Most cheaper printers (esp. on Windows) use the GDI protocol for printing. These printers only know how to print rasterised images, so the document is rasterised by the OS/Print driver and only this final rasterised image is sent to the printer. This is different from higher end PCL/PS printers where the document is translated into a page description language and the printer is (partially) responsible for rasterising the final document for print.

Since Windows uses the GPU to render fonts I wouldn't be surprised if the same code is used to rasterise the fonts for GDI printing.

That being said I'm very surprised they can identify the GPU just from that, unless there is some specific bug in the driver for the card which produces an obvious font rendering artefact.

Gaussian blur is your friend if you wanna send a death note, I guess.

Also, I doubt you can get conclusive evidence from a single letter. Luckily, your average random note has lot of them, even duplicated ones. I would carefully align and average out as many capital A’s as I had, and work with that.

You missed the explanation above why you're wrong, at least on consumer non-PostScript printers. Most cheap printers nowadays passes the buck of rasterisation to Windows (and its horrible, security headache spooler). You can even check if which is which: in Windows 10, open Settings, then Devices, select Printers & scanners, select [your name of printer], press Manage, press Printer Options (not Printing options), open the Advanced tab and then click on the Print Processor... button. If it says "winprint" then Windows handles the rasteriser.

I think they meant something like "ink smears".

Pinpointing a version of Windows, if it was printed from a stock OS font, could be as simple as comparing tiny differences in the vector files and knowing if one pixel would rasterize at 60% black versus 50%. To the extent that the rip goes through a graphics card, it would be knowing whether that card rendered the 60% as 58% or 62%.

I'm pretty sure if you scale it down, the printer driver will do an extra layer of downsampling and add its own anti-aliasing; but the printer hardware doesn't do that, it just sprays the dots it's told to spray, and in general the drivers replicate the pixels that are sent from Photoshop or in this case, MS Word, which uses something like QuickDraw used to be on a Mac, an embedded system process, to rasterize the fonts.

Prior to this it had never occurred to me. But yeah, more randomness. Noise filter and blur, then a bit more noise, then photograph it, print the photo, scan it on another device, put it in the washing machine, leave it on the porch for a week and repeat.

Could also be by design, similar to printer identification dots. Have the artifacting vary every so slightly from one GPU to another. Then again, I feel (emotional statement, not of fact) that this would be known by now if it was a thing.

So...evidently from a sample of me, I can tell from which monitor a screenshot was taken...

https://www.eff.org/pages/list-printers-which-do-or-do-not-d...

Edit: Specifically, what about printers that only print black and white?

https://www.eff.org/press/archives/2005/10/16

The implication is that only color printers are affected.

"The U.S. Secret Service admitted that the tracking information is part of a deal struck with selected color laser printer manufacturers, ostensibly to identify counterfeiters."

https://en.m.wikipedia.org/wiki/EURion_constellation

https://upload.wikimedia.org/wikipedia/commons/7/7b/Obverse_...

Probably just idiots playing around with the copier rather than dedicated gangs, but if it worked they might be tempted to say "Well ..."

https://old.reddit.com/r/mildlyinteresting/comments/1s8rl9/i...

[1] https://www.eff.org/pages/list-printers-which-do-or-do-not-d... [2] https://en.m.wikipedia.org/wiki/Machine_Identification_Code

I'm not sure these days when most GPUs are IEEE-754-compliant. But back in the mid to late 2000's I worked on a GPU renderer for video editing and we had a few filters that gave noticeably different results on different GPUs. One filter did a hard black and white threshold, then blurred the result, did another hard threshold, etc., in a loop. Because of differences in precision of the floating point values (24-bit on AMD at the time, if I recall correctly), the thresholds could produce minor differences that got magnified by the blurring, and then created new thresholds with minor differences, etc.

Even if all the GPUs are using IEEE-754 floats, there are driver differences that can cause the results to be slightly different, too. Like a simple GLSL mix() function could be implemented as result = x * a + y * (1 - a) (where x and y are 2 input pixels and a is the alpha of x). Or it could be implemented more efficiently as result = a * (x - y) + y. Doing the same math in a slightly different way can sometimes lead to slight differences in intermediate results which compound in the final result. So yeah, it may be possible to tease out some of these things by examining something like font rendering.

Becomes extremely easy for malicious actors (out or inside the police) to fake evidence and frame anyone they'd like.

Bite mark identification was used forever until blown up by particularly shameless grifting, and has never been shown to work as practiced. [1]

Tennessee still uses dowsing rods. [2]

Fingerprinting as practiced is a bundle of folk practice, guesses, and some science. Quality varies wildly. [3]

Fiber analysis, lie detectors, spatter analysis and many more techniques are all crap. When one bogus method is finally found legally unreliable, cops and prosecutors find a new one.

[1] https://innocenceproject.olemiss.edu/radley-balko-reports-on...

[2] https://www.themarshallproject.org/2022/03/17/witching-dowsi...

[3] https://www.aaas.org/resources/latent-fingerprint-examinatio...

And I'm wondering about that, because we know the criminal must have been a pretty hard-core cryptocurrency nut. There aren't THAT many of them in Norway (they've already concluded they are a fluent Norwegian speaker).

Instead of facing it just as a lead, it might influence investigators to, consciously or not, build confidence in framing a (wrong) person and and end up building a compelling case against them.

The person who wrote the ransom they believe to be fluent in Norwegian, there could easily be other people involved who're foreign.

Do we?

I was still under the impression everybody was blaming her husband for the disappearance.

As for malicious actors, wouldn't that be a risk for most forms of evidence? Likewise, wouldn't many of the techniques used to establish the validity of other forms of physical evidence be applicable when these techniques are used?

Ah yes, I see they’re using the default formatting options. That narrows down our search to 99.9999% of the population.

That narrows it down to coming from 10s of millions of computer perhaps?

⇒ I don’t think it’s fair to call this speculation, let alone pure speculation.

Seriously though, I thought printers have been using microdots as identifiers for years. Is this just an old wives tale?

Just make sure you pay with cash.