Physical finite UART-progress comparison

The coordinator ran the reviewed controller at 3dc78e07 in /home/jadams/tmp/k230-mainline-probe-integration, branch integrate/mainline-probe-path. Root alone reserved the board/UART. Camera and build slot were not used. The fixed public result preserves the observations, immutable artifacts, completion timestamp, private capture hash and limits; raw UART, protected baseline and reception nonce remain private.

Actual matching host qualification had passed. The board downloaded and verified the new candidate export, imported its eight additional paths and staged the matching boot files; staging returned zero. The controller then independently verified protected normal system/kernel/profile, boot file hashes and services, rebooted into U-Boot, qualified every loaded artifact’s size/CRC and verified the volatile arguments. Persistent normal boot selection was not changed.

Operator command (private directory contains no repository credentials):

python3 tools/mainline-drm-initrd-shell-trial.py \
  --mode minimal --same-image-shell-pid1 --uart-progress \
  --bundle /nix/store/gmsmqkjcb8vmh6h44xvq7y59dd9xihsh-k230-mainline-drm-trial-boot-files \
  --manifest "$PRIVATE_RUN/candidate-manifest.json" \
  --normal-report "$PRIVATE_RUN/normal-report.json" \
  --log "$PRIVATE_RUN/trial.private.log" \
  --result "$PRIVATE_RUN/result.private.json"

The command returned 2 (observation saved, recovery required), rather than a successful mainline acceptance. The 180.079-second passive capture saw the fresh 7.3.0-rc5 banner, exact received kernel arguments, /bin/sh init entry and anchored Bash primary prompt. It attempted exactly one fresh builtin receipt command. No receipt and no complete reporter records arrived. There were no parser protocol errors; an empty set of records is not proof of healthy counters. No candidate retry, proc command, guard or reboot followed the input attempt. No fresh protected normal return was observed within the passive capture. The subsequent operator reset passed guarded protected normal postflight: fresh boot identity, exact system/profile/kernel/init, eight protected boot hashes, three active shell services and registration absence. Home IPC returned zero and a separate camera observation showed normal clock/icons/background, with angle and focus limits. Reset recovery receipt. This is manual recovery; automatic mainline restart, mainline root, timer health, UART delivery and panel/glass acceptance remain UNVERIFIED.

Interpretation and next discriminator

The kernel printed that the SBI DBCN extension was detected. In selected arch/riscv/kernel/sbi.c, that detection sets sbi_debug_console_available. This grounds availability detection, not successful reporter initialization or firmware output. The reporter currently creates a normal-priority kthread and its first output follows worker scheduling, msleep(5000) returning, the cached snapshot and direct firmware write. No report distinguishes none of those steps. There are no UART or timer counter values to compare in this trial. Earlier source/config/object evidence does not replace that missing physical observation.

The next narrow comparison should add a separately qualified fixed direct-SBI worker-entry breadcrumb before the first sleep, preserving the existing six samples, runtime opt-in and normal priority. It stays off IRQ, TTY and PID1 paths. Presence proves the worker reached that firmware call, not that the call returned; absence still leaves scheduling versus output unknown. A later sample would then prove return from the entry call and a delayed wakeup. This is a proposed source comparison, not implemented or tested hardware behavior. Stronger markers on the init/PID1 path need an explicitly reviewed scope change. Repeating the identical sleeping-worker experiment offers no new discriminator.