# JLCPCB order files — kicad_mainboard

## Current state (2026-11-10)

Read this first. The sections below are a dated log; where one contradicts this summary, this summary is right.

- **Copper: order at 1 oz** (JLCPCB's default) since 2026-11-10; the 2 oz notes further down are history. The ring current is held to 4.5 A by the LED engine and fused at 7.5 A in the harness; the narrowest point of the 5 V path (6.5 mm) carries about 9.2 A at a 10 °C rise at 1 oz (IPC-2221).
- **3.3 V:** no regulator on the board. The ESP32-S3-ETH's own JW5060 buck makes `+3V3` from `VSYS` (`J1-H1` pin 2), fed from `+5V_LED_LOGIC` through `D33`; `3V3_EN` (pin 4) is open. `U2` (K7803-2000R3), `C12`, `C13` and `C14` are gone.
- **Expander:** `U1` is a TCA9555PWR (TSSOP-24, **C465732**; UMW C42420608 as an alternate), same address 0x20, same board pins. Its CPL rotation (+270, as the old SOIC) is set from JLCPCB's usual TSSOP drawing because the EasyEDA lookup failed on the day: check it in the placement preview.
- **Ring power:** `J5` is a JST VH 4-pin vertical (B4P-VH(LF)(SN), **C160317**): 1-2 `+5V_LED_RING`, 3-4 GND, mating the ring's B4PS-VH through a VHR-4N cable.
- **Ring signal and temperature:** `J-LED-SIG` is a JST XH 5-pin (B5B-XH-A, **C157991**): 1 DATA, 2 GND, 3 CLK, 4 GND, 5 `NTC_RING_J`. New: `D34` (PESD3V3S1UL, C70371), `R65` (10k 1 %, C25804, pull-up to +3V3), `R66` (1k, C21190) and `C16` (100 nF, C14663) into GPIO3.
- **Pin moves:** GPIO3 = ring NTC (`RING_TEMP`), GPIO40 = hue encoder B (`HUE_DT`), main encoder button (`ENC_SW`) on `U1` IO0_3.
- **Part counts** (from these files, 2026-11-10): 86 placed SMD parts (`Mainboard_BOM_SMD.csv`, 37 lines) and 104 for full assembly (47 lines). Solder joints: 254 (SMD-only) and 370 (full).
- **Fuses:** `F1` is one 2 A resettable 1812 fuse, RUILON SMD1812P200TF16, LCSC C20812; the 5 V side has a 7.5 A ATO fuse in the harness at the supply.
- **CPL positions fixed (2026-11-10):** JLCPCB places each part by its own footprint's origin (the pad centre), but KiCad's connector footprints have theirs on pin 1, so every header showed up half its length off in JLCPCB's placement preview (the 1x20 sockets by 24 mm). `tools/check_cpl_rotation.py` now also writes `cpl_position_offsets.json` (the move from KiCad's origin to JLCPCB's, for the 18 connectors and `Q6`), and `tools/build_jlc_assembly.py` adds it to the CPL. `Q5`'s turn (+270) is set by hand in that tool because its pad numbers differ from JLCPCB's footprint.
- **Still to check at upload:** `F1`'s stock and library type, and the orientation of `Q5`, `C6`, `C11` and `U1`, and the position of every connector, in JLCPCB's preview.

## 2026-11-10 — U2 removed (3.3 V from the module), TCA9555, JST VH ring power, ring NTC input, 1 oz

- **Removed:** `U2` (K7803-2000R3), `C12`, `C14` and the unfitted `C13`, with their copper. `J1-H1` pin 2 (`VSYS`) is fed from `+5V_LED_LOGIC` by a 0.5 mm track off the `D33` lane; pin 4 (`3V3_EN`) is open; pin 5 (`3V3`) is now the `+3V3` source (a `PWR_FLAG` on `+3V3` in the schematic).
- **`U1`:** MCP23017-E/SO (SOIC-28W, C47023, about $1.25 and 900 in stock) -> TCA9555PWR (TSSOP-24, C465732, $0.33-0.70, 26.6k). The KiCad symbol puts every used pin where the MCP23017's was; `~RESET`'s tie went, `~INT` is open. Its 19 connections were re-routed with 0.2 mm tracks; each GND pin has its own via, none in a pad.
- **`J5`:** Mini-Fit 2x3 (C20539148) -> JST VH B4P-VH (C160317), moved 2.6 mm down to clear `D5`; pin labels updated.
- **Ring NTC:** `J-LED-SIG` 4-pin -> 5-pin (pins 1-4 unchanged); `D34`, `R65`, `R66`, `C16` below it; `RING_TEMP` routed to GPIO3, `HUE_DT` re-routed from `R33` to GPIO40, `ENC_SW` from `R29` to `U1` pin 7.
- ERC 0; DRC with schematic parity 0 violations, 0 unconnected, 0 parity issues (warnings included). Gerbers, drill, BOM/CPL, renders, schematic PDF and reports regenerated. Not tested on hardware.

## 2026-11-09 — single-chip LED drive: WLED DevKitC-1 removed, clock buffer and ring signal header added

- **Removed:** `J6-J1`, `J6-J3` (the DevKitC-1 sockets). The ESP32-S3-ETH now drives the ring itself (in-firmware LED engine): GPIO18 = data, GPIO2 = clock, the two pins that used to be the WLED serial link (nets renamed `LED_DATA_3V3` / `LED_CLK_3V3`; `TP8` stays on the data line).
- **Data:** `R18` (330R) -> `U3` (74AHCT1G125) -> `R63` (33R, new) -> `J-LED-SIG` pin 1. `R26` keeps the input low at boot. `U3`, `R18`, `R26`, `C10` and `TP12` moved into the freed area right of the module.
- **Clock (new):** `R61` (330R) -> `U5` (74AHCT1G125, C7484) -> `R64` (33R) -> `J-LED-SIG` pin 3, with `R62` (10k pull-down), `C15` (100 nF) and `TP13`.
- **`J-LED-SIG`:** JST XH 4-pin (B4B-XH-A, C144395): 1 DATA, 2 GND, 3 CLK, 4 GND. It mates the ring's `J2` in its own cable. **`J5` (Mini-Fit) is power only** now: pin 1 unused.
- **Buffer supply:** `D33` (SS34) now feeds `+5V_LED_LOGIC` (the two buffers) from `+5V_LED_PSU` over the track that fed the DevKitC; it still blocks a reversed supply. The buffers see about 4.8 V, inside the 74AHCT's 4.5-5.5 V.
- ERC 0; DRC with schematic parity 0 violations, 0 unconnected, 0 parity issues. Gerbers, drill, BOM/CPL, renders, schematic PDF and reports regenerated. The ESP32-S3-ETH antenna area is unchanged. Not tested on hardware.

