A signature encodes motor habit, not intent. It was never good security. I want to make both halves of that case and then say something about what handwriting actually was, as a technology, because we're watching it get deprecated and nobody has written the migration guide.

What a signature is

When you sign your name, you aren't drawing letters. You're executing a motor program. The cerebellum and motor cortex have compiled a sequence of muscle activations, refined over thousands of repetitions, that runs open-loop: once started, it does not need visual feedback to finish. This is why you can sign with your eyes closed, or on a screen you cannot see, and get something recognizably yours. It is also why your signature has degraded over your adult life into a shape that bears no relation to your name. The program optimized for speed and lost the letters along the way.

Forensic document examiners know this, and their real tools are not the shapes. They're the dynamics. Pen pressure, which shows up as line width and indentation. Speed, which shows up as line smoothness: a fast stroke is clean and tapered, a slow traced one is wobbly and blunt-ended. Pen lifts. Stroke order. A forger copying the shape of your signature has to go slowly and look at it, and slowness leaves a trace. The best forgeries are made by people who have practiced your signature until it is their motor program, at which point it's fluent and also not really a copy anymore.

So a signature verifies that a particular motor program ran. It doesn't verify who ran it, and it says nothing at all about what the person meant. That last part is the one contracts pretend it does.

It was never security

Signatures worked as authentication for the same reason passwords worked in 1985: not because they were hard to forge, but because forging one was rarely worth it and the cost of getting caught was high. The security was social. The signature was a commitment device, a way of making a person physically do a thing that they could later be confronted with. The check cleared because the bank knew you, not because your loops were unique.

The proof is what happened when signatures went online. The little box on the delivery pad, the mouse-drawn squiggle on the loan form, the typed name in a cursive font on a PDF: none of these preserves any of the dynamics a forensic examiner would use, and everyone knows it, and everyone accepts them anyway. They're legally sufficient because the law never required the signature to be unforgeable. It required an act. The act is now a click, the squiggle is theater, and the theater is kept because dropping it would make people uncomfortable.

If you want to authenticate a person, you use something with actual entropy: a key, a device, a secret. The signature was a ritual. It still is. It's just a ritual that now happens on glass.

The format

Set signatures aside. Handwriting as a whole was a compression format, in the sense that it was a standardized encoding of language into marks, optimized for a specific channel, and the interesting part is which constraints shaped it.

Cursive is the clearest case. It exists because quills and later steel nibs didn't like being lifted. Every pen lift risked a blot, cost time, and broke the ink flow. So the letters were redesigned to connect, and the connections were standardized, and the result was an encoding where a word is a single continuous stroke. That's a speed optimization for a specific hardware constraint. Ballpoints, which lift freely, removed the constraint in the 1940s, and cursive started dying about a generation later, once the people who had been taught it under the old constraint stopped teaching it.

The Palmer Method, which dominated American schools from about 1900 to the 1950s, was explicitly a standard. Uniform slant, uniform letter forms, drilled by arm movement rather than finger movement so that clerks could write all day without cramping. It was designed for throughput. A Palmer-trained clerk could produce legible copy at a sustained rate that a typewriter of the era couldn't match, and the encoding was interoperable: any other Palmer-trained reader could decode it at speed. It was, in effect, a font that humans rendered.

Shorthand went further. Gregg and Pitman shorthand are real compression: phonetic encodings that drop vowels, merge common sequences, and represent whole words with single strokes. A trained stenographer wrote at 200 words per minute, faster than most people speak. It was lossy, it required a trained decoder, and it was the highest-bandwidth text input method available to a human for about a century. It died the moment a machine could do it.

The recognition problem

The attempts to get computers to read handwriting are a nice illustration of how much the format depended on a human decoder.

The Apple Newton in 1993 tried to recognize ordinary cursive and became a punchline, mostly because cursive was never designed to be decoded by anything but a human brain with a language model and forty years of context. Palm's Graffiti, a few years later, did the opposite: it defined a new alphabet of single strokes that a computer could read reliably and made the humans learn it. That worked. The humans adapted in a week. The format bent toward the decoder.

Modern recognition, which is quite good, works by doing what the Newton could not: running an actual language model over the stroke data and picking the most likely word given the context. Which is to say, it reads handwriting the way a person does, by guessing, and it took thirty years of progress to get there.

Deprecation

The format is dying because the encoder stopped shipping. Schools in most of the US stopped requiring cursive around 2010, when it was left out of the Common Core standards, and a lot of them stopped teaching it. Kids now learn print, then keyboards. There's a cohort of adults, mostly under thirty, who cannot read a handwritten letter from their grandparents.

I find that I don't have strong feelings about whether this is bad. The constraints that produced cursive are gone. The throughput it optimized for is beaten by any phone. What is lost is what's always lost in a format migration: the archive. Letters, diaries, marginalia, the entire handwritten record of a few centuries, is now in a format that a shrinking number of people can read natively, and there is no batch converter. There's only the slow, imperfect, machine-assisted transcription of things that someone decides are worth it. Most of it won't be.

That's what deprecation looks like. Not a shutdown. Just a slow loss of readers, until the marks are still there and nobody can hear them.

My own signature, for the record, looks like a cardiogram of someone having a mild event. It's the same cardiogram every time. That's all it was ever going to prove.