LILYGO T-Display-K230 · specification ledger

80 of 281 requirements are unverified

On this board the documentation and the hardware disagree routinely. A datasheet described a working radio whose enable line is never driven. So a requirement counts as grounded only when something was observed on the board, or vendor source was read and cited by path — never because a datasheet says so.

201 grounded
80 unverified

Progress · September 23, 2026

What has changed lately

Installed and verified

Room for apps and updates

The system expanded its compact root filesystem on the physical 128 GB card, leaving about 119 GB free. A second boot preserved the capacity, boot files, shell and Wi-Fi.

Storage evidence

Experimental shell · measured on board

Finding the card rendering cost

A new profile attributes about 98% of repaint CPU to constructing and painting frames. After correcting a test setup race, all 13 injected interaction checks passed again. A scaled-pixel cache passed interaction checks but increased CPU cost in its board comparison, so it stays disabled. Performance and real-finger acceptance remain open.

Repaint profile and remaining performance gates Corrected test and board repeat Cache comparison and decision

Experiment complete

CPU vectors tested on the board

Vector state and 192 pixel comparisons passed. Card performance gains were small and mixed. Work is underway to bring the proven vector path into the normal image; the candidate still needs its own boot and recovery checks.

Measured decision

Next: reduce card repaint cost, verify the feel with a finger, and refine the shell, launcher, settings and notifications into a coherent webOS-inspired interface. That interface is still in development. Try the gesture and motion study · See the proposed screens · Read the design overview.

The requirement count tracks archived specifications. This progress summary and the feature demonstrations also report verified work in changes that remain open. Landing a proposal does not make it an installed feature.

project work · dated snapshot

What is next?

The public work board separates planned changes, implementation, review and evidence gates, and archived work. It shows task counts and distinguishes landed source from device proof.

Open the project work boardproposals, next actions and evidence →

handheld shell · 2026-09-22

The shell responds to a hand on the glass.

A portrait NixOS session with an on-screen keyboard, desktop application launcher, window controls and system actions. Real-touch recordings now show typing, application interaction and a reboot back into the shell. Home recovery is confirmed by the operator. The shell and launcher are archived, with the remaining drag verification gap visible in the ledger.

Keyboard, apps and reboot: screenshots and videos by feature handheld demonstrations →

the screen · 2026-09-21

The panel lights, and the touch controller reports.

The AMOLED emits under Linux, the kernel console renders on the glass at 71×77 characters, and the GT9895 delivers multitouch to /dev/input/event0. Both faults were the same shape: a driver that reported itself healthy while producing nothing. The panel had been sent a different panel’s initialisation sequence — correctly transcribed, from a function with no callers — and never told to leave zero brightness. The touch controller had its interrupt declared edge‑triggered on a part that holds the line low until it is acknowledged: 0 interrupts against 2173.

The board's AMOLED panel showing the Linux kernel console, photographed off the board.

docs/evidence/panel-photos/05-console-on-panel.jpg — photographed off the board, not rendered. Five more photographs are committed beside it.

51 FE — SET_DISPLAY_BRIGHTNESS A DCS‑brightness AMOLED powers up at zero, so it can be fully initialised, accept every command, and emit nothing.
0 → 2173 interrupts Same kernel, same image, one device‑tree word: IRQ_TYPE_EDGE_FALLING to IRQ_TYPE_LEVEL_LOW.
a patch that blanked it 25% DSI burst headroom, built and never looked at. The persistent journal dated the regression to the image. Reverted, remeasured, and the brightness flicker largely went with it.
panel and touch requirements archived Panel and touch bring-up are now included in the published ledger. Boot-splash handoff remains a separate change in progress.

The photographs, the dead function, and the clock that blanked it panel note → Why the controller was awake and silent for a day touch note →

first boot · 2026-09-20

NixOS reached a root shell on the silicon.

Not under emulation, not on the build host. A kernel this repository cross-compiled, an image this repository wrote, an SD card in the board — and a prompt on /dev/ttyACM0 that answered four questions no x86 machine and no qemu-system-riscv64 -machine virt could have answered the same way.

[root@nixos:~]# cat /proc/cpuinfo
processor	: 0
hart		: 0
isa		: rv64imafdcv_zicbom_zicboz_zicntr_zicsr_zifencei_zihpm_zba_zbb_zbs_svpbmt
mmu		: sv39
uarch		: thead,c908
mvendorid	: 0x5b7
marchid		: 0x8000000009140d00
mimpid		: 0x50000

[root@nixos:~]# uname -a
Linux nixos 6.6.36 #1-NixOS SMP Tue Jan  1 00:00:00 UTC 1980 riscv64 GNU/Linux

[root@nixos:~]# cat /proc/cmdline
console=ttyS0,115200n8 root=fstab loglevel=4 lsm=landlock,yama,bpf init=/nix/store/2h0xrlf5n74fjjhxqd18y6vl55w4zyrr-nixos-system-nixos-26.11.20260919.20b1ddd/init

[root@nixos:~]# findmnt -no SOURCE,FSTYPE /
/dev/mmcblk1p2 ext4

[root@nixos:~]# nproc
1

thead,c908 · 0x5b7 T‑Head’s vendor ID and the C908 microarchitecture. This is the real K230D, not a host and not virt.
6.6.36 #1-NixOS The Xuantie kernel built by this flake, running as a NixOS system — not the vendor image the board shipped with.
init= arrived intact /proc/cmdline is exactly the string the SD image writes to bootargs.txt. Making that true took a fix in the U‑Boot environment.
root=fstab resolved No explicit root=, yet / is /dev/mmcblk1p2. The NIXOS_SD label path had never been exercised before; it works.

