# Arduino + ESP32 Bench — log

One entry per session: date · what changed · what is next.

- 2026-11-14 scaffolded from templates/PROJECT.*.tmpl

## 2026-11-14 — assistant built, toolchain in place

- Registered in `projects/index.json` + `PROJECTS.md` and in `hermes project`.
- Auto-load proven from a fresh process (`hermes -z`): it recited the objective,
  the rules and the next step from this folder's `.hermes.md` alone.
- **Toolchain installed and verified on the 3050** (no sudo needed — single static
  binary under `$HOME`): `arduino-cli` 1.6.0 in `~/.local/bin`; cores `arduino:avr`
  1.8.8 (Uno/Nano/Mega), `esp32:esp32` 3.3.12 and `esp8266:esp8266` 3.1.2. Board-manager
  URLs registered for all three. Verified by compiling a plain blink for each: Uno
  **924 bytes / 2%**, ESP32 **271,440 bytes / 20%**, ESP8266 generic
  (ESP-12x) **232,148 bytes / 22%**.
- `mike` was added to the **`dialout`** group in the same session that remounted the
  T130 share. Takes effect at the next login — the running Hermes process does not
  have it yet.
- **Upload/flashing is still UNTESTED.** No board has ever been attached to the 3050.
  The honest claim is "this compiles on the 3050", never "this runs".
- Rules encoded from Mike's own answers: the main job is **write AND explain**; the
  parts are **an unsorted shelf** (never assume a labelled kit, never ask for a photo);
  boards flash **from the 3050**.
- **Next:** the parts inventory is **done** (2026-11-14) — ESP32 DevKit V1, Elegoo Uno R3,
  Arduino Mega 2560, an ESP32-WROOM-32 module, plus a Pi Zero W and a Libre Renegade filed
  separately as Linux computers, out of scope. All of it is in `docs/parts-inventory.md`,
  and the toolchain is compile-verified for every family: Uno 924 B (2%), Mega 1,536 B,
  ESP32 271,440 B (20%), ESP8266 232,148 B (22%).
- **Next step, and the only unproven thing:** the **upload**. No board has ever been attached
  to the 3050. Plug one in — the Elegoo Uno R3 is the simplest — and flash a blink with
  `arduino-cli upload`, which also confirms the `dialout` group fix took effect.
- **Standing preference added to the rules (rule 10):** Mike has a pile of add-on cards and
  shields and said, in his words, *"those I guess we should do as I need them"*. So they are
  **not** inventoried up front — identify one at a time, only when a build depends on it.
  Do not open with a shelf audit; that is now explicitly off the table.

### Upload attempt — FAILED, and what it proved (2026-11-14, ~19:15–19:40)

Mike plugged the **ESP32 DevKit V1** into the 3050. It never appeared. Recorded here so the
next session does not repeat the same 25 minutes.

- **Cross-checked both machines:** the 3050's USB bus showed nothing but its own root hubs, and
  the 9010 showed only its keyboard/mouse/wifi/bluetooth. `arduino-cli board list` → "No boards
  found". No `/dev/ttyUSB*` or `/dev/ttyACM*`.
- **The board is powered:** red (power) LED lit, and the blue GPIO2 LED too. So it is alive and
  getting 5 V.
- **The 3050's ports are GOOD** — this is the important control. At 19:19 a USB keyboard was
  plugged in and the kernel enumerated it immediately (`hid-generic`, `/dev/input/input13-16`).
  A port that enumerates a keyboard is not a broken port.
- **Two different cables** were tried, with the same result, and a 20-minute watch on the USB bus
  (polling for a tty or for `ch341`/`cp210x`/`ftdi_sio`/`cdc_acm` to load) saw nothing at all.
- **Conclusion:** powered board + proven-good port + two cables + zero kernel events = the failure
  is on the **board's own USB side**, i.e. its USB-serial chip or the connector's solder joints.
  Not the port, and probably not both cables.
- **Still to try, cheap:** a *proven* data cable (one that makes an old phone or camera appear),
  and wiggling the plug in the board's socket in case it is a dry joint.
- **Recovery route if the chip really is gone:** the ESP32 itself is fine — flash it through an
  external **3.3 V USB-serial adapter** wired to TX/RX/EN/IO0. Worth digging out *only* if needed.

**So the upload remains unproven, and the ESP32 is no longer the board to prove it on.** The
**Elegoo Uno R3** is the better next candidate: simplest board, no boot-mode tricks, CH340.

### Context, for a later session

- This project was created while the T130 share was down: the `.automount` unit had
  latched off (`mount-start-limit-hit`) after the T130's network dropped at ~05:24 UTC.
  Mike remounted it with `sudo bash ~/fix-share.sh` (written for the occasion).
- The rclone Google Drive token died the same day (`invalid_grant`), so the hourly
  `3050-to-Drive` bisync and the 4am/5am backups were failing until re-auth.
