People see normal content. Machines read a second layer from the same physical medium. No app, no QR code, no cloud, no network — decoding and rendering happen entirely on your device.
The receiver has two sensors: the camera reads light, the microphone listens. Images and audio never leave your device.
Traditional media serves people only. This system lets one medium serve two kinds of reader — and the human layer is completely untouched. It just looks like normal content.
Ordinary images, ordinary video, ordinary music. Content for people, exactly as it was.
The other layer in the same medium: asset ID, scene, state, timecode, command.
The real animation file is carried whole inside the image. What the receiver gets back is that file itself — not an ID, so it needs nothing in advance and no network round trip to look anything up.
The print channel yields only tens of bits per capture, so "micro" is its physical limit — and the right design. The hard part is detection and localisation on a busy layout, not decoding.
No speech recognition, no large language model. The sound itself is the packet. Web Audio gives full control over the signal, which makes this the channel most likely to ship first.
A carrier that moves can put a different slice of data in every frame, and the receiver accumulates them back into a whole file — fountain-coded, so losing half the frames changes nothing. The protocol for this is finished; the interface is not wired to it yet.
Once the camera or microphone captures the signal, decoding, parsing and rendering all happen in the browser. No image upload, no cloud inference, no server round-trip. Lower latency, better privacy, and it still works in a gallery basement with no signal.
Physical channels differ enormously in what they can carry: roughly 1,900 bytes from a single photograph, about 3 bits per second through the air. So the protocol is tiered — each channel carries what it can, sharing one format. The content itself travels in the signal, so the receiving device needs nothing prepared in advance.
Figures marked "Unverified" come from published literature and commercial systems. They have not been confirmed by our own measurements. Replacing them with measured values is the entire purpose of Stage 0.
The protocol layer and both channels — image and audio — are usable. Drop an image into the studio, type a paragraph, and the file that comes out looks identical to the eye; point a phone at it and the paragraph comes back.
What is missing is measurement on real hardware. Both channels pass a full simulated channel (speaker response, room reverb, lens blur, JPEG), but the real world always holds something the simulation does not. Until then every capacity figure is marked unverified.
Both the optical and acoustic channels are invisible now: over half the pixels move by 1/255, and the audio perturbation sits below the psychoacoustic masking threshold. Perspective distortion is still unsolved, and every number here comes from a simulated channel — none of it has been measured on real hardware.