measuring-device-cache.md

What is eating my storage? Measure your own device (no data published)

This is the one topic in this repo where we deliberately publish no dataset — only the method. A disk breakdown is personal (it lists every app you run), and one device is not a representative sample. So run benchmarks/device-cache-scan.sh on your own machine; the numbers stay on your screen.

macOS: where a Mac quietly fills up

du -sk ~/Library/"Application Support"/* ~/Library/Caches/* 2>/dev/null | sort -rn | head -20

• ~/Library/Caches/<app> is regenerable and usually safe to delete (quit the app first). Browsers, package managers (pip, npm, Homebrew), and design tools are the usual top entries — often multiple GB each.
• ~/Library/Application Support/<app> is persistent app data, frequently not safe to delete (it can hold your settings, projects, or local databases). Electron apps and anything with a local cache of documents live here.

The pattern nobody tells you: on a developer or creative Mac, a handful of app caches routinely add up to tens of gigabytes, and they come back after you clear them, because the apps refill them. Clearing buys time, not a fix.

Android: clear cache vs clear data

Without root, Android does not expose per-app cache size to adb (dumpsys storagestats <pkg> returns nothing on Android 15). What you can read is the aggregate:

adb shell dumpsys diskstats | grep -iE "Data-Free|App Data Size"

To act on it, go to Settings → Apps → → Storage, where Android shows the split:

• Clear cache removes regenerable files (thumbnails, downloaded media, temporary data). Safe. The app keeps your login and settings. This is what reclaims space with no downside, and the app slowly refills it.
• Clear data wipes the app back to a fresh install — you are logged out and local content is gone. Only do this to fix a broken app, not to save space.

So the answer to "clear cache vs clear data": clear cache for space, clear data only to reset. The messaging apps (WhatsApp, Telegram) and video apps (YouTube, TikTok) are where cache grows fastest, because they hoard downloaded media.

Why no dataset

We measured on our own devices while building this and chose not to publish the numbers: one Mac's ~/Library reveals its owner's entire software stack, and one phone's app list is equally personal, while N=1 tells you nothing general. If you want a representative version, the honest way is a controlled experiment on a clean emulator with a fixed set of apps — which is a different project. The script above lets you measure the only device whose data is yours to see: your own.