soft3/proposals/tade-one-exchange.md

cyber soft3 architecture proposal tade process cyber
π 0.01% 875 words · 5 min

tade — one exchange format, from the wire to the screen

builds on cybergraph + sync + tade (implemented 2026-06), which made tade the wire substrate under radio and foculus. this proposal finishes the thought: tade is the one format data takes on every boundary of the stack — wire, the payload of a link inside a signal, a CLI's machine channel, the durable log, the stream a painter reads — and the serialization libraries the stack borrowed along the way go.

the claim

the stack has one data model already: the nox noun (atom / pair), provable and canonical; neuron data is encoded in it. tade today is a second structural format where it carries kv / tables, and postcard / serde_json are a third and fourth wherever a crate needed a quick codec. four truths for one value.

the resolution is a division of labour, not a winner:

noun   (nox)    computes      — the value, provable, canonical bytes
tade   (frame)  carries       — (sigil · render) + payload, text-safe, streamable, nestable
spark  (dialect) shows        — the catalog of the render byte: what a (sigil, render) pair renders to, in cells or pixels
prysm  (paint)  draws         — pixels / world units / terminal cells from spark's answer

one value, four roles, four components. a tade frame's structural payload is a canonical noun; tade adds exactly what the noun lacks — a type and a way to show — and nothing else.

what changes

# step where size
1 canon. one value, one byte sequence: ordered keys (no HashMap on the read or write side), minimal LEB128, fixed child order, conformance vectors that pin every rule tade spec + crate one spec page, ~50 lines, vectors
2 payload = noun. the structural payload of a frame is the canonical nox noun encoding; tade frames of structured data and nox data are the same bytes under a 2-byte annotation tade spec (one page on the binding), nox reference cross-link one page
3 catalog → spark. the (sigil, render) vocabulary staged in tade/spec/7-catalog.md and molecule.rs is spark's: spark is the catalog of the render byte. prysm reads spark; it does not hold a vocabulary of its own tade, spark, prysm move + links
4 proofs own their bytes. zheng and lens stop encoding proofs and commitments with postcard — the bytes a verifier hashes must be defined by zheng's spec, not by a crates.io version. a proof on any boundary is a tade frame whose payload is zheng's own layout zheng (12 postcard call sites), lens (6) the real work of this proposal
5 CLI: tty → ANSI, pipe → tade. every CLI follows zheng's tade_out: styled text on a terminal, a tade chunk stream when piped, ending in a STATUS chunk. serde_json leaves the machine channel (≈130 call sites across bbg, foculus, neuron, cybergraph, soft3, cyb, lens, zheng, hemera); serde derives (≈60) go with it all CLIs mechanical, repo by repo
6 ladder. tade's chip moves from rung 7 (network) to rung 4 (cybergraph), beside file and signal: it is the bytes of a file and the payload of a link on any boundary. done 2026-10-07 — the page already reads so soft3.org (the ladder, front page) —

radio's upstream iroh messages (QUIC session chatter) stay postcard: that is transport noise, not stack data. the line is: what has meaning is tade; what only moves bytes between endpoints is iroh's business.

why not the alternatives

  • noun everywhere, no tade. a noun stream is not text-safe and not self-delimiting; it cannot be cat-ed into a terminal next to plain output or rendered incrementally. tade's 0x1F marker never occurs in valid UTF-8 — that property is the whole reason a typed stream can live inside a terminal.
  • tade everywhere, no noun. then provability and canon would have to be re-derived for a second structural format, and nox would compute over one thing while the graph stored another.
  • postcard / serde as the canon. a borrowed codec's version becomes the definition of a proof's identity. unacceptable for anything a verifier hashes.

acceptance

this proposal closes only when all of the following hold — not when the pages are written:

  1. tade conformance vectors exist and pass in every crate that encodes or decodes frames (tade, foculus, cyb, prysm, rune, zheng cli, bbg cli);
  2. rg postcard in zheng and lens returns nothing outside the vendored transport;
  3. no CLI in the stack writes serde_json to a pipe; each has the tty / pipe split;
  4. cargo test is green in every repository touched, and every PR is merged — the proposal is closed by the last merge, not by the first.

sequencing

specification first (steps 1–3): small, and it fixes what the implementation must match. implementation (4–5) after the harness work in flight is finished — not before, and not interleaved with it. the chip move (6) is already live.

tade · nox · spark · prysm · cybergraph · zheng · lens

Graph