## 2026-10-19 — six Extended parts swapped for Basic equivalents

Checked against JLCPCB's parts library on 2026-10-19; same footprints, so no board change.

| Was (Extended) | Now (Basic) | Used for |
|---|---|---|
| C1590 0.1uF 0603 | **C14663** Yageo CC0603KRX7R9BB104, 50V X7R | `C7`, `C9`, `C10` |
| C602037 22uF 25V 0805 | **C45783** Samsung CL21A226MAQNNNE | `C12`, `C14` |
| C105881 330R 0603 | **C23138** 0603WAF3300T5E | `R18` |
| C98220 10k 0603 | **C25804** 0603WAF1002T5E | `R19`, `R26`, `R60` |
| C99782 4.7k 0603 | **C23162** 0603WAF4701T5E | `R21`, `R22` |
| C22548 1k 0603 | **C21190** 0603WAF1001T5E | `R9`, `R23` |

LCSC's retail stock showed 0 for C21190 on 2026-10-19 while JLCPCB listed 22.7M; confirm at upload. The sections below are dated history and keep the old numbers.

## 2026-10-07 — board re-layout and re-route; `R24`, `R25`, `R28` removed

- **Removed parts:** `R24` (10k, C25804), `R25` (1k, C22548) and `R28` (1k, C22548); `R23` stays, as the ENABLE pull-up. Drop the three from any earlier order.
- **Board:** new placement and routing, new GPIO pin map (see the CHANGELOG), TMC2209 socket rotated 90 degrees, panel headers in a new order. Check the silkscreen names under the panel strip when wiring the panel.
- **BOM/CPL: 73 SMD parts / 198 joints, 97 full-assembly / 392 joints.** Gerbers re-exported (same 12-file set). ERC 0; DRC with schematic parity: 0 violations, 0 unconnected, 0 parity issues.

## 2026-10-06 — motor rail 12 V; `U4`, `Q5`, `R19` removed; `D5` now SMBJ15A

- **Removed parts:** `U4` (the 24 V to 12 V fan converter: the fan now runs straight off the 12 V rail), `Q5` and `R19` (the 5 V
  reverse-polarity FET and its gate resistor: the keyed Mini-Fit Jr connector already prevents reversal), and the optional `C12` pad.
  `F3` stays. Drop `U4` (C5369645), `Q5` (C727764) and `R19` from any earlier order.
- **`D5`** is now an **SMBJ15A**: GOODWORK **C908796** (68,540 in stock, about $0.04 on 2026-10-06; confirm Basic/Extended at upload), alternate
  DIODES SMBJ15A-13-F C135046 (3,095 in stock). 15 V standoff, 16.7 V minimum breakdown, 24.4 V clamp.
- **Supply:** the motor/logic input on `J-PWR-IN` pin 5 is now **12 V** (silkscreen *+12V*). Do not connect a 24 V supply.
- **BOM/CPL: 76 SMD parts / 204 joints, 100 full-assembly / 398 joints.** Gerbers re-exported (same 12-file set). ERC 0; DRC with
  schematic parity: 0 violations, 0 unconnected, 0 parity issues.

## 2026-10-05 — fan header now 12 V (`U4` replaces `F5`)

- **`U4`** is now the YLPTEC K7812-500R3 (**C5369645**, JLCPCB Extended, 1,598 in stock, $1.08 at 1+, on 2026-10-05): 24 V to 12 V, 0.5 A, 15 to 36 V in, 3-pin SIP, 7812 pinout,
  on the board's existing TRACO-style footprint (11.5 x 7.55 x 10.2 mm). It replaces the TRACO TSR 0.5-24120 (C5993053: 18 in stock, $8.05, 32 V max in), which stays a valid alternate, as does
  DEXU K7812-500R3 (C909660, 1,016 in stock, $0.95 at 1+). Its datasheet (checked 2026-10-05) confirms the pinout and body size; the 22 uF/25 V ceramic input and output capacitors are optional.
  **`F5` (C976305) is no longer on the board; drop it from any earlier order.**
- **Optional capacitor pads, not in the BOM or CPL (do not populate):** `C12` (10 uF/50 V X7R, 0805) across `U4`'s 24 V input, and `C13` (47 uF/6.3 V X5R, 1210) on `+3V3`, in the strip between the
  ESP32-S3-ETH and DevKitC sockets. Both are marked DNP in the schematic and board, so `tools/build_jlc_assembly.py` skips them (counts unchanged). No LCSC numbers chosen. Their paste openings are in the
  gerbers, so an unfitted pad gets a small solder dome at assembly (harmless). A 1210 for `C12` does not fit: it hits the `U4` and `MH3` courtyards. Fit them only if the first-power-up scope check shows ripple.
- **BOM/CPL: 78 SMD parts / 219 joints, 103 full-assembly / 416 joints.** Gerbers re-exported (again after `C12`/`C13`, same 12-file set). ERC 0; DRC with schematic parity: 0 violations, 0 unconnected, 0 parity issues.
- `J-FAN` is a 12 V header (silkscreen *12V FAN*): never plug a 5 V fan into it.

## 2026-10-05 — fan header added

- **New parts:** `J-FAN` Molex 22232041 (C164168; alternative 470531000, C240840), `F5` BHFUSE BSMD1206-075-30V (C976305; alternative
  JDTfuse ASMD1206-075-30V, C2982268), `Q6` 2N7002 (C8545), `R57` 1k 0402 (C11702), `R58` 10k 0402 (C25744) and `R59` 4.7k 0402
  (C25900). Numbers, specs and stock checked on the LCSC/JLCPCB pages on 2026-10-05; JLCPCB re-validates stock on upload.
- **JLCPCB library type** (from JLCPCB's parts search, 2026-10-05): `Q6` C8545, `R58` C25744, `R59` C25900 and `R57` C11702 are **Basic**;
  `J-FAN` C164168 (9.6k in stock; the alternative C240840 is also Extended) and `F5` C976305 (64k in stock; the alternative C2982268
  is also Extended) are **Extended**, so expect JLCPCB's per-part fee for those two. None is on its preferred-extended list.
- **Orientation, not verified in JLCPCB's viewer** (that needs an upload): the community rotation table of the KiCad Fabrication Toolkit adds
  180 degrees for `SOT-23` (`Q6`, KiCad rotation 180), `SOT-353` (`U3`) and `CP_Elec_` (`C8`, `C11`); this CPL carries KiCad's rotations
  unchanged. `tools/build_jlc_assembly.py` now prints these parts after each run. Check `Q6` (gate = pin 1, on the right, with the
  drain pad toward `J-FAN`) in the preview and rotate it there if it is shown reversed. `J-FAN` is a through-hole wave-solder part, so the
  silkscreen decides: its friction-lock ramp side must face the board edge side shown in the footprint outline, and pins are marked G + T P.
- **BOM/CPL: 79 SMD parts / 221 joints, 103 full-assembly / 415 joints.** Gerbers re-exported (same 12-file set).
  DRC with schematic parity: 0 violations, 0 unconnected, 0 parity issues. Check `Q6`'s pin-1 orientation in JLC's
  placement preview.

