Add wiring doc with the series-LED failure mode

Records the diagnosis: green only lights when red is on, because the green
return reaches the red LED's anode instead of the ground rail.

docs/WIRING.md

@@ -0,0 +1,69 @@
+# Wiring
+
+![Wiring pinout](wiring.svg)
+
+## Board to board
+
+| Board A (esp32_bluepad32) | | Board B (esp32 2.0.17) |
+| --- | --- | --- |
+| GPIO17 TX | → | GPIO16 RX |
+| GPIO16 RX | ← | GPIO17 TX |
+| GND | — | GND |
+| 5V | — | 5V (model setup only, see README) |
+
+TX goes to RX, not TX to TX. The ground is a shared voltage reference so the
+two boards agree on what a logic 1 is — without it you get garbage or silence.
+It is not a return path for anything else, so do not hang LED current off it.
+
+## LED circuits
+
+Both live entirely on **board B**. Each is its own complete loop:
+
+```
+GPIO25 ──[220Ω]──▶|── RED pair ────┐
+                                   ├── ground rail ── board B GND
+GPIO26 ──[100Ω]──▶|── GREEN pair ──┘
+```
+
+The two returns meet **only at the ground rail**, and the rail goes straight to
+a GND pin on board B. LED long leg (anode) faces the resistor and GPIO side.
+
+Resistor values are a starting point: roughly 220Ω for red, 100Ω for blue or
+white. A pin is comfortable at about 12mA and hard-limited near 40mA. If a pair
+draws more than ~20mA — which pre-wired "3V" LEDs sized for a coin cell often
+do — drive it through a small NPN transistor or MOSFET instead of off the pin.
+
+## Known failure mode: LEDs wired in series
+
+Symptom seen on this build — green only lights when red is already on, and
+green never lights alone:
+
+| Software state | Observed |
+| --- | --- |
+| red on, green off | red on ✓ |
+| red off, green on | **both off** ✗ |
+| red on, green on | both on ✓ |
+| red off, green off | both off ✓ |
+
+Cause: the green return lands on the *anode* side of the red LED rather than on
+the ground rail, putting the two LEDs in series. Green alone then has to push
+through both — roughly 2.1V plus 1.9V — and a 3.3V pin cannot do it, so neither
+lights. With GPIO25 also high, the junction is lifted and both conduct.
+
+Two rows of that table are impossible from a software fault, which is what
+identifies it as wiring.
+
+**Test:** with both LEDs off, meter from the shared return wire to a GND pin.
+It must read 0V. Press Circle and measure again — if it rises above 0V, that
+wire is not ground.
+
+**Simplest fix:** give each LED its own separate wire back to its own GND pin
+on board B, with no shared return at all. Board B has several grounds, all
+common internally. Slightly less tidy, and it removes the junction entirely.
+
+Worth checking polarity at the same time. A reversed LED simply stays dark, so
+it looks like a dead circuit rather than a wiring error.
+
+---
+
+Created by: Jess Rogerson (yelling commands at Claude.AI)