IO diagnostics · round 2

Plug the rack in and the panel says what is connected, what isn't, and what to check, in plain words. Printer-style utilities wrap your d_* tools, and every button dry-runs first.

io-diagnostics main @ a6853bde (local repo) · drivers on bam_ws claude/traj-tracking-sim-to-real-a3da60 @ bb393a35, not merged · live captures 2026-10-22T19:35:23-07:00, real rack, read-only. No button that writes or moves was pressed.

One frame from the hero video. The video itself is left out of the public copy.

A still from the middle of the video. The video is not in the public copy: its frames show the machine's name and device serials.

Hero: the live dashboard on this PC.

fdcanusb XXXXXXXX holds CAN 1 / 14 / 15.
CAN 1 → Run calibration… dry run, showing the exact command and preconditions.
The confirm modal won't enable until you tick "joint is free".
Cancel.
The second adapter XXXXXXXX: "frames never reached the bus" and what to check.

The rack, live

tree-light — Default view (structure P1, tree v6, skin Claude): entities only. The adapter holds its boards; the computer is not a fold.

adapter-28B3A25D — The second adapter: on USB, but its frames never reach the CAN wire (8,192 broadcast queries left lec 7 / dlec 7, err 0). The panel says so and lists what to unplug and measure.

adapter-272813EF — fdcanusb XXXXXXXX selected: the Utilities card, printer-style. Checks, then Maintenance, then Motion set apart.

can1-calibrate-dry-run — CAN 1 → Run calibration…: the dry run shows the exact moteus_tool command, each precondition with its state, and the confirm text naming CAN 1's hard stop and gravity load.

can1-calibrate-modal — Motion needs a ticked physical precondition before the button enables. The capture pressed Cancel.

table-light — The table uses the same rows as the tree, and the Message column no longer collapses (your screenshot showed “ME”).

Every button, and what it would run

These are real dry runs through the host (io-diagnostics act … --dry-run) against this PC's adapter and camera. Nothing ran. Same name = same meaning on every driver (docs/actions-vocabulary-v0.1.md).

Button	Danger	Exact command its dry run reports	Replaces
Check communication fdcanusb XXXXXXXX	query	`py3 -m moteus_servo.diagnostics --bus-lock ~/.cache/rack-commissioning/bus.lock --firmware-cache ~/.cache/moteus-firmware --inventory ~/archive/datasets/bam_rig/rack_inventory diagnose --device /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00 --range 1 126`	d_find + d_triage
Check firmware fdcanusb XXXXXXXX	query	`py3 -m moteus_servo.diagnostics --bus-lock ~/.cache/rack-commissioning/bus.lock --firmware-cache ~/.cache/moteus-firmware --inventory ~/archive/datasets/bam_rig/rack_inventory diagnose --device /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00 --queries 0 --no-config`
Stop all fdcanusb XXXXXXXX	interrupts	`py3 ~/bam_ws/src/core/mdp/agent/actuators/moteus_servo/utils/moteus_with_targets.py ~/.local/bin/moteus_tool -t 1,14 --stop --fdcanusb /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00`	d_stop
Open console fdcanusb XXXXXXXX	interrupts	`py3 ~/bam_ws/src/core/mdp/agent/actuators/moteus_servo/utils/moteus_with_targets.py ~/.local/bin/tview -t 1,14 --fdcanusb /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00`	d_tview
Update firmware… fdcanusb XXXXXXXX · CAN 1	changes	`py3 a private note --target 1 --port /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00 --cache-dir ~/.cache/moteus-firmware --tool ~/.local/bin/moteus_tool`	the update-moteus-firmware skill
Run calibration… fdcanusb XXXXXXXX · CAN 1	moves	`~/.local/bin/moteus_tool --target 1 --calibrate --fdcanusb /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00`	d_calibrate
Update configuration… fdcanusb XXXXXXXX · CAN 1	changes	`~/.local/bin/moteus_tool --target 1 --restore-config ~/.local/share/moteus-config/can-1.cfg --fdcanusb /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00`	d_config (its script is gone; stock moteus_tool --dump-config / --restore-config)
Clear faults fdcanusb XXXXXXXX · CAN 1	changes	`py3 -m moteus_servo.diagnostics --bus-lock ~/.cache/rack-commissioning/bus.lock --firmware-cache ~/.cache/moteus-firmware --inventory ~/archive/datasets/bam_rig/rack_inventory act --device /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00 --action clear_faults --node 1`
Reboot… fdcanusb XXXXXXXX · CAN 1	interrupts	`py3 -m moteus_servo.diagnostics --bus-lock ~/.cache/rack-commissioning/bus.lock --firmware-cache ~/.cache/moteus-firmware --inventory ~/archive/datasets/bam_rig/rack_inventory act --device /dev/serial/by-id/usb-mjbots_fdcanusb_272813EF-if00 --action reboot --node 1`
Change CAN id… fdcanusb XXXXXXXX · CAN 1	changes	`ran in 0.098 s · 1 lines`	d_reassign
Check communication RealSense D405 · 6036	query	`pyrealsense2.context().query_devices() device.get_info(pyrealsense2.camera_info.<field>) for <field> in (name, serial_number, asic_serial_number, firmware_version, recommended_firmware_version, usb_type_descriptor, physical_port, product_id, product_line, firmware_update_id, connection_type), skipping any the device does not support`
Check firmware RealSense D405 · 6036	query	`pyrealsense2.context().query_devices() device.get_info(pyrealsense2.camera_info.<field>) for <field> in (name, serial_number, asic_serial_number, firmware_version, recommended_firmware_version, usb_type_descriptor, physical_port, product_id, product_line, firmware_update_id, connection_type), skipping any the device does not support`
Reboot… RealSense D405 · 6036	interrupts	`intel_rgbd.real.RealSenseCamera('XXXXXXXXXXXX').hardware_reset()`
Update firmware… RealSense D405 · 6036	changes	`exited 3 after 0.1 s`
Run calibration… RealSense D405 · 6036	changes	`pyrealsense2.context().query_devices() device('XXXXXXXXXXXX').as_auto_calibrated_device().run_on_chip_calibration('{"speed": 3, "scan parameter": 0}', 120000) set_calibration_table(table) then write_calibration(), only when \|health_before\| >= 0.25 and \|health_after\| < \|health_before\|`
Open console RealSense D405 · 6036	interrupts	`/usr/bin/realsense-viewer`