## 2026-10-05 — second via on each +3V3 trunk layer change

- Three extra 0.6/0.3mm vias (and two short links) on the +3V3 trunk. Gerbers and the drill file
  re-exported; BOM/CPL unchanged. DRC with schematic parity: 0 violations, 0 unconnected, 0 parity
  issues.

## 2026-10-05 — panel ESD protection cells; board re-laid out

- **New parts:** 21 x 1k 0402 (UNI-ROYAL 0402WGF1001TCE, LCSC C11702), 7 x 100R 0402 (0402WGF1000TCE,
  C25076), 21 x PESD3V3S1UL (C70371, the same part as `D4`), `D32` SMBJ5.0A (C113974). Removed:
  `J-EXP`, `J-ENC-EXP4`. The generated cell references (`R29`-`R56`, `D6`-`D26`) are discovered from
  the board by `tools/build_jlc_assembly.py` rather than listed by hand.
- **BOM/CPL: 74 SMD parts / 210 joints, 97 full-assembly / 400 joints.** Gerbers re-exported (same
  12-file set). DRC with schematic parity: 0 violations, 0 unconnected, 0 parity issues.

## 2026-10-01 — `J3` (motor) re-pinned to the standard stepper order

- **`J3` is now 1 A+, 2 A−, 3 B+, 4 B−** (pins 1+2 one coil, 3+4 the other), so a standard
  stepper cable such as StepperOnline's 17HS08-1004S plugs straight in. One via added and the coil
  traces near `J3` re-routed. Gerbers re-exported; only F.Cu, B.Cu, F.Silkscreen (the new pin
  labels) and the drill file changed. BOM/CPL unchanged. DRC with schematic parity: 0 violations,
  0 unconnected, 0 parity issues.

## 2026-10-01 — `R28` added (1k series resistor on GPIO43)

- **`R28`**, 1k 0603 (YAGEO RC0603FR-071KL, LCSC C22548, the same part as `R9`/`R23`/`R25`), in
  series with the saturation encoder's DT line, next to `J1-H1` pin 20. One via moved. Gerbers
  re-exported; only the copper, the F.Mask/F.Paste openings for `R28`, the silkscreen and the
  drill file (that via) changed. BOM/CPL: **24 SMD parts / 68 joints, 49 full-assembly / 267
  joints.** DRC with schematic parity: 0 violations, 0 unconnected, 0 parity issues.

## 2026-10-18 — Q5, C6, D1 swapped for cost and stock

- **`Q5`:** Siliup SP30P03NK, **LCSC C41355006** (was AON6403, C2760089). **`C6`:** Rubycon 35ZLH220MEFC8X11.5, **C107639** (was SamYoung NXH35V220, C346959). **`D1`:** SS34, SMA, **C8678** (was the through-hole 1N5822, C82806; now an SMD part: SMD parts 82 (was 81), through-hole parts 24 (was 25), 106 in total, 420 joints).
- **`J2` sockets:** HCTL PM254-1-08-Z-8.5, **C2897371** (was Ckmtw C124416, 0 stock). **`U3`:** TI SN74AHCT1G125DBVR, SOT-23-5, **C7484** (was Nexperia 74AHCT1G125GW C12495, SC-70-5, 30 in stock); `U3`'s CPL rotation is 0 (KiCad has 180): the EasyEDA footprint has pin 1 bottom-right, which is how it sits on the board.
- **`F1`:** RUILON SMD1812P200TF16, 2 A hold resettable fuse, **C20812** (status *candidate*: confirm JLCPCB library type and stock at upload), replacing the `F1A`/`F1B` T508VS clips. (Superseded 2026-10-21: no clips remain on the board.)
- **`F3` removed** (and with it the T508VS clips, C41371869): the reason given then, that the RD-85A limits its own 5 V current below 15 A, is not supported by its datasheet (see the Hardware Overview, section 8). The 5 V side is unfused; use an inline blade-fuse holder in the harness with a larger supply.
- **Crowbar removed** (`Q7` C17702951, `D34` C173408, `R60`): no overvoltage protection beyond `D32`; see the changelog entry of 2026-10-21.
- **CPL rotations (2026-10-18):** the CPL now carries JLCPCB-frame rotations: `cpl_rotation_offsets.json` (made with `tools/check_cpl_rotation.py`, which compares each part's pad pattern with its EasyEDA footprint) gives the turn added to KiCad's rotation. Undecided and still at KiCad's rotation: `Q5`, `Q7` and the fuse clips `F1A/B`, `F3A/B`; check those in JLCPCB's placement preview.
- **`U1`:** Microchip MCP23017-E/SO, SOIC-28W, **C47023** (was MCP23017-E/SP DIP-28, C647352). Same pinout; about $1.05 cheaper per board at 100; now assembled in the SMD pass. CPL rotation 0 (KiCad has it at 90): JLCPCB's EasyEDA footprint has pin 1 bottom-left with the rows horizontal, which matches the board as placed.
- **Fuse clips (T508VS, C41371869):** (Superseded 2026-10-21: no clips remain.) 250 in stock covers 62 boards at 4 per board. Decision 2026-10-18: order in batches of up to 60 boards rather than change the footprint; the only well-stocked through-hole ATO holder (Littelfuse 178.6165.0002, C207061) costs about $3.50 more per board.
- BOM/CPL regenerated; gerbers re-exported; DRC 0/0/0.

## 2026-10-17 — overvoltage crowbar on the 5 V input (removed 2026-10-21)

- **New SMD parts:** `Q7` MSKSEMI AON6312-MS, **LCSC C17702951** (30 V N-ch DFN5x6, status *candidate*; same land pattern and CPL rotation hint as `Q5`), `D34` BZT52C4V7, **C173408**, `R60` 10k, **C98220**.
- BOM/CPL regenerated (SMD parts 81, full 106, joints 420); gerbers re-exported; DRC 0/0/0.

## 2026-10-17 — power input split, D33 for the DevKitC

- **`J-PWR-IN` removed.** `J-PWR-5V`: HCTL HC-5566-2*2A-05, **LCSC C20539147** (Mini-Fit Jr 2x2). `J-PWR-12V`: JST B2B-XH-A(LF)(SN), **LCSC C158012** (XH 2-pin). Harness parts are user-supplied: 5557-04R housing with Mini-Fit terminals for 5 V; XHP-2 housing with SXH-001T-P0.6 terminals for 12 V.
- **New `D33`:** MDD SS34, **LCSC C8678** (SMA), DevKitC 5V0 feed. Check its orientation in the placement preview.
- BOM/CPL regenerated (SMD parts 78, full 103, joints 407); gerbers re-exported; DRC 0/0/0.

