Case study / Product build, own
Know what a file really is before it burns you
Born from a real prepress incident: a file that looked fine and was not. DropCheck tells you what a file actually is, where it will break, and what it is leaking, without it ever leaving your machine.
Live, in-browser- 100% client-side
- Nothing uploads
- Range-read forensics

The engagement
Challenge
A file that looks fine can be the one that breaks the run
A file that looks fine can carry the flaw that breaks a print run or quietly leaks metadata. And you often cannot send a sensitive file to a server to find out what is wrong with it. The check has to happen locally, on files that may be large.
Action
Forensics that never leaves the browser
An in-browser forensics tool with a framework-free core designed to be extracted as a standalone package, range-reads only so a 500 MB file costs a few MB of memory, and a hard privacy invariant: nothing uploads, every parse and fix runs client-side.
Result
Drop a file, learn the truth, at zero privacy cost
Drop a file and learn what it really is, where it will break, and what it leaks, with the privacy guarantee as the product, not a setting. A real static build exists.
How it is built
The privacy guarantee is the architecture
There is no upload step to remove, because there is no server. That is the whole design.
- 01Framework-free core. The forensics core has no framework or DOM dependency, so it lifts out as a standalone package.
- 02Range reads. A RangeReader over Blob.slice never loads a whole large file; a 500 MB PSD costs a few MB of memory.
- 03Per-format analyzers. Marker walks for JPEG, PSD, and TIFF that surface real verdicts, not trivia.
- 04Privacy invariant. Nothing transmits file bytes or metadata off the machine, ever. The guarantee is the product.
- 05Honesty. A published known-limitations list; when a scan is capped, the UI says so.
What it proves
The privacy guarantee is the product: forensics that never sees a server.
See the rest of the work →Built from a real prepress incident, described here in the abstract with no client or file detail.