# Conformance — kido-interop vs kido.json (2026-10-15)

Line-by-line check of `imports/kido.json` against the interop draft
(`modules/kido-interop.h/.c`, `architecture/doll-layers.md`).
Prior conformance reviews left two items DEFERRED: doll-layer placement and
the inter-kernel protocol. Both are addressed here.

## MET

1. **"Kido is the outer kernel-node, for the outer part."**
   `doll-layers.md` fixes the seats: Kido = outer seat, BUBO = inner seat.
   `KIDO_KERNEL_KIDO = 0` is this kernel's declared seat in every envelope.

2. **"BUBO (original C kernel) holds the inner system; Kido holds the
   outer (language ball)."**
   Placement documented; the `KIDO_TYPE_LANG` packet type is Kido's
   outer-seat identity made into protocol — synthesized language fragments
   are the traffic this boundary exists to carry.

3. **"Compensates BUBO's flaws: ... Kido is designed portable and
   adaptive from the start."**
   The protocol's asymmetry law puts ALL adaptivity on Kido's side (probe,
   retry, solo-degradation); BUBO owes one behavior (ACK). The module is
   portable C with zero arch-isms (ARM64 `-ffreestanding` cross-compile
   clean), and the driver contract names no medium — Kido adapts to
   whatever transport a future Gate approves.

4. **"The outer doll's kernel seat — counterpart to BUBO's seat in the
   inner system."**
   Seats are fixed and named in both the doc and the code
   (`kido_kernel_t`); only the outer seat moves.

5. **"Two kernels, one organism: ... They divide the stack by doll layer,
   not by competition."**
   The protocol contains no leader election, no preemption across kernels,
   no resource arbitration — the kernels do not compete, so the protocol
   does not arbitrate. Cooperation is structural (asymmetric duties), not
   negotiated.

6. **"Architectural node in the system design today; a runnable Kido ...
   is a deep build, Phase 4 territory."** — MET by discipline.
   Kido's side of the protocol is real and self-tested; the BUBO peer is
   declared simulation, the bare-metal transport is a declared stub (loud
   `KIDO_ERR_STUB_REACHED`, never a guessed medium). No bootable kernel is
   finalized; first boot remains N8toriousAi's to witness.

## Prior DEFERRED items — now resolved

- **Doll-layer placement** (boot conformance DEFERRED) → DEFINED in
  `architecture/doll-layers.md`: seats, what lives where, what crosses,
  what never crosses.
- **Inter-kernel protocol** (boot/HAL conformance + resonate C5 DEFERRED)
  → DRAFT. The envelope wraps the carried-whole `kido_signal_t` exactly as
  resonate C5 required ("plugs in without changing the API");
  `kido-resonate.h` is unmodified.

## Still DEFERRED (declared, not hidden)

- The real BUBO peer (Phase 4; the test peer is declared simulation).
- The bare-metal transport medium (no medium guessed; driver contract is
  medium-agnostic).
- Ordering / duplicate guarantees across the link (explicitly unclaimed;
  receivers idempotent-tolerant by contract).
- The 5%/hop decay constant is a draft value — N8toriousAi tunes it.

## Honest limits of this draft

- The self-test's "peer" is test-authored packets, not BUBO. It proves
  Kido's envelope, handshake, decay math, version rejection, and loudness
  — 75 checks, 0 failures — and nothing about BUBO.
- The ANNOUNCE wait consumes stray arrivals (documented in the header);
  real traffic starts after LINKED.
- ARM64 paths cross-compiled (`-ffreestanding`, zero warnings), not
  executed — execution proof belongs to the QEMU harness sessions, and
  the interop layer is not yet wired into any boot image.
