How do I recover files deleted by Purgley?

Because Purgley prioritizes user control and safety, files marked for removal are not instantly destroyed. Instead, they are moved directly to your system's native 'Recently Deleted' holding area. You have a 30-day window to recover any asset directly through native system apps before permanent hardware clearance takes place.

Does Purgley collect any personal data or usage metrics?

No. Purgley is designed as a local-first, privacy-isolated utility. Your media library and file system are analyzed entirely on-device via local CoreML models; no data ever leaves your hardware. To monitor application stability and catch crashes, we use lightweight, anonymous telemetry signals via TelemetryDeck that strip away all identifying network footprints.

Can Purgley damage my device's operating system?

No. Purgley operates strictly within safe user-space boundaries and respects native OS permissions. System-critical assets and locked directories are completely protected by default. While users have the authority to bypass certain target filters, the app cannot modify or delete files essential to your device's core operating system.

How does Purgley's on-device AI handle my photos and files?

Purgley utilizes optimized, native CoreML machine learning models to analyze asset characteristics and map them into a localized vector space. This mathematical clustering groups similar items, heavy files, and system bloat purely within your device's secure memory. No embeddings, vector data, or raw files are ever transmitted to external servers or remote AI APIs.

Why Is My iPhone's Duplicates Album Missing Similar Photos?

If you upgraded iOS expecting the native Duplicates utility folder to instantly fix your messy camera roll, you have likely run into a frustrating limitation.

The system catches identical file copies perfectly. However, it completely ignores the five near-identical burst shots you took of a moving subject, the slightly edited copy of a photo you shared, or a cropped screenshot. Your eyes tell you they are duplicates, so why is Apple blind to them?

The answer comes down to an intentional design limitation in how native mobile operating systems process image files.

1. The Strict Definition of a Duplicate

Apple's native duplicate detection is primarily engineered around exact or near-exact system matches.

Byte-Level vs. Perceptual Matching

When iOS scans your library in the background while your phone is charging, it looks for specific, strict markers:

Identical Metadata: Assets sharing nearly exact timestamps, filenames, and capture locations.
Byte-Level Properties: Direct file format twins or simple re-encodings (such as a raw .HEIC file alongside a compressed .JPEG counterpart of the exact same pixel matrix).

When you tap Merge, the system safely blends these direct clones together, preserving the highest quality version and combining tags into a single asset.

2. Why Similar Photos Break the Apple Pipeline

The native system explicitly fails on intentional visual variations, what software engineers categorize as Near-Duplicates.

When you take four photos in a row to ensure nobody blinks, or crop the border out of a landscape shot, the binary configuration of that file changes completely. To a computer checking raw metadata or direct pixel strings, a 48-megapixel original and a cropped square snippet look like completely unrelated pieces of data.

Apple strictly limits its detection sensitivity to prevent false positives. The system is intentionally conservative; it would rather leave 1,000 burst frames cluttering your device storage than risk an automated engine accidentally deleting an irreplaceable memory because the background geometry looked similar.

3. How Perceptual Hashing Catches What iOS Misses

To clear out visual clutter instead of just exact database duplicates, software has to look past file properties and analyze the actual layout, shapes, and structural signature of the image. This is exactly how independent tools like Purgley bridge the gap.

Instead of basic data matching, every photo runs through an optimized local Perceptual Hashing (pHash) pipeline:

The Perceptual Hashing Pipeline

Downsample and Grayscale: The asset is scaled down to a minute grid (such as 32 X 32 pixels) and stripped of color variables.
Discrete Cosine Transform (DCT): A mathematical formula decomposes the image into frequency blocks, isolating large structural layouts from fine noise.
Binarization: The lowest 64 frequencies are compressed into a dense, unique 64-bit fingerprint.

If you shift your camera position slightly, change a facial expression, or crop the canvas, the core visual fingerprint remains virtually unchanged.

4. Engineering Clean Clutter Detection Responsibly

Processing thousands of complex visual fingerprints requires significant mobile computing power. Brute-forcing an image library can easily trigger severe thermal throttling or cause an Out-Of-Memory (OOM) crash on mobile hardware.

Optimizing Memory for On-Device Processing

By skipping heavy cloud servers and utilizing hardware-level downsampling (CGImageSourceCreateThumbnailWithTransform), optimized utility engines can analyze image structures safely inside the local sandboxed chip environment.

The logic is simple: comparing these signatures takes mere nanoseconds using bitwise operations (nonzeroBitCount). This allows the device to cluster massive bursts and near-identical tracking shots together without sacrificing battery health or privacy.

Great storage cleanup should not require sending your private photo gallery to a third-party server. It just requires an engine that analyzes layout the way your eyes do.

Want to experience the speed of clean, local-first engineering? Download Purgley on the App Store and take control of your storage privacy today.