# Spec conformance — kido-entry-arm64 (2026-10-14)

Against import://kido, the Gate-approved kido-boot interface, and the
kido-hal-arm64 bootinfo contract. Verdicts: MET / PARTIAL / DEFERRED /
MET-by-discipline.

## kido.json (import://kido)

1. Portable and adaptive from birth; compensates BUBO's x86-bound
   heritage — **MET.** Zero x86-isms; ARM64-only file with a
   compile-time arch guard. Adaptivity preserved: the shim fills the
   contract, selection still happens in boot's adapt stage.
2. Outer kernel-node, second in the BUBO line — **MET-by-discipline.**
   The shim is the outer layer's power-on path; it does not reach into
   BUBO's inner-system concerns.
3. First boot is N8toriousAi's to witness — **MET-by-discipline.** The shim
   boots to a loud halt today (probe stub); no bootable kernel is
   claimed. The QEMU run is a shim proof, not a kernel boot.

## kido-boot interface (Gate-approved — NOT modified)

4. _start calls kido_boot(); every stage return honored — **MET.**
   kido_entry_c_main calls kido_boot(), then kido_halt(status) on any
   return. Loud-halt discipline preserved end to end.
5. kido_boot_early()'s stack/BSS checks now verify real work — **MET.**
   Previously vacuous on hosted; on bare metal they verify this shim's
   SP setup and BSS loop. The checks ran during the QEMU boot.
6. kido_halt() spins with the status code — **MET.** Reached in the
   QEMU boot (exit 124, zero exceptions); the code is preserved for a
   debugger/watchdog per the boot header's contract.

## kido-hal-arm64 bootinfo contract

7. Fill kido_bootinfo_t before probe() — **MET.** Shape filled:
   board_id (loud assumption), mem_* zeroed with mem_valid=0 (loud
   stub), DTB magic-checked and stashed.
8. Run at EL1 or higher (ID_AA64* traps at EL0) — **MET.** EL2->EL1
   drop via HCR_EL2/SPSR_EL2/eret; EL1 proceeds; EL0/EL3 spin.
9. Never invent platform facts — **MET.** The memory map is NOT
   fabricated (mem_valid=0); the board is a documented assumption, not
   a detection claim; an invalid DTB is flagged, not worked around.
10. DTB pointer preserved for kido-mem — **MET.**
    kido_entry_dtb_ptr() / kido_entry_dtb_valid() accessors; the raw
    pointer survives in BSS past the C handoff.

## Gaps (honest)

- **PARTIAL:** board detection — assumed QEMU virt, pending kido-mem's
  DTB compatible-string parse. Loud and documented, not silent.
- **DEFERRED:** memory map (mem_valid=0) — kido-mem's bounded space.
- **DEFERRED:** MMU/cache bring-up, kido_main, RPi4 — unchanged from
  prior sessions.
- **Unproven:** which EL path executed in QEMU; cycle-level BSS
  correctness (QEMU -kernel loads zero pages — see D1 in the design
  record). Stated, queued for a poison-fill debug build.

**Tally: 10 MET (incl. 2 by discipline), 1 PARTIAL, 3 DEFERRED, 2
explicitly unproven.** Nothing claimed beyond what ran.