## 2026-10-17 (later) — Q5 reverse-polarity FET and R19 restored

- **`Q5`:** AOS AON6403, **LCSC C2760089** (status *candidate*), footprint `Mainboard_local:DFN-8_5x6_Single_GSD`; check its orientation in JLCPCB's placement preview (pin 1 = first source pad; the CPL carries KiCad's rotation of 90 degrees and the community table suggests +270) and the land pattern against AOS's drawing.
- **`R19`:** Yageo RC0603FR-0710KL, **C98220**, gate pull-down.
- BOM/CPL regenerated (SMD parts 77, full 101, joints 407); gerbers re-exported; DRC 0/0/0.

## 2026-10-17 — U2 → YLPTEC K7803-2000R3, C12/C14 added, F1 silkscreen 2A, gerbers re-exported

- **`U2`:** YLPTEC K7803-2000R3, **LCSC C18212378** (status *candidate*), replacing the TSR 2-2433N (C45406999, 4 in stock). Same pads; body outline/courtyard redrawn for 11.5 x 9.0 mm; 17.5 mm tall.
- **New SMD parts:** `C12` (U2 input) and `C14` (U2 output), 22uF 25V X5R 0805, **C602037** (Samsung CL21A226MAYNNNE).
- **`R26`:** C25804 (0 stock) replaced by Yageo RC0603FR-0710KL, **C98220**.
- **`F1`** silkscreen now "F1 2A ATO" (was 2.5A); the blade is user-supplied, a 2 A ATO.
- BOM/CPL regenerated (SMD parts 75, full 99, joints 396); gerbers re-exported (same 12-file set, Protel names); DRC with `--schematic-parity --refill-zones`: 0 violations, 0 unconnected, 0 parity issues. `mainboard_bom_cost.csv` and `tools/bom_cost.py` price the BOM from LCSC.

## 2026-09-30 — fuse ratings on the silkscreen

- **"F1 2.5A ATO" and "F3 15A ATO"** added between each clip pair. Both positions take the same
  standard ATO fuse, so nothing else stops a 15A fuse going into `F1` (leaving the 24V motor
  rail unprotected) or a 2.5A one into `F3`. Silkscreen only; gerbers re-exported, BOM/CPL
  unchanged. DRC: 0 violations.

## 2026-09-30 — fuse clip spacing corrected (F1, F3)

- **The blade fuses wouldn't have fit.** XFCN's T508VS drawing specifies L = 19.04mm between
  the *outer edges* of the two clips' single pads (standard/low-profile ATO). The board had
  19.04mm between the clip *origins*, 23.6mm edge to edge, so the grips sat ~13.9mm apart
  against an ATO fuse's 9.3mm blade pitch (Littelfuse drawing). `F1A`/`F1B` and `F3A`/`F3B`
  each moved 2.28mm inward; the edges are now at x = 8.48 and 27.52 (19.04mm apart). Routes
  onto the clips were carried along. **CPL changed** (the four clip positions). DRC with
  parity: 0 violations, 0 unconnected, 0 footprint errors; the +5V feed cross-section is
  unchanged (10.8mm). Gerbers re-exported. `U2` in `ASSEMBLY_COMPARE.md` updated to the
  TSR 2-2433N.

## 2026-09-29 — U2 → TSR 2-2433N, DevKitC pins, parity cleanup

- **`U2` is now a TRACO TSR 2-2433N** (2A; JLC **C45406999**, status *candidate*, confirm stock),
  on KiCad's `Converter_DCDC_TRACO_TSR2-24xxN_TSR2-24xxxN_THT` footprint (same pads; body to the
  south). **CPL changed:** `U2` rotation is now 90°, so check its orientation in JLC's THT
  placement preview. TSR 2N datasheet checked (2026-09-30): no external input or output capacitor
  is required (internal input filter; the 1500µF part is only for surge-immunity testing).
- **DevKitC-1 spare pins tied:** `J6-J1B.2`, `J6-J3B.1/.2` to GND, `J6-J1.2` to +3V3. Same parts.
- **Parity cleanup:** footprint values and library IDs now match the schematic; mounting holes are
  board-only; `lib_footprint_mismatch` is ignored after a pad-by-pad check showed every footprint's
  copper matches the KiCad 10 library. DRC with parity: **0 violations, 0 unconnected, 0 footprint
  errors**.
- BOM/CPL regenerated (only `U2` changed); gerbers re-exported.

## 2026-09-29 (twelfth pass) — `R27` bleeder across `C8`/`C11`

- **New SMD part `R27`:** 4.7kΩ 1206, **LCSC C17936** (UNI-ROYAL 1206W4F4701T5E, 250mW, ±1%).
  Across `+24V_MOTOR` / GND, ~5mA. Not polarity-sensitive; placed at 0°.
- **BOM/CPL: 23 SMD parts / 66 joints, 48 full-assembly / 265 joints.** Gerbers re-exported
  (same 12-file set). ERC: 0 errors. DRC: **0 violations, 0 unconnected**.

## 2026-09-29 (eleventh pass) — `R26` LED data pull-down

- **New SMD part `R26`:** 10k 0603, **LCSC C25804** (same part as `R24`). Pull-down from
  `LED_DATA` to GND at `U3`'s input, so the ring stays dark while the WLED module boots.
- **BOM/CPL: 22 SMD parts / 64 joints, 47 full-assembly / 263 joints.** Gerbers re-exported
  (same 12-file set). ERC: 0 errors. DRC: **0 violations, 0 unconnected**.
- **At order time:** nothing polarity-sensitive. `R26` is placed at 90° just east of `U3`.

## 2026-09-28 (tenth pass) — `D5` TVS, `C11` second VM cap, `F4` removed

- **New SMD part `D5`:** SMBJ26A, SMB, **LCSC C123820** (MDD; alternate DIODES SMBJ26A-13-F C135062).
  Unidirectional TVS across `+24V_MOTOR`. Cathode (banded) is pad 1 on the `+24V_MOTOR` side;
  check the polarity in JLC's placement preview.
- **New SMD part `C11`:** 100µF/35V, same part as `C8` (**LCSC C176683**, candidate — confirm stock).
  Placed at 90°; check its rotation and polarity in JLC's placement preview.
- **`F4` (2A PTC, C2153878) is gone, with nothing in its place:** `U2`'s output now lands directly
  on `+3V3` (`+3V3_PSU` net merged into `+3V3`). `R23` (1k, STEP pull-up) moved north; same part.
- **BOM/CPL: 21 SMD parts / 62 joints, 46 full-assembly / 261 joints.** Gerbers re-exported.
  DRC: **0 violations, 0 unconnected**.

