Y2B-245 — Real screen result and sleep failure

> **Historical record.** The dates, candidate identity, "current" claims, > next steps and permissions below belong to this recorded boundary. See > [current state](../CURRENT_PLATFORM_STATE.md) for the latest physically observed result.

Historical failed iteration. The owner subsequently reports successful periodic PID1 execution; retain the -38 screen below as the evidence that led to the time32 fix.

Owner-reported hardware evidence, recorded 2026-09-08 in #20:

WDT: STOPPED
FB: GUARD OK
PID1: 1
BEAT: 0
STAGE: STOP: SLEEP ERROR
PROC MOUNT RC: 0
SYSFS MOUNT RC: 0
LAST ERR: SLEEP -38
SLEEP RC: -38

This is a real device report, not the synthetic screen preview. It confirms Linux 6.18 reaches initramfs PID1, the guarded inherited framebuffer text renderer works, and proc/sysfs mounts return success. The displayed watchdog/guard status records checks at the successful frame write. No completed sleep, increasing heartbeat/frame sequence, timer accuracy or additional runtime fields are claimed. M1's Linux-plus-PID1 objective remains achieved; periodic execution is unproven. No new flash/restore chronology or stock verification is inferred from this report.

Focused cause: missing time32 ABI, not POSIX timers

Audited local upstream v6.18 (7d0a66e4bb9081d75c82ec4957c50034cb0ea449):

arch/arm/tools/syscall.tbl:179: syscall 162 uses sys_nanosleep_time32.
kernel/time/hrtimer.c:2204: that implementation is gated by
CONFIG_COMPAT_32BIT_TIME and calls hrtimer_nanosleep with CLOCK_MONOTONIC.
kernel/sys_ni.c:337: COND_SYSCALL(nanosleep_time32) provides the weak fallback.
arch/Kconfig:1466: time32 is a visible bool; allnoconfig disables it despite
its normal 32-bit default. It is independent of CONFIG_POSIX_TIMERS.

The tested out/m1-y2b245-diagnostics/kernel.config disables both options. Actual ELF sys_call_table[162] = 0xc00309a8, identical to sys_ni_syscall and weak sys_nanosleep_time32. This explains the immediate -38 / ENOSYS without requiring a timer, UART or framebuffer redesign. Existing mocked syscall tests validated error handling but could not detect a missing kernel implementation.

Y2B-250 #21 adds only CONFIG_COMPAT_32BIT_TIME=y to the kernel fragment. POSIX_TIMERS stays disabled. The new validator inspects the real linked table and rejects an ENOSYS target or weak fallback, including when config text incorrectly claims support. D08, CPU0, watchdog, framebuffer, PID1 source and loader packaging are unchanged.

The fix still needs a device test: real syscall presence does not independently prove timer interrupts/wakeup. The current task prepares a new offline candidate; it does not authorize flashing it or beginning M2.