Screenshot lab — the protected viewer

Open a protected document, try each capture method against it, then feed the captured file back into its row. The image is read with OCR and scored against the reference text, in this browser. The question each row answers is not whether the capture looked damaged, but whether it produced a reusable copy.

Flicker note. The experimental level modulates the overlay every frame at a higher amplitude. It is disabled automatically under prefers-reduced-motion. Do not leave it running in front of anyone who is sensitive to flicker.

Document id Level screenshotProtectionResponse captureProtection

Reference text — paste what the captured page actually says, or OCR has nothing to be scored against.

Method	Platform	OCR chars	OCR words	Usable copy?
PrintScreen key	Windows	—	—	Score a capture
Win + Shift + S	Windows	—	—	Score a capture
Snipping Tool	Windows	—	—	Score a capture
Cmd + Shift + 4	macOS	—	—	Score a capture
Browser screenshot	Chrome / Edge / Firefox	—	—	Score a capture
DevTools capture	any	—	—	Score a capture
Screenshot extension	any	—	—	Score a capture
Phone camera	physical	—	—	Score a capture
Phone camera, HDR on	physical	—	—	Score a capture
Screen recording, frame extracted	any	—	—	Score a capture

OCR words is the column that decides it. Someone copying a document needs the words, not the pixels — a capture can look ruined and still score 98%.

Capture events seen by the page

Recorded while the viewer is open. Repaint is the time from the event handler running to the response being on screen. It is the number that decides whether reacting is worth anything: if the operating system grabs the framebuffer before it elapses, the response changed nothing about that capture.

Nothing yet. Open the viewer and try a capture — a method that produces no row here cannot be reacted to at all.

How to run this honestly

Open the viewer on the machine you actually deploy to.
Take a real capture with one method. Do not use a browser or automation capture as a stand-in — they take the pixels by a different route and will tell you nothing about the OS ones.
Check the event table above. No row means the page never knew, and no response was possible.
Open the captured file and fill in its row below, then score it by OCR.

Expect PrintScreen and Win+Shift+S to be unaffected. Windows delivers PrintScreen as a keyup after the framebuffer is taken, and the Win+Shift+S overlay freezes an image of the screen the moment it is invoked — so in both cases the page is told once it is already too late. Snipping Tool opened as an app is the plausible case, because seconds pass between it taking focus and the user dragging a rectangle. Measure it rather than trusting that.