Five ways to see the entities

All six are live in the bottom-right Structure switcher, on the same live data. P1 is the default because it is literally “fdcanusb → Moteus c1 · CAN 15” plus a Ports toggle for the USB / PCIe detail. The single computer is never a fold in any of them.

structure-p1 — **p1 · Devices, transport as columns (default)** — Only the things you plugged in: the adapter holds its boards. The USB / PCIe path is a quiet Connection column and tooltip; the Ports toggle brings the layers back.

structure-p2 — **p2 · Two views, one toggle** — Devices by what they are for (Actuators / Cameras / Network / Computer) or by connection, like Windows Device Manager's two views; the same row stays selected across the toggle.

structure-p3 — **p3 · Path on demand** — The entity tree, with the connection path as a breadcrumb under the selected row, a Path column, and a Paths toggle.

structure-p4 — **p4 · Compact folders** — Today's physical tree with every single-child chain joined into one row (VS Code's compact folders); tty and video nodes become attributes.

structure-p5 — **p5 · Bus lens** — The buses first (CAN · fdcanusb XXXXXXXX, USB cameras, PCIe, Ethernet), the computer's own topology folded into one Computer row.

structure-p0 — **p0 · Physical (round 1)** — Round 1's physical tree, kept for comparison.

Eleven skins, light and dark

The seven you named, plus four of mine:

ISA-101 HMI: grey, colour only for abnormal. It is the standard behind “quiet when OK”.
macOS Settings.
Linear.
High contrast: AAA.

Every skin and mode passes the WCAG gate in the test, measured: text ≥ 4.5:1 on four surfaces, and problem shapes ≥ 3:1 on the surface and the selected row. OK rows stay uncoloured. Claude is the default.

skin-claude-light — **Claude (default)**

skin-foxglove-light — **Foxglove / medkit**

skin-foxglove-dark — **Foxglove / medkit**

skin-unstyled-light — **Unstyled (bring your own)**

skin-unstyled-dark — **Unstyled (bring your own)**

What the live rack caught

Our own checks raced for the rack bus lock: one camera reported "lock held". Plugins sharing a gate now queue.
Config reads failed or garbled values: the inventory and config check read raw streams past the frame guard. Now paced; CAN 1 and 14 read every key and firmware 657b5c5, read-only.
Firmware 1.1.2 has no servo.derate_temperature: every board read "unknown". Now "not on this firmware", and derating comes from fault_temperature − temperature_margin.
A silent adapter was blamed on "frames are ACKed": now FramesNotSent when dlec stays 7.
Your d_* aliases point at a moved folder: src/core/mdp/actuators → src/core/mdp/agent/actuators.
d_find --range re-launched the triage script: fixed.
A silent board printed "MISMATCH kp, ki, kd…": now "NO REPLY".

Decisions

Done and proved

The live panel shows both adapters, CAN 1 / 14 / 15 and both D405s, with plain-words reasons (captures above, real rack).
Every utility dry-runs to the exact command (the table above). Gates: 876 tree/skin checks with mutations caught, 118 view-core, 37 utilities, 182 host pytest, 687 moteus_servo, 163 intel_rgbd, 275 rack twin.

Needs you

Pick a structure and a skin. Default if you say nothing: P1 + Claude.
XXXXXXXX: unplug its CAN cable at the board end, then press Check communication. The panel says whether the board or the adapter holds the bus.
CAN 1 now reads ratio 0.1, but the record holds 1. Say “record it” and I run config-record --can-id 1. CAN 1 motion stays refused until then.
Update configuration has no recorded per-board .cfg files yet, so it refuses with a dump command. Default: I seed them from the 10-10 / 10-19 firmware-run dumps when you say so.

Deliberately not done

No write or motion button was pressed by any agent: Stop all, firmware, calibration, config and ids are yours to click.
Nothing is merged; the drivers stay on their branch.