | 1.19 s average of 5 (1.14–1.23 s); a later run under load had four launches at 1.74–1.87 s and one 19 s outlier |
| OCR, one A4 page at 300 dpi, Accurate | 0.13 s and 0.23 s (median of 5, two runs) |
| OCR, one A4 page at 300 dpi, Fast | 0.09 s and 0.12 s |
| Full-resolution render of a 12-megapixel capture, including JPEG encoding | 0.03–0.09 s per filter across two runs |
| Editor preview (1 800 px) | 0.03 s and 0.06 s |
| Recognize a 25-page document (A4, 300 dpi) | 2.8 s in one run, 5.3 s in another |
| Export 25 pages as searchable PDF | Maximum 2.2–2.7 s · Balanced 1.5–2.0 s · Small 1.2–1.3 s |
| Size of that export (stored originals: 9.2 MB) | Maximum 9.0 MB · Balanced 3.1 MB · Small 1.5 MB |
| Memory across the 25-page batch | Footprint did not grow between the start and the end of the batch; peak was not sampled |
| Camera-to-preview latency | Not measured (needs a physical device) |
♿ Accessibility
| Area | State |
|---|---|
| Dynamic Type | All text uses Dynamic Type styles; layouts switch to vertical stacks at accessibility sizes. The largest size (AX XXL) is exercised by a UI test on Home and Document |
| VoiceOver | Controls carry labels, hints, traits and values; decorative images are hidden; the crop corners and the before/after split are adjustable elements. Not verified with VoiceOver on a device |
| Touch targets | 44 pt minimum on custom controls |
| Colour | State is never carried by colour alone (icons and text accompany it) |
| Motion | The scanning beam animation is disabled with Reduce Motion |
| Dark mode | Every colour token has a light and a dark value; exercised by a UI test |
🌍 Localization
| Language | State |
|---|---|
| English | Source language |
| French (Canada) | 463 strings translated, 12 with singular / plural forms; Home and Library verified by a UI test. Other French screens were not individually reviewed for truncation |
Strings live in App/Resources/Localizable.xcstrings and InfoPlist.xcstrings. Scripts/localize.py re-extracts keys from the
build, merges Scripts/translations-fr.json, and fails if a key is untranslated. Recognition languages are independent of the
interface language.
🔁 Development Workflow
- Change engine code in
Packages/ORIXKitand runScripts/build.sh kit(seconds, on the Mac). - Change the app, run
Scripts/build.sh debug, then the UI tests on both simulators. - New user-facing strings:
Scripts/build.sh debug && Scripts/localize.py, add the French toScripts/translations-fr.json. - Before committing:
Scripts/check.sh(author header on every source file, secret scan). - Commit messages follow
feat:,fix:,design:,test:,docs:,release:.
🚦 Release Status
| Gate | State | Evidence |
|---|---|---|
| Compiles, Debug and Release | ✅ | 0 errors, 0 warnings |
| Package tests | ✅ 79 / 79 | Scripts/build.sh kit |
| UI tests, iPhone and iPad simulators | ✅ 13 / 13 each | Scripts/build.sh uitest |
| On-device OCR verified | ✅ in tests and Simulator | CER / WER table above |
| Searchable PDFs verified | ✅ | PDFKit read-back tests, in-app PDF search test |
| Persistence | ✅ | relaunch tests (package and UI) |
| Export | ✅ PDF, text, CSV, images | package tests, UI tests |
| Icon assets | ✅ compiled into the archive | light, dark, tinted |
| Signed Release archive | ✅ 1.0.0 (1) | Apple Distribution, App Store profile |
| App ID registered | ✅ ai.spboucher.orix.scan |
created by automatic signing |
| Physical-device testing | ❌ not done | no device connected |
| Scanner (camera) | ❌ never exercised | Simulator has no camera |
| App Store Connect record | ✅ created 2026-11-20 | Apple ID 6837545072, name only; metadata not entered yet |
| Apple validation of the IPA | ✅ | altool --validate-app: no errors |
| TestFlight upload | ✅ 1.0.0 (1), authorized by the owner | processing state VALID |
| Internal testers | ✅ available | internal group "Test", 1 tester (invited) |
| External testers | ⏳ not submitted | needs Beta App Review and a separate decision |
| App Store | ⏳ not submitted | separate approval |
Details and outstanding approvals: RELEASE-REPORT.md,
docs/release/testflight.md,
docs/release/release-checklist.md.
🗺 Roadmap
Not commitments; candidates in rough order.
- Validation on physical iPhones and iPads with the TestFlight build
- Layered Icon Composer icon in addition to the flat three-appearance asset
- Apple Pencil annotation
- Handwriting, if it can be validated on real samples
- Optional iCloud Drive export destinations in ORIX Actions
- Moving
ORIXDesigninto a package shared by the ORIX apps
⚠️ Known Limitations
- Never run on a physical device. The camera scanner, Face ID / Touch ID, VoiceOver, background continuation, memory pressure and real performance are unverified on hardware.
- OCR accuracy figures are from a synthetic corpus, not photographs of paper.
- Handwriting is not a feature. Only computer script fonts were tried.
- Isolated single characters can be missed. In a table drawn for testing, Vision did not return cells holding a single digit (
3,7) in any mode, while12and25were read. Tables with one-character cells can therefore come out with empty cells. - Table extraction is heuristic. Merged cells, nested headers and wrapped cell text are not reconstructed. Corrections made in the table preview apply to the exported CSV and are not saved.
- Edited lines lose word-level positioning in the PDF text layer (the corrected line is laid over the line box as one run).
- Imported pages that contain text are never recompressed.
- Unlocking a protected PDF redraws its pages, which drops form fields, links and outlines of that file.
- Quality warnings are estimates and can be wrong in both directions.
- French beyond Home and Library was not screen-by-screen reviewed.
- No Apple Pencil annotation, no iCloud sync, no widgets.
- Folders are one level deep.
🤝 Contributing / Internal Development
This is a private, single-author project. Internal rules:
- Apple frameworks only; a third-party dependency needs a written reason.
- No feature is shown to users unless it works and has a test or a documented manual check.
- Numbers in documents come from a command in this repository.
- Every source file starts with the author header;
Scripts/check.shenforces it. - No credentials, signing material or personal documents in the repository (
.gitignore,Scripts/check.sh).
📜 License
All rights reserved © 2026 Simon-Pierre Boucher. See LICENSE.
🪐 ORIX Ecosystem
ORIX is a family of independent, premium Apple utilities that share one idea: useful the moment they are installed, no account, private by design.
| App | State |
|---|---|
| ORIX Scan | This repository |
| ORIX Audio, ORIX Calc | Separate projects under development |
| ORIX PDF, Image, Video, Measure, Dev, Files, Math, Secure | Ideas, not started |
ORIX Scan contributes two reusable pieces: the ORIXDesign design system (tokens, components, the scan-beam signature, the
accent pair model) and the ORIX Actions workflow model. See
docs/design/design-system.md,
docs/design/brand-guidelines.md and
docs/design/icon-design.md.
Simon-Pierre Boucher — design, development
www.spboucher.ai · git.spboucher.ai · contact@spboucher.ai