The full transcript, and why Linux gets one hart first-boot note → How the kernel command line got through boot-chain note →

Capabilities

Hardware notes

Hand-authored rather than rendered from the specs, and maintained here rather than anywhere else. They are split by where their facts came from, because that is what decides how much they are worth: a note written from a console transcript outranks a note written from a device tree, and both get corrected in place when the repository learns something newer.

Observed on the board

Written from console transcripts. The board did these things and they were captured.

Read out of the files

Written from the device trees, the schematic and the vendor BSP. Correct descriptions of source, which is not the same as working hardware.

Evidence on file

Committed in the repository and rendered here in full, so a boot log or a photograph is readable without a checkout. An observation on the board outranks vendor source; a datasheet grounds nothing. The two lists are kept apart because they carry different weight.

grounding a requirement what the count above rests on

cited by a hardware note real, and holding up no requirement yet

docs/design/handheld-prototype/README.md 3.5 KiB
docs/design/handheld-shell/README.md 3.7 KiB
docs/design/handheld-shell/surfaces.svg 8.4 KiB
docs/evidence/card-shell/injected/two-live.png 29.3 KiB
docs/evidence/card-shell/mirror-order/board/README.md 3.5 KiB
docs/evidence/card-shell/render-cost/board/README.md 3.1 KiB
docs/evidence/card-shell/repaint-stages/README.md 6.6 KiB
docs/evidence/card-shell/scaled-cache-board/README.md 6.8 KiB
docs/evidence/card-shell/throw-fixture-sync/README.md 6.0 KiB
docs/evidence/card-shell/throw-sampling/README.md 6.3 KiB
docs/evidence/card-shell/touch-timestamps/board/README.md 2.6 KiB
docs/evidence/gt9895-port.md 17.2 KiB
docs/evidence/launcher-gestures/edge-states/README.md 1.7 KiB
docs/evidence/launcher-gestures/integrated-injected/overview.png 31.3 KiB
docs/evidence/launcher-gestures/real-finger/gestures.mp4 3717.7 KiB
docs/evidence/launcher-gestures/real-finger/window-selection.jpg 27.6 KiB
docs/evidence/network-video/installed-app/README.md 3.5 KiB
docs/evidence/offline-wifi-image/editor.png 25.0 KiB
docs/evidence/offline-wifi-image/help.png 46.9 KiB
docs/evidence/offline-wifi-image/nnn.png 17.1 KiB
docs/evidence/panel-dark.md 6.7 KiB
docs/evidence/panel-photos/01-urandom-noise.jpg 24.6 KiB
docs/evidence/panel-photos/02-all-black.jpg 20.9 KiB
docs/evidence/panel-photos/03-all-white.jpg 23.9 KiB
docs/evidence/panel-photos/04-normal-mode-fullscreen-noise.jpg 24.2 KiB
docs/evidence/panel-photos/06-sheared-console-31.68mhz.jpg 62.9 KiB
docs/evidence/rm69a10-init-sequence.md 8.3 KiB
docs/evidence/rtsmart-panel-bringup.md 27.3 KiB
docs/evidence/shell-real-touch-apps/after-test-native.png 117.3 KiB
docs/evidence/shell-real-touch-apps/physical-apps-menu-29s.png 476.2 KiB
docs/evidence/shell-real-touch-apps/video-audit.md 1.7 KiB
docs/evidence/shell-real-touch-keyboard/after-test-native.png 41.5 KiB
docs/evidence/shell-real-touch-keyboard/keyboard-toggle-cropped.jpg 25.0 KiB
docs/evidence/shell-real-touch-keyboard/visibility-followup-audit.md 1.2 KiB
docs/evidence/shell-real-touch-system/after-reboot-native.png 16.1 KiB
docs/evidence/shell-real-touch-system/system-post-reboot.png 265.7 KiB
docs/evidence/shell-real-touch-system/video-audit.md 2.3 KiB
docs/evidence/storage-capacity/README.md 7.5 KiB
docs/evidence/touch-probe.txt 8.5 KiB
docs/evidence/uboot-env.txt 5.6 KiB
docs/evidence/vglite-scene-board/context-lifetime/README.md 5.2 KiB
docs/evidence/video-acceleration/360p-native.png 222.3 KiB
docs/evidence/video-acceleration/360p-physical.jpg 15.7 KiB
docs/evidence/video-acceleration/360p-physical.mp4 407.9 KiB
docs/evidence/video-acceleration/README.md 7.8 KiB
docs/evidence/wifi-persistent/README.md 2.1 KiB
docs/research/handheld-ux/webos-comparison.md 2.1 KiB
docs/thermal.md 5.2 KiB