## 2026-09-28 (ninth pass) — LED data buffer, boot-strap resistors

- **New SMD parts:**
  - `U3`: SN74AHCT1G125DBVR, SOT-23-5, **LCSC C7484**. It must be the AHCT part, not AHC (C7468).
  - `C10`: 0.1µF (C1590).
  - `R23`/`R25`: 1k (C22548).
  - `R24`: 10k (C25804).
- **ENABLE/DIR swapped** between GPIO45 and GPIO3 (firmware updated to match), and re-routed.
- **Motor coil traces widened** from 0.2mm to 0.6mm on most of their length. Only short necks
  between header pins stay at 0.2–0.4mm, at most ~5mm per net. Each coil lead is now ~20–26mΩ.
- **BOM/CPL: 20 SMD parts / 60 joints, 45 full-assembly / 259 joints.** Gerbers re-exported.
  ERC: 0 errors. DRC: **0 violations, 0 unconnected**.
- **At order time:** besides `Q5`, check `U3`'s pin-1 orientation in JLC's placement preview.
  SOT-353 is small and asymmetric (3 pins on one side, 2 on the other), so a 180° error is
  visible: pin 1 (OE) should sit at the bottom-right corner, on the side facing away from
  `J5` and directly above `C10`'s right-hand pad.

## 2026-09-28 (eighth pass) — firmware ↔ netlist ↔ Waveshare pinout cross-check

- Every firmware GPIO checked against the netlist and Waveshare's ESP32-S3-ETH schematic (H1/H2):
  all 20 match. That covers STEP/DIR/ENABLE, TMC UART, WLED link, I2C and the four native encoders.
  The MCP23017 button bits (GPB0/1/3) match too. ERC is unchanged (cosmetic off-grid warnings only).
- **Silkscreen:** `J-ENC-EXP3`'s SW pin is unconnected (GPB3 went to the panel SAT button) but
  was still labelled `S`. It now reads `x`. Gerbers re-exported; BOM/CPL unchanged. DRC:
  **0 violations, 0 unconnected**.

## 2026-09-28 (seventh pass) — C6 on the pour, C8 GND via, 3V3 widened

- **`C6` (220µF ring bulk cap) hung off a 0.2mm B.Cu trace plus a via.** The `+5V_LED_PSU` pour
  now extends down over it, and the trace and via are deleted.
- **`C8` (TMC VM bulk) returned to GND through 10mm of 0.4mm trace to one via.** A GND via now
  sits right at its GND pad.
- **The +24V hop vias were upsized** from 0.8/0.4 to 1.0/0.5mm.
- **+3V3 widened wherever DRC allows** (0.5mm on most segments, stepping down to 0.25mm or the
  original 0.2mm where clearance needs it). U2→F4 went from 49 to 20mΩ, and F4→ETH module from
  96 to 60mΩ.
- No footprints moved: **BOM/CPL are byte-identical**, gerbers re-exported. DRC: **0 violations,
  0 unconnected**.

## 2026-09-28 (sixth pass) — +5V pour widened at the power connector

- **The +5V feed still necked to ~1.3mm right below `J-PWR-IN`.** Two things walled the pour off
  under the pad row: the F.Cu +24V trace (pin 5 → `F1A`) crossing at y=15.2, and an F.Cu GND
  trace looping pin 6 → pin 8 around NC pin 7. Both GND pins are through-hole on the B.Cu plane,
  so the GND loop was redundant.
- **`F1A`/`F1B` swapped places.** They're identical clips; the schematic is unchanged. The
  24V-in clip `F1A` is now on the west side, with pin 5 → `F1A` running on B.Cu through a 1.0/0.5mm
  via. The fused side `F1B` (east) → `D1` is now a short spur inside the pour, not a wall across it.
  The redundant GND loop was deleted, and the +5V zone widened west (x=11) and east (x=34.5).
- **Result:** the narrowest single path is 3.3mm, and total copper across the pad row is 10.8mm
  (was 1.3mm). The ring's worst cross-sections elsewhere are 5.5mm at the `Q5` source leads and
  ≥16mm GND on B.Cu. **CPL changed** (the `F1A`/`F1B` positions); BOM, gerbers and CPL were
  re-exported. DRC: **0 violations, 0 unconnected**.

## 2026-09-28 (fifth pass) — F2/D2 removed, LED-ring +5V feed fixed

- **`F2`/`D2` removed from the board** (they had been DNP since 2026-09-20; the branch fed nothing).
  BOM/CPL unchanged, since neither was in them.
- **Every export before this one fed the LED ring's +5V through a 0.2mm trace.** The +24V leg
  from U2 split the `+5V_LED_PSU` pour, leaving a single 0.2mm B.Cu trace between `J-PWR-IN` and
  `F3`. The +24V now hops under the pour on B.Cu, so the +5V pour is continuous on F.Cu (details
  in the repo README). DRC: **0 violations, 0 unconnected**. Gerbers re-exported.

## 2026-09-28 (fourth pass) — silkscreen legibility, schematic hygiene, orphan cleanup

- The 75 silkscreen pin labels added on 2026-09-27 went from a 0.12mm to a 0.15mm stroke
  (JLCPCB's legibility minimum is ~0.15mm). The encoder pin letters moved 0.2mm right to clear
  the header outlines. DRC: **0 violations, 0 unconnected**. Gerbers re-exported.
- Schematic only (netlist proven identical): `J3` was drawn on top of `J2-MOTOR`, and `J6-J1B`
  on top of `R18`. Both were moved clear, and a dangling leftover wire+label was deleted.
- The placement/routing scripts named in the history below (`build_pcb_100.py`,
  `route_pcb_100.py`, `place_check.py`) were removed; see git history (last present in `b2d7025`).

## 2026-09-28 (third pass) — module sockets un-mirrored, 24V widened

**Every export before this one had both plug-in module footprints mirrored. Don't order those.**
Swapping each socket pair's x positions fixed it, and all traces to those pins were re-routed
(details in the repo README). Also: the +24V chain is widened to 1.0mm (0.5mm floor), the
TMC2209 GND tie is widened, and the ETH silkscreen now reads "RJ45 SIDE UP". BOM is unchanged;
the CPL has new positions for `J1-H1`/`J1-H2`/`J6-*` (THT, full-assembly file only). Gerbers/drill
re-exported (same 12-file set). DRC: **0 violations, 0 unconnected**.

## 2026-09-28 — Q5 pin mapping fixed, Q6/R20 removed, LED-ring pour un-cut

**Every export before this one had Q5/Q6 wired to the wrong pins. Don't order those.**
The `Q_PMOS_GSD` symbol's pin numbers (1=G, 2=S, 3=D) were matched against KiCad's
`PowerPAK_SO-8L_Single` pad numbers (1–3 source, 4 gate, 5 drain tab). That put the gate and
drain nets on source pins and left the real gate and drain unconnected (see the repo README).

- `Q5` now uses `lib.pretty/PowerPAK_SO-8L_Single_GSD` (pads renumbered to match the symbol).
  Its position/rotation and CPL row are unchanged. In JLCPCB's placement preview, confirm the
  part's pin-1 dot lands on the footprint's pin-1 mark.
- `Q6` + `R20` removed from the design. **BOM/CPL now: 15 SMD parts / 47 joints, 40
  full-assembly parts / 246 joints.**
- `+5V_LED_RING` pour is now continuous and full-width from Q5 to `J5`. `LED_DATA` moved to B.Cu
  under it, and `J5` pin 2 joined the ring net; the silkscreen now reads "+5V" there.
- Gerbers/drill re-exported (same 12-file set). DRC: **0 violations, 0 unconnected**.

