Method for the Adverb StonesBook One
How to tell a machine how, with your hands.
A stone on a tray changes what an AI does: cheaply, by Friday, secretly, ask me first. Your phone is the bow. A slow arch means take care. A flick means now. Stones that contradict each other sound wrong before anyone acts on them.
Playable today: motion, sound and a recognizer that learns your strokes, all in this page. Not built: the physical stones, and any connection to an agent. A phrase ends as a sentence you can copy, share or hand to an Apple Shortcut. Nothing is for sale.
A phrase is a spell in the plain sense: set objects, a set motion, and something happens. Stones are adverbs. Strokes are verbs. How you move is the inflection. The pause between strokes is the punctuation. These are written the way you'd write a score, and the sentences below come from the same code the instrument runs.
The first example has no stroke at all. Tap two objects and the agent has to work out the verb; a stroke is how you stop it guessing.
| Part | On a violin | Here |
|---|---|---|
| Stones | the tuning and the key | Set the scale, the chord and the tone. They also set the adverbs. Order is priority: the first stone placed wins an argument, and it's the root of the chord. |
| Motion | the bow | A still phone makes no sound. Rotation speed is volume, tilt picks the note, roll opens the tone, a sharp jerk plucks. |
| Thumb | a finger on the string | Moving without the thumb is humming. Moving with your thumb on the pad is speaking: that motion gets read as a stroke. |
| Speed and size | dynamics | Same shape, different meaning. A flick says right now. A slow arc says take care. A wide stroke says everything like this; a tiny one says only this. So a lazy flick isn't a shortcut. It says something else. |
| Rests | phrasing | Two strokes close together happen at once. A longer rest means "then". |
| Cadence | the ending | Speaking the phrase ends it. "Ask me first" hangs on a suspended chord until you let it go. Contradictions beat against each other, so you hear the problem first. |
Start at the dot. Hold the phone like a stick, top edge pointing away, and move from the wrist.
What each one tells the agent, and what it does to the sound. Print a tray: each stone is a QR code and a URL you can write to any NFC tag.
The page measures you on three exercises and remembers your best on this phone. You can decide for yourself whether they're fun.
Beyond those, the path is pitch by feel. There's no fret, so finding a note by tilt alone is like playing a theremin, and that's the hard part on purpose.
It becomes a plain sentence and a small block of JSON. Today you carry it the last step yourself: copy it into any AI, share it to an app, or run an Apple Shortcut called "Stones" that you build to do whatever you like, such as asking an assistant or dimming a lamp. That last hop takes one tap each time, because a web page can't act on its own. A native app could connect directly to an agent. We haven't built that.
Yes for short, deliberate strokes you teach it yourself, with your thumb marking when a stroke starts. No for working perfectly on strangers, or for spotting strokes in everyday motion with no thumb. The uWave study (8 people, 4,480 gestures) reports 98.6% when templates keep adapting, 93.5% when frozen, and 75.4% across people. iPhone Safari delivers motion at a nominal 60 Hz, and a stroke under two seconds doesn't need a compass.
We ran this page's matching code on the original uWave recordings. It's real hands, on a Wii remote's accelerometer. Not an iPhone, and not this page's gyroscope pipeline:
That's why teaching comes first in the instrument. Nobody has swung a phone at this page yet. We don't have a measured number for a real iPhone in a real hand.
You blew into an iPhone and it played. Tilt added vibrato for people who stayed. It cost 99 cents, about 700,000 people had bought it by March 2009, and its two versions later passed ten million users. The motion copied something people already knew how to do. So sound comes first here, with depth hidden until you want it.
Launch motion sensing read swing direction but not strength, and players soon learned that a small flick did the same as a full swing. Once that's true, the gesture is dead. It's why a flick here means something different from a full stroke.
RCA built a run of about 500 and advertised it as easy. It wasn't, and playing with no physical reference felt disorienting. It lived on through virtuosos like Clara Rockmore, not living rooms. Stones are the physical reference it lacked. We don't claim this is easy.
Place objects on a table and their relationships make the sound. Björk toured with it and it won awards, yet the company has been dissolved and its products are gone. Audiences understood it at once. It also needed a table and a camera, and a phone needs neither.
€630 million in 2025 revenue, most of it figurines. People keep buying objects when each one contains something new. A stone that means "cheaply" contains nothing new the second time, so this may never work as a collection. We don't know of any product where objects act as combinable modifiers at this scale. That's the untested part.
One tap to start sound and ask for motion access. Safari asks every so often.
Sound hasn't started. Tap anywhere here to try again, and check the ring switch.
The Shortcut button only works on an iPhone where you've made a Shortcut named Stones that takes text as input. It's up to you what that Shortcut does. No agent has been told anything until you send this.