# Conformance review: kido-image vs kido.json + scaffold law (2026-10-16)

kido.json is a placement/role spec ("portable and adaptive from the
start"; "compensates BUBO's flaws"), not a module checklist. This
review maps the session's work against the spec's laws and the
scaffold's own standing decisions. Verdicts: MET / DEFERRED /
MET-by-discipline.

## Spec law

1. **"Portable and adaptive from the start" (kido.json role) — MET.**
   kido-image.c is portable C with one arch seam: bare-metal ARM64
   reads linker symbols; hosted uses an explicit test injection. No
   x86-isms, no arch #ifdef beyond the scaffold's standard
   ARM64-or-hosted discipline (matching every other module).
2. **"Compensates BUBO's flaws" — MET-by-discipline.** BUBO's second
   flaw (per the boot deliberation) was untestable-until-hardware.
   The reserve path is genuinely testable on the dev host via the
   injection seam (sched self-test section 30-39 pins the carve);
   the bare-metal symbol path is verified by cross-compile + link +
   readelf symbol-order proof. Nothing is claimed to have executed
   on bare metal that hasn't.
3. **Honesty law (scaffold standing: loud stubs, never silent) — MET.**
   Mislinked image (end <= start) -> KIDO_ERR_EARLY_ENV, and
   sched_init propagates it before the allocator initializes. The
   hosted STUB_REACHED -> vacuous-0,0 mapping is declared in the
   design record and both headers, not a silent fallback. No new
   status codes invented (sched D1 precedent: new header codes are
   N8toriousAi's call).

## Standing decisions honored

4. **Sched deliberation D2 — CLOSED.** reserve 0,0 replaced by the
   kido-image bounds on both build paths (hosted: injected or
   declared-vacuous). The "safe by measured layout" era ends here.
5. **Alloc-sched-integration P1 ("alloc init in one place") — MET.**
   The bounds query lives inside kido_sched_init; kido-boot.c is
   untouched, the allocator's interface is unchanged, and the
   reserve carve-out needed zero logic changes.
6. **No Gate-approved logic modified except sched_init's reserve
   args — MET.** kido-boot.c/h, kido-alloc.c, kido-hal-arm64.*,
   kido-mem.*, kido-entry.*, kido-interop.*: zero logic changes.
   kido-resonate.c: one TESTING-comment line (build line gains
   kido-image.c), no logic. kido-link-arm64.ld: symbols added,
   `_stack_top` byte-identical (proven via old-vs-new link
   comparison, 0x40088b10 both).
7. **"No bootable kernel finalized without N8toriousAi" — MET-by-discipline.**
   This session drafts a module and wiring; the full build plan and
   first boot remain his.

## Deferred / unproven (honest limits)

- **DEFERRED:** bare-metal execution of the reserved tree (QEMU).
   Assembled and linked clean; the symbol order is proven; the
   shared carve path is tested hosted. Execution proof is the
   natural next verification session.
- **DEFERRED:** DTB-region trimming (deliberation C4) — belongs to
   kido-mem's discovery output, not this module.
- **DEFERRED:** relocation detection (deliberation D1) — a future
   kido-reloc module's job; the link-at-load law is stated, not
   enforced.
- **UNPROVEN (by design):** `_image_end <= _image_start` loud path on
   bare metal — only reachable via a deliberately broken script;
   the hosted NULL-guard paths are tested.
- D1 (KIDO_ERR_NO_TASKS header code) remains N8toriousAi's call — untouched.