## 2026-09-27 (second pass) — C9 + connector pin-label silkscreen

- **`C9` added:** 0.1µF 0603, LCSC **C1590** (same part as `C7`), decoupling `U1`'s VDD pin.
  **BOM/CPL now: 17 SMD parts / 62 joints, 42 full-assembly parts / 261 joints.**
- **Silkscreen:** pin labels outside the housings of `J-PWR-IN`, `J5`, `J3`, all 8 encoder
  headers, `J-BTN-TAP` and `J-EXP`. The encoder, tap and expansion headers' reference designators
  are hidden and replaced by short names. `J-PWR-IN`, `J5` and `F1A` reference text moved to make room.
- Gerbers/drill re-exported (same 12-file set). DRC: **0 violations, 0 unconnected**.

## 2026-09-27 — C6 polarity fix, new C8, module orientation silkscreen

- **`C6` (220µF electrolytic) was reversed** in the previous gerbers/CPL. Its + pad was on GND and
  its − pad on `+5V_LED_PSU`, so JLCPCB would have fitted it reverse-biased across the 5V input.
  Now rotated 180° in place. The CPL row changed from `24.0, 0°` to `27.5, 180°`, same
  footprint origin convention (pad 1 = +).
- **`C8` added:** 100µF/35V SMD electrolytic, Lelon VZH101M1VTR-0607, **LCSC C176683**. It's an
  Extended part, so check stock when uploading. It's the TMC2209 VM bulk cap, placed beside
  `J2-MOTOR` pin 1. **BOM/CPL now: 16 SMD parts / 60 joints, 41 full-assembly parts / 259 joints**
  (was 15/58 and 40/257).
- **Silkscreen:** orientation marks for the ESP32-S3-ETH, DevKitC-1 and TMC2209 sockets.
- Gerbers/drill re-exported with KiCad 10.0.6, same layer and file set as before (12 files).
  `kicad-cli pcb drc --schematic-parity --refill-zones`: **0 violations, 0 unconnected**.
- `tools/build_jlc_assembly.py` now picks the newest installed `kicad-cli` by version number.
  Before, it sorted install folders as text, picked KiCad 9.0, and failed to load this KiCad 10
  board.

## 2026-09-20 — F2/D2 dropped from the BOM (unused after the pull-up rail fix)

`F2` (2A PTC fuse) and `D2` (1N5822 Schottky) formed the general-purpose `+5V_LED` protection
branch. The encoder pull-up rail fix (below) removed the only load that branch ever fed, leaving
both parts wired on the board with zero function. Removed from `tools/build_jlc_assembly.py`'s
`PARTS` table and re-ran it — **BOM/CPL regenerated: 15 SMD parts / 58 joints (was 16/60), 40
full-assembly parts / 257 joints (was 42/261)**. Both footprints stay unpopulated (DNP) on the
PCB itself; nothing in the gerbers/layout changed, only what gets assembled. `F4`'s BOM row split
out from the old `"F2,F4"` consolidated line since it's the only PTC fuse left on that part number.

## 2026-09-20 — gerbers re-exported after encoder pull-up rail fix

All 8 panel-encoder headers' pull-up supply pin was moved off `+5V_LED` onto `+3V3` (was putting
~5V on 3.3V-only ESP32-S3/MCP23017 inputs — see the repo README's Power rails section and Known
open items). Fixed in both schematic and PCB copper, then verified with a real `kicad-cli` DRC
pass (zones refilled, schematic-parity checked): **0 violations, 0 unconnected items**. Only the
copper changed (no footprints moved), so **`Mainboard_gerbers.zip` was regenerated; `Mainboard_BOM_*.csv`
and `Mainboard_CPL_*.csv` were re-run too but came out byte-identical** to the prior export — placements
didn't change. Gerbers exported via `kicad-cli pcb export gerbers`/`export drill` at the same
layer set as every prior export here (`F.Cu`/`B.Cu`/paste/silkscreen/mask + `Edge.Cuts`, Protel
filenames, separate PTH/NPTH Excellon drill). `F2`/`D2` (the now-unloaded `+5V_LED` general
branch) remain in the BOM/CPL — they're wired but unused; drop them from the assembly if desired.

## 2026-09-10 — shrunk to 100×100mm + full high-current copper + assembly breakout

The board was **re-laid-out from 150×150 to 100×100mm** (JLCPCB's cheap price tier) and fully
re-routed — **0 unconnected, 0 DRC violations** (silkscreen fully de-conflicted: connector labels
are vertical, one per header; crowded left-column power labels relocated to clear space).
Regenerated by:
`tools/build_pcb_100.py` (re-shape + re-place) → `tools/route_pcb_100.py export` → Freerouting →
`tools/route_pcb_100.py import` (re-route + pours). Placement was overlap-checked by
`tools/place_check.py`.

**High-current copper — the LED-ring path now has real copper on *every* leg** (the old board
only poured 2 of the 4 high-current nets and fed them through 0.2mm traces):
- `GND` — full **B.Cu ground plane** (the 10.7–14A return + all signal grounds).
- `+5V_LED_PSU` / `+5V_LED_RINGFUSE` / `+5V_LED_RING` — three **F.Cu pours**, ~12–18mm wide,
  covering J-PWR-IN → F3 → Q5 → J5. All ring-path +5V legs stay on F.Cu (Q5 is a top-side SMD
  FET), so 14A never crosses a via. High-current pads forced to **solid** zone connection.
- Q5 is placed at rot180 so its ring-fuse pin faces F3 and its ring pin faces J5, letting the two
  F.Cu zones stack without overlap.
- **⚠ Order at 2oz (70µm) copper.** IPC-2221 (external, 10°C rise): 14A needs ~5.7mm-equivalent at
  2oz; the pours exceed that with margin. Copper weight is a JLCPCB order-form selection (not
  encoded in gerbers) — **set it to 2oz for both outer layers at checkout.**

## Assembly: two upload sets for cost comparison (2026-09-09)

You can quote **SMD-only** vs **full (SMD+THT)** assembly — see
[`ASSEMBLY_COMPARE.md`](ASSEMBLY_COMPARE.md). Generated by `tools/build_jlc_assembly.py`
(reads placements live from the board, so re-run after any layout change).

- SMD-only: `Mainboard_BOM_SMD.csv` + `Mainboard_CPL_SMD.csv` (16 parts, 60 joints)
- Full:     `Mainboard_BOM_FULL.csv` + `Mainboard_CPL_FULL.csv` (42 parts, 261 joints)
- Rollup:   `assembly_cost_compare.csv`

---

## History (150×150 SMD-only export)

Originally exported 2026-08-29 from `Mainboard.kicad_pcb` / `Mainboard.kicad_sch`, after fixing a
real footprint mismatch on `Q5`/`Q6` (see below) and re-routing.

