DZG STUDIO — measurements and field notes

What the AI label can mean, how we communicate across languages, and why useful work needs checks against reality.

An “Other” bucket is not a fact about your notes.

Every image in 1,231 Codex conversations fingerprinted. 92% of the duplicate screenshots sit in conversations used this week, so the age slider points at the wrong end; a fork depends on another file's byte offset; and the clever fix was the smallest.

No single local model wins every language, so Alder measures quality and routes each language quietly.

How large local models reach real Macs reliably, recover from interruption, and stay verifiable.

How Alder protects a spoken thought through silent microphones, interrupted sessions, and blocked fields.

A confident detector named 16.9 GB of abandoned iOS runtimes.

What coding agents leave on a developer's disk, measured with read-only scan hooks: a summary that photocopies screenshots, an archive that frees nothing, and how to read 14 GB without killing the machine.

A first-use test turned a proud, feature-heavy interface into something quieter.

Why a name that suggests care and renewal fit better than one that simply described cleanup.

The harder question was not how to classify files, but when Mint should act and when it should ask.

What Mint learned by moving from a fast cross-platform prototype to software designed around the Mac.

Different files, same pattern: digital clutter accumulates faster than most people can organize it.