Provenance.
Verify what you can't reconstruct.
Every other page rebuilds content from instructions. This one is the other half of the thesis:
WAI carries whatever it can reconstruct, and verifies whatever it carries. Some content
can't be reduced to instructions — a photograph, a recording, a sensor stream. WAI does not
decode or regenerate it; it carries the opaque bytes, content-addressed, with a signed,
energy-accounted lineage. The image below is a real JPEG — your browser's own
<img> decodes it, while this tab proves its identity (recompute the content
hash, compare to the receipt), verifies the receipt's signature, and shows where each byte came
from and what it cost. WAI is asset-agnostic here: the mechanism is identical whether the origin
is a camera or a synthesizer.
Carried, not reconstructed
WAI never decodes the bytes — the sink hands them to whatever native element renders that media. What WAI adds is a content-addressed envelope and a signed receipt, so the asset is tamper-evident and its lineage is auditable, even though WAI cannot rebuild a single pixel.
Energy-accounted lineage
The receipt is a chain of steps — origin → master → publish — each binding its input and output content hashes and the real measured joules it cost (IOReport, no sudo). The same JWP Merkle + Ed25519 discipline as every other receipt here; it chains across edits.
Tamper-evidence in a tab
Press corrupt one byte: the recomputed content hash diverges from the one the receipt signed, the check goes red, and the image dims. No server, no trust — the proof is a BLAKE3 comparison and a signature check the browser runs itself.