**Gerbers/drill re-exported again same day** after fixing a socket-orientation bug: the WLED
co-processor DevKitC-1 sockets (`J6-J1`/`J6-J3`) had their real USB-C end facing into the board's
interior instead of an edge (see below). `CPL`/`BOM` were re-checked against the new board and
are byte-identical to the previous export — none of the 12 SMD parts moved, only the two THT
DevKitC sockets and the 4 mounting holes did — so only the Gerbers/drill/job file were replaced.

**Gerbers re-exported again 2026-09-03** after moving the I2C bus off strapping pin `GPIO46`:
the mandatory I2C pull-up (`R21`) was holding that strapping pin high at every reset. Fix moved
`SDA` to freed `GPIO44` (the panel sat button was pushed onto the MCP23017 expander, `GPB3`), so
`GPIO46` is now unused and no pull-up sits on any strapping pin. Schematic nets changed → PCB
re-placed-in-position (placement unchanged) and fully re-routed via the usual Freerouting flow;
**0 DRC violations, 0 unconnected**. `CPL`/`BOM` verified byte-identical again — none of the 12
SMD parts moved (`R21`/`R22` kept position + value, only their net changed) — so only the
Gerbers/drill/job file were replaced.

## Files

- `Mainboard_gerbers.zip` — Gerbers (F/B Cu, F/B Paste, F/B SilkS, F/B Mask, Edge.Cuts) + Excellon
  drill files (PTH/NPTH separate) + the `.gbrjob` job file. Upload this directly to JLCPCB's
  "Add Gerber file" step.
> **2026-09-10: full assembly needs zero hand-soldering** — all 8 previously
> sourcing-risk parts were substituted with in-stock JLCPCB parts (module sockets,
> Molex→HCTL 5566 power headers, J6 1×22→1×20+1×2, F1/F3→T508VS SMD blade clips).
> See `ASSEMBLY_COMPARE.md`. The blade fuses and plug-in modules remain user-supplied.

- `Mainboard_CPL_SMD.csv` / `Mainboard_CPL_FULL.csv` — Pick-and-place / position files (SMD-only
  = 16 rows; full = 42 rows). Already in JLCPCB's expected format
  (`Designator,Mid X,Mid Y,Layer,Rotation`, `mm`-suffixed coordinates, `Layer` as `T`/`B` — the
  reformatting recorded in `feedback_jlcpcb_cpl_format`). Regenerated live from the board by
  `tools/build_jlc_assembly.py`, so re-run that after any placement change.
- `Mainboard_BOM_SMD.csv` / `Mainboard_BOM_FULL.csv` — Bills of materials, consolidated one row
  per unique part, designators comma-separated (e.g. `F2,F4`). SMD file matches JLCPCB's sample
  structure (`Comment,Designator,Footprint,LCSC Part #`); the FULL file adds `Assembly`,
  `Source status`, and `Note` columns (drop those 3 columns if JLCPCB's parser is strict).
  See `ASSEMBLY_COMPARE.md` for the SMD-only vs full cost comparison.

  *(The old single-variant `Mainboard_BOM.csv/.xlsx` + `Mainboard_CPL.csv/.xlsx` were removed —
  the bare CPL held stale 150×150 positions. `.xlsx` variants are no longer produced; JLCPCB
  accepts the `.csv` directly.)*

## Real bug found and fixed before this export: Q5/Q6's footprint didn't match the real part

> **2026-09-28:** this fixed the package outline but not the pin numbering, which stayed wrong
> until the 2026-09-28 entry at the top of this file.

The schematic specified `Q5`/`Q6` as Vishay's real **SQJ407EP** (electrically verified against
its datasheet earlier this project — RDS-on, current rating), but the PCB footprint assigned
was a generic `SOIC-8-1EP_3.9x4.9mm` — the wrong package. The real part ships in Vishay's
**PowerPAK SO-8L** (body 5.13 × 6.15mm, segmented exposed drain pad, 4 tied source pins), a
completely different, larger footprint that would not have matched the physical part had this
board been ordered as-is. Confirmed against the real Vishay datasheet
(`vishay.com/docs/62806/sqj407ep.pdf`) and fixed by switching to KiCad's stock
`Package_SO:PowerPAK_SO-8L_Single` footprint (dimensions cross-checked against the datasheet's
package outline table before trusting it). Board was fully re-placed and re-routed after the
change — 0 DRC violations, 0 unconnected items.

## Real bug found and fixed same day: DevKitC-1 sockets weren't edge-oriented

`J1-H1`/`J1-H2` (the Waveshare ESP32-S3-ETH sockets) and `J6-J1`/`J6-J3` (the Espressif
DevKitC-1 sockets) were both placed pin1-at-y=8 (near the board's top edge). That's correct for
the ETH module — confirmed against Waveshare's real dimensioned board diagram that pin 1 of
`H1`/`H2` is the RJ45/USB-C end — but wrong for the DevKitC-1: confirmed against Espressif's
real board photo (silkscreen pin labels) that pin 1 of `J1`/`J3` is the **WROOM-module** end,
and the USB-C end is at the opposite (pin 22) end. With both sockets placed pin1-up, the ETH
module's RJ45/USB-C ended up correctly at the board edge, but the DevKitC's USB-C port would
have landed at y≈61mm — deep in the board's interior, inaccessible once the module is socketed.
Fixed by rotating `J6-J1`/`J6-J3` 180° and repositioning so pin 22 (their real USB-C end) is the
end at y=8, matching the ETH sockets. Required a full board regeneration (placement + Freerouting
+ manual fix for one net Freerouting couldn't resolve, `MOTOR_COIL_B1`, routed by hand via a
verified-clear F.Cu/B.Cu/F.Cu path) — 0 DRC violations, 0 unconnected items.

## D4 swapped: PESD3V3L4UF (SOT-886) -> PESD3V3S1UL,315 (D_SOD-882/DFN1006-2)

`D4` only ever used 2 of the original part's 6 pins (`+3V3` and `GND` — a simple single-line
clamp, not the 4-line array the part provided), so it was swapped for a single-line unidirectional
TVS in a smaller, better-stocked package. Verified against the real Nexperia PESD3V3S1UL
datasheet (fetched via LCSC's hosted copy) that pin 1 is the cathode (K) and pin 2 the anode (A),
matching the schematic's `Device:D_Zener` symbol and the `Diode_SMD:D_SOD-882` footprint's pad
numbering — cathode wired to `+3V3`, anode to `GND`, same polarity convention as the original
part. Schematic (symbol + footprint field) and PCB (footprint swap + hand-routed reconnection of
both nets) both updated — 0 ERC violations, 0 DRC violations.

## Assembly scope (the SMD-only option)

> As of 2026-09-10 a **full-assembly** BOM/CPL also exists — see the breakout note at the top and
> `ASSEMBLY_COMPARE.md`. The section below documents the **SMD-only** option (`*_SMD.csv`), which
> remains the cheapest path; the THT parts it excludes are hand-soldered after the board arrives.

