[luismiKode]

Three fixes from tonight's runs of the end-of-line station (E5) with a DVT in the fixture, and the SDK branch that carries four more.

fix(kgui): Take an icon's PSRAM when its row first has an icon

The launcher took its 128 icon slots at start: 6 MB of PSRAM held for good since the icons went to 128 px, where it was 1.5 MB. The station's camera part takes 18 MB when it wakes and the OV5640 another 4.9 MB for its frame buffers; on a fresh boot 4.82 MB were left, the camera never opened ("no PSRAM for 3 frame buffers"), the placement ended as a badly seated Kode and the memory row failed. A slot is now taken the first time an icon is written to its row, and kept; a row whose slot cannot be had draws the apps glyph, as before.

On the board: 27.3 MB free at rest where there were 23.0, 9.2 MB after the station's wake where there were 4.82; placement and the six camera and LED rows PASS. The launcher with four apps installed shows their icons (an app with no icon of its own keeps its letters); two captures of it are with the station's validation notes. There is no capture of the same board before the change.

fix(tools): Wait for the C5's link before the station's --discover sweep

--discover asked for the C5 as soon as the port opened, which resets the board, and found it not linked yet: radio ok=0 why=link and no channels. It waits for the link first: radio ok=1 seen24=1 seen5=1 with the channels written.

fix(tools): Make the station's NFC and RFID steps prove a write of this run, and count NFC

The RFID step wrote the ID made from the MAC and passed when what it read was what it wrote: the same Kode back at the station finds its ID already on the tag and passes with a dead write. The Kode's own match=yes was printed and not used, and with an unreadable MAC any ID passed. The NFC step wrote the serial the Kode's own row had just left on the tag. Both steps now write something else first (the ID with its bytes reversed; "KODE E5") and then what counts; RFID asks for match=yes on both writes, with one retry a write; an unreadable MAC is a FAIL; and NFC is no longer optional.

The bench operator presses at most one wrong button a pass (it could press three and fail a good board).

fix(tools): Hold the station's NFC and RFID steps until both of the Kode's rows are in

With the two writes in, the PC's NFC step failed 3 passes of 11 when it ran beside the operator's rows. It was not the buttons: the step started as soon as the Kode's own NFC row answered, and the Kode was then running its RFID row, with the 125 kHz carrier up on an antenna that sits beside the NFC one. The PC's two steps now wait until both of the Kode's rows are in, and run beside the operator's rows again: 20 passes with the writes on and no failure of either.

chore(sdk): the RFID decoder, injected buttons and touch, the sd row, the C5 card's bus, the mag row and the NFC reader's SPI

The pointer is on kode-sdk's fix/station-rfid-and-inject (kodediy/kode-sdk#127), at 10ca19d. It has to move to develop's commit once that one is merged; #66 moves the same pointer to e446c67, which this branch sits on.

What was run

• The build with its four checks: size (7 % of the slot left, the warning develop already gives), internal RAM 409 356 of 417 000, starts, MISRA with none past the baseline.
• tools/factory's own tests with the change (test_factory.py, test_dot_parallel.py).
• tools/tool_tests.sh: all but test_kble.py, which fails on this Mac for macOS's Bluetooth permission and is not touched here.
• On a DVT in the E5 fixture, the station's programme with --operator-sim and the NFC and RFID writes on. With this branch's binary (sha256 4e232fcd…a971): 20 of 20 passes with all 29 rows PASS, 27 to 31 s, with the PC's NFC and RFID steps beside the operator's rows. Earlier in the night, with the commits up to the NFC and RFID writes: 3 of 3 passes; and with the first two only: 45 passes in which the faults they fix did not come back. Not run in any of them: the four pad rows (charge, delivery, GND, data loop).

🤖 Generated with Claude Code

https://claude.ai/code/session_01UAYt4sNGU1gkMrcPpLheZ3

[github-actions]

kodeOS build

Check	Result
Image size	kodeos.bin 3872768 of 4194304 bytes, 321536 left (7%)
Internal RAM	kodeos.elf 35 statics of 2048 bytes or more in internal RAM, all listed
Starts	starts: 23 Active Objects and 6 door set-ups, each beside what it starts
MISRA	MISRA: 1823 findings in 187 files, none past the baseline

::warning::kodeos.bin is within 10% of its ceiling (321536 bytes, 7% left of 4194304)