evidence · 3.3 KiB

README.md

Committed at docs/evidence/mainline-uart-progress-memory/no-stimulus-physical-2026-10-04/README.md. This is what grounds the requirements that cite it.

Open raw source

Zero-input comparison: no memory summary observed

The same built Memory image reached a fresh 7.3.0-rc5 banner, exact arguments, /bin/sh entry and primary Bash prompt. During 180.0995 seconds of passive capture, no final memory summary or normal return was observed. The controller sent zero candidate bytes after boot, reporting receipt NOT_REQUESTED and RX NOT_TESTED. No protocol error or candidate reboot was recorded.

Root exclusively operated board/UART from ~/tmp/k230-mainline-probe-integration, branch integrate/mainline-probe-path, controller revision d12091bd. The matching built bundle was unchanged from the preceding one-input comparison: /nix/store/lznjjfx1kzm8mp2j0r06c87h2vdymkhd-k230-mainline-drm-trial-boot-files. Result preserves exact artifacts, UTC times, return code 2 and the private UART hash/size. Fixed records is exactly empty. Raw UART, boot identities and the protected normal baseline remain private.

Actual host preparation passed before UART access. The new operator reset before this test passed fresh protected normal postflight and reviewed Home camera observation. The test repeated normal preflight, all five load/CRC guards and the exact 381-byte Memory argument command. No new kernel build, export or transfer was needed. Persistent normal boot selection was unchanged.

python3 tools/mainline-drm-initrd-shell-trial.py --mode minimal \
  --same-image-shell-pid1 --uart-progress --uart-progress-memory \
  --uart-progress-memory-no-stimulus \
  --bundle /nix/store/lznjjfx1kzm8mp2j0r06c87h2vdymkhd-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"

Withholding the one input attempt did not produce a summary in this run. That observation does not locate the failure or exclude an input-dependent effect under other timing. Worker/observer progress, advancing kernel jiffies and the 45-second completion timeout remain unknown. A summary’s final DBCN output call can itself fail or block; silence cannot separate timer/scheduling/idle/firmware and output dependencies. No camera or real-glass proof was obtained in candidate mode. Ordinary mainline /init, usable root and task 5b.5 remain UNVERIFIED.

The capture is complete. A subsequent NEW operator reset passed protected normal postflight: distinct boot identity, exact system/profile/kernel/init, eight boot hashes, three services and registration absence. Home IPC returned 0 and a reviewed camera image showed its clock/icons/background, with oblique upside-down framing, glare and soft focus. This is not touch/orientation acceptance. Recovery receipt preserves fixed facts and hashes without raw UART or boot identities. Automatic return remains UNVERIFIED.

The separately reviewed next comparison changes only idle/tick policy through one bare volatile nohlt argument, using this same built image and passive controller. Actual host qualification passed. Its candidate observation and recovery remain distinct from this verified normal recovery and ordinary mainline acceptance.