Ever since I was a young boy, words fascinated me — and so did physics. I loved the way words spill down a page, but also the patterns you can see in them that have nothing to do with the meaning: their lacunarity. And I loved physics — the idea that underneath what a thing appears to be, there's a measurable structure you can get at with the right instrument. For most of my life those were two separate loves. PasadoDocs is what happened when they finally fused: when I stopped reading text only for its meaning and started treating it as a physical construct — a thing with shape and signal you can measure.
Theorizing that the way your words line up on a page when you write is not accidental led me to the notion that, perhaps, page lacunarity represents some sort of background signal — a subconscious shape that, maybe, we could detect and measure.
Thusly began this project, PasadoDocs — and thusly began several of the most disappointing weeks of my life, as experiment after experiment after experiment failed to show anything at all. It was random — until it wasn't. Finally, we had a signal, but it was faint.
That is when the journey began. I sort of gave up on page lacunarity — rivers, banks, deltas, diagonals — but, being stubborn and hard-headed, I dived into existing work in lexicology, as well as deriving some of my own tests that had to do with word impact, phrase shape, verb-form patterns, and some other stuff. I even reached deep down inside, back to my days as a lab assistant with a physics group. I started to treat the text as a physical construct and ran it through various modeling techniques. Some worked better than others, and the experiments are ongoing. The ones that worked the best ended up in the stack.
All numbers are based on preliminary science. We've done work building tampered cores — where we told the AI to deviate, to try to imitate Agatha Christie. We gave the AI the full stack of analyzers; it had access to the source code. Interestingly, it failed harder — the harder it tried to get everything right, the worse the prose got. It's like an untrained child trying to forge a Van Gogh. Sure, give them all the tools — it's still a child who can't paint.
The intended pattern of use is whatever you want it to be. Writers can use it and export documents to whatever format they like. Or you can share the document and, in settings, turn off the write protection and transcripts — your transcript will still collect, but it will be stored in the .scribble instead. Or you can be 100% transparent, and prove that your words are 100% yours.
At the end of the day, this is an experiment: can I use physics to build a thumbprint of how writers write, differently from how machines write? Join the experiment, become part of the community, and we can learn together. No accusations. No adverse inferences. Just you, at your keyboard, writing on a surface that protects you from yourself — that shows you where something you pasted might not be yours, tracks it for you, and helps you be a better writer, non-adversarially.