U-Boot motion measurement attempt

The 30-second U-Boot target recording could not be registered: the unchanged tools/panel-measure.py found only markers 0 and 1, below its minimum of three. Measurement output is a failed measurement, not a zero-motion result. Splash task 3.7 remains open.

Provenance and procedure

The board runs the corrected initial-scene diagnostic image from source 12c1b0e, integrated as opt-in code in master 61dcc90. Its system is /nix/store/24h4sb1msncgbasnwl3ckdlxqf9v2g6r-nixos-system-nixos-26.11.20260919.20b1ddd. No image flash or readback was performed for this trial.

The deterministic default grayscale array from tools/panel-target.py (seed 20260921) was encoded as B,G,R,X, with X=255, for the existing U-Boot 568x1232 raw-logo path. The RGB565 bytes emitted by that tool are unsuitable for this loader. The XRGB target is 2,799,104 bytes, SHA-256 e2bb58206db1083d8078b5ea8cd83d86569eca9aca950b8d60bac9432e3ebd60. Transfer log records the board-side digest check before installation. The capture script uses the repository’s serial Session helper to reboot and stop at U-Boot, without saving an altered environment.

Serial transcript records the direct XRGB logo load and UBOOT_MOTION_TARGET_READY before the camera starts. The video and physical frame at five seconds show the target while U-Boot is held. Starting kernel occurs after recording ends. Camera focus was 90, manual exposure 200; camera settings record the full controls. This is a warm reset with USB attached.

Analysis command, exit 1:

python3 tools/panel-measure.py --video docs/evidence/splash-uboot-motion/20260922T224020Z-uboot-motion-target.mp4 --label 'U-Boot 61dcc90 candidate, fixed ArUco registration'

Focus investigation

The same generated target PNG was displayed through a temporary Linux swaybg process while the terminal and bar were hidden. This is an optics check, not another U-Boot measurement. The fixed-focus stills detected:

See coarse detection, fine detection, and focus-*.png. No tested fixed focus resolved three markers. The steep view and reflection over the far-right marker limit registration. Combining positions from changing focus settings would introduce a different calibration and is not used to manufacture a passing result. The next valid measurement needs a less oblique, less reflective physical camera/board arrangement.

Even a registered result would be in panel rows. The old dsi-hsfreqrange-hardcoded.md A/B table is in raw camera pixels under a fixed earlier geometry. They cannot be numerically compared. A new Linux 0x87 reference with the same target and rig is needed for a like-for-like comparison; no PHY sweep is justified by this failed optical registration.

The keyboard at focus 30 shows lower-panel key rows and labels more clearly than the original focus 90 recordings. Focus 50 retains more terminal detail but less sharp keyboard text. This supports a follow-up keyboard visibility recording, not finger-input accuracy or a calibrated full-panel geometry claim.

Restored state

The serial transcript records restoration of the immutable logo with SHA-256 6c7a36086297597b359657ab53925ee0725e5201461484d543cbd950f5baa3af and active shell, seatd and firewall. The focus scripts stop only their own temporary swaybg, restore the bar, and bring back the terminal. The follow-up keyboard-visible-layout.txt in ../splash-initial-scene-ready/ restores the terminal to tiling after scratchpad use. Camera focus/exposure return to 90/250. Keyboard visibility signals and cleanup are recorded in keyboard-optics-show.txt and keyboard-optics-restore.txt. No home backup, application default, or persistent boot configuration changed.