Only 16 of this board's 42 components are SMD; the rest are connectors/sockets (13 of them,
including the `J6-J1B`/`J6-J3B` 1×2 split-off pair), `U1` (MCP23017, DIP-28 THT), `U2` (TSR
1-2433, TO-220-3 THT), the bulk cap, and the THT reverse-polarity diodes — all easy to
hand-place and, for the connectors especially, worth hand-placing anyway to verify fit against
the real mating modules (ESP32-S3-ETH, the DevKitC, the TMC2209 module) before soldering.
Sending only the 16 fiddly SMD parts (a PowerPAK SO-8L FET pair, ten 0603 passives, and the 4
T508VS SMD blade-fuse clips) through JLCPCB's SMT line avoids THT assembly fees/lead time for
parts that are genuinely easier to place by hand. (`F1`/`F3` themselves are no longer on this
hand-solder list — as of the fuse-clip swap above, they're SMD-assembled as `F1A`/`F1B` and
`F3A`/`F3B`; only the blade fuses that drop into those clips are still user-supplied.)

**Excluded from the CPL/BOM, hand-solder after the board arrives:**
`C6`, `D1`, `D2`, `J-BTN-TAP`, `J-ENC-EXP1`–`4`, `J-ENC-HUE`, `J-ENC-MAIN`,
`J-ENC-SAT`, `J-ENC-VAL`, `J-EXP`, `J-PWR-IN`, `J1-H1`, `J1-H2`, `J2-MODE`, `J2-MOTOR`, `J3`,
`J5`, `J6-J1`, `J6-J3`, `J6-J1B`, `J6-J3B`, `U1`, `U2`. `MH1`–`MH4` are mechanical mounting
holes, nothing to assemble.

## LCSC part numbers

| Part | Value | LCSC # | Notes |
|---|---|---|---|
| `C7` | 0.1uF 0603 | C14663 | Same part as the LED ring board (`kicad_led_ring_offset`) — Yageo CC0603KRX7R9BB104, 50V, Basic. |
| `D4` | PESD3V3S1UL,315 | C70371 | Nexperia unidirectional TVS diode, D_SOD-882/DFN1006-2 package. Pin 1 = cathode (to `+3V3`), pin 2 = anode (to `GND`) — confirmed against the real Nexperia datasheet. |
| `F4` | 2A PTC resettable | C2153878 | Littelfuse 1210L200PR, 6V rating — fine for the 3.3V rail it sits on. (`F2`, which shared this part, was removed 2026-09-28.) |
| `Q5` | SQJ407EP-T1_GE3 | **C727764** | Vishay P-channel ideal-diode FET, PowerPAK SO-8L package — confirmed directly from LCSC's product page (`productModel":"SQJ407EP-T1_GE3"`, brand VISHAY) after the base part number alone didn't surface in several earlier search attempts. Package cross-checked against the real Vishay datasheet before trusting it (see above) — this is the part the PCB footprint was fixed to match. |
| `R9` | 1k 0603 | C21190 | UNI-ROYAL 0603WAF1001T5E, Basic. |
| `R18` | 330R 0603 | C23138 | UNI-ROYAL 0603WAF3300T5E, Basic. |
| `R19` | 10k 0603 | C25804 | Common LCSC basic part. |
| `R21`, `R22` | 4.7k 0603 | C23162 | UNI-ROYAL 0603WAF4701T5E, Basic. |

JLCPCB's BOM upload tool re-validates all of these against live stock — worth a last check there
regardless, stock levels change often.

## Known open items before ordering

- **Copper weight**: select **2oz (70µm)** in JLCPCB's PCB spec form (defaults to 1oz). Required
  for the LED-ring current path — see the high-current copper note at the top. Copper weight is not
  encoded in the gerbers, so it must be set on the order form.
- **Module mechanical clearance** (100×100 packs the two plug-in modules closer than the old
  board did): verify against the real hardware that the ESP32-S3-ETH and DevKitC-1 bodies don't
  collide with each other or overhang tall left-column parts (blade-fuse holders, `C6`, `U2`).
  The RJ45/USB-C ends of both modules sit on the top edge (correct); confirm module length doesn't
  overhang the TMC2209 (`J2`) socket below the ETH module. From the placement (2026-09-27 review):
  the two modules sit ~7mm apart side by side, and the left-column parts sit clear of the ETH
  module's footprint. **The one real limit: the ETH module can extend at most ~4mm past the centre
  of its last header pin (the end away from the RJ45)** before it overlaps the TMC2209 module
  in plan view. Both sit on the same socket height, so that would be a real collision. Measure
  that distance on the actual module with calipers. **Update 2026-09-28, from Waveshare's
  dimensioned drawing:** the module is 72.8 × 21mm and ends ~3mm past its last pin, so it
  clears the TMC2209 module by ~1.4mm. The USB/RJ45 end overhangs our board edge by ~16mm,
  and the RJ45 is on the module's top side.
- **Silkscreen**: fully de-conflicted — 0 DRC violations (connector labels vertical, one per
  header; left-column power labels moved clear).
- **ERC**: still 0 *errors*, and down to 149 *warnings* (`erc_report.rpt` regenerated 2026-09-12,
  live from the current schematic — the previously-committed copy was stale from 2026-09-03,
  predating the fuse-clip swap and J6 socket split). 147 are sub-mil `endpoint_off_grid` pin/wire
  stubs (cosmetic), 1 is a likely stray micro-wire near `J2-MOTOR` pin 1 (the PCB itself shows 0
  unconnected pads, so this isn't a real net break), and 1 remaining `lib_symbol_mismatch` is `U1`
  (`MCP23017_SP`) — verified pin-for-pin identical to the current library, purely a Description-text
  divergence, no connectivity impact; syncing it needs KiCad's own "Update Symbol from Library"
  (GUI-only, since there's no byte-exact target to hand-reconstruct outside the tool).
  Fixed along the way: a missing `fp-lib-table` entry for the project-local `FuseClip_T508VS`
  footprint (also cleared 4 related `footprint_link_issues` warnings), `D4`'s one-line divergence,
  and the `Conn_01x02` mismatch shared by `J6-J1B`/`J6-J3B`/`J-BTN-TAP` — that one's library
  revision actually moved the pin *positions*, so fixing it meant re-deriving each pin's new
  absolute position (via the symbol-to-sheet Y-flip transform) and moving the connected wires/
  no-connect markers to match, verified afterward against a fresh netlist export (`J-BTN-TAP` pin 1
  still nets to `GND`, pin 2 to `TAP_BTN`; `J6-J1B`/`J6-J3B` remain correctly unconnected).
- No PCB has been hand-reviewed in the KiCad GUI yet — worth a visual pass before ordering,
  particularly the three F.Cu high-current pours, the B.Cu ground plane, and `Q5`'s rot180 placement.
