Nobotic USV, Cape Dory 25D conversion.
What this is. The plain-language write-up of the boat's electronics as they now stand: the two compute boxes, the two team-built release and monitoring nodes, the CAN network, the battery watchman, cameras, AIS, storage, lights, connectors, and the parts to buy.
Who wins on what. The Compute Bay Housing and Wiring Plan v9.6 stays the design authority for the two-island architecture, lightning safing, and the surge doctrine. This plan is the box-by-box implementation and the shopping list, and it carries the per-box hole schedule (Appendix A). Where the two disagree, the Compute Bay Plan wins on architecture and this plan wins on part numbers and hole counts.
How to read it.
| Part | What it holds |
|---|---|
| Sections 2 to 7 | The design as it stands |
| Section 8 | The buy list |
| Section 9 | What still needs a decision or a bench check |
| Appendix A | Every box and every hole |
| Appendix B | Order status |
Change history and the glossary live in the Changelog document. This plan describes only the current design.
Companion documents: Changelog (all history, decisions, and the glossary); Node Boxes Explained v9.6 (the wiring pictures); Box Wiring Sheets v9.6 (one drawing per box); Battery Box Building Guide v9.6; Lightning Ground System Guide v9.6; Steering System Handbook v9.6; BMS Software Change Notes v9.6; Serial and Network Summary (the port audit of 2026-09-08/09).
Terms used here.
| Term | Plain meaning |
|---|---|
| Island | A group of gear fed from one battery, with no wire path to the other group |
| Node | A small team-built box with a microcontroller in it |
| CAN bus | The shared two-wire network the boat's devices talk over |
| Gland | A screw-in fitting that clamps and seals around a cable where it enters a box |
| Drip loop | A downward sag in a cable, so water runs to the bottom and falls off |
| Ferrite | A clip-on magnetic collar that soaks up surge current riding on a cable |
| Staked | A dab of adhesive on a plug or connector so vibration cannot walk it loose |
| De-energized-closed | A relay that passes current when it has no power. Losing the relay does not lose the load |
| Opto board | An input board that reads switch contacts across a gap of light, so no wire crosses |
| PoE | Power over Ethernet: the network cable carries the device's power as well as its data |
| PoE class | The power level a PoE device asks the switch for when it is plugged in (802.3af up to 15.4 W, 802.3at up to 30 W). The switch sets that much aside whether or not the device uses it |
| SFP module | A small plug-in module that turns a switch's SFP slot into a port. The copper SFP module gives an ordinary RJ45 jack, data only, no power |
| Web API | The switch's own documented command interface over the network. A program on the Pi calls it to turn a port on or off |
| SNMP | A standard way to read a network device's state over the network. On the TSW202 it is read-only: good for checking, not for control |
| H-bridge | A four-switch circuit that runs a DC motor in either direction |
| RS-232, RS-485 | Two wired serial standards most science instruments speak. RS-232 is one device per cable; RS-485 is a shared two-wire line |
| FTDI | The maker of the USB-to-serial chips this plan allows. Each has a unique serial number, so the computer can tell one cable from another |
Section 1, the change history, moved to the Changelog.
Both are the CXCESNS 11.8 x 8.3 x 3.9 in aluminum boxes already on hand.
Standard fit for both: all cable entries on the bottom face, drip loops, one ferrite per entry outside, vent, desiccant, humidity sensor, and all boards conformal-coated.
Inside each box, a 1/8 in (3 mm) G10/FR-4 plate, blank about 11.2 x 7.6 in, rides on the ledge holes along both long walls.
Through-bolt the box back wall to the 3/16 to 1/4 in solid glass aft of the settee.
Bed on butyl through build and trials; re-bed on permanent sealant at the final flight install. Add the four holes to the shared drilling template.
The carrier stands vertically on a slotted 1/8 in 6061 L-bracket, about 3 x 3 x 4 in with two slotted base holes, bolted to the plate.
Heat path: SoC, thermal pad, a flat aluminum block, a gap pad, the box wall.
After first power-up, run the CM5 hard for ten minutes. The wall over it should feel warm. SoC temperature is a trended health number in telemetry.
The Pi A box is bonded to the hull skin below the waterline, aft of the settee and as far from both lightning straps as the space allows, so the box wall stays near sea temperature whatever the cabin does.
Bond method:
The CM5 block and gap pad press on the back wall.
This is the coldest box aboard, so condensation is expected and managed:
The pass/fail stays: interior at or below 122 F, with a 122 F simulated cabin and the hull spot at sea temperature.
Pi B, under 5 W, stays through-bolted to the settee panel.
| Inside | Note |
|---|---|
| CM5 on the carrier on hand, boots from its own eMMC | The carrier seats both the 4 TB NVMe SSD (one WD Blue 4 TB bought) and the Hailo-8L M.2 accelerator (three bought; one flies on Boat 1, two are Boat 2 stock, not spares). Bench-prove both slots before the wiring is committed |
| Waveshare 2-CH CAN FD HAT | Control and house buses, isolated; stack staked |
| WD-A watchdog (Nano Every + regulator) | Cycles Pi A's rail on 5 min silence. Pass element defaults ON: a dead watchdog leaves the rail powered. Bench check: pull the Nano, the CM5 stays up |
| GPS A (hiBCTR NEO-7M, bought), IMU A (BNO085) | Position only. Heading comes from the BNO085 IMUs (three bought 2026-09-08) and the track; there is no moving-base pair and no heading UART. The NEO-7M is single-band GPS with GLONASS, accurate to a few yards, which is all the boat needs; two of the 4-pack bought 2026-09-09 fly, two are Boat 2 stock |
| LoRa bonnet, 915 MHz, jumpered to an SMA on the wall | Wing pod link |
| dAISy AIS receiver (USB stick) | Always-on listener on its own below-deck antenna. Cable clamped, plug staked. The em-trak is the second, independent AIS receiver |
| One Ethernet run to the TSW202 at the aft junction | No switch in this box. Pi B and Starlink reach Pi A through the TSW202, the boat's only network switch (sec 6.1) |
| Digital isolator (ADuM1201) for Iridium A serial; optocoupler for the inter-box alive discrete | The inter-box link is now only the main-island-alive discrete. The heading UART and its two isolators are gone with the ZED-F9P pair |
| Metal screw-mount powered USB hub with a captive cable, the USB isolator on hand on its link | Carries the three FTDI USB-to-RS-232 cables and the one FTDI USB-to-RS-485 adapter for the science instruments (sec A1). FTDI silicon only, never CH340. A udev rule keyed to each cable's serial number, so a sensor keeps its name after a reboot |
| The SBMS0 USB lead (via the WiFi/USB add-on board), on the isolated hub | Pack telemetry only, optional. The watchman's one hand, the charge pause on its EXTIO4 output (sec 6.7), is hard-wired and never crosses USB: a quiet stream means eyes lost. No VE.Direct-to-USB cables come to this box (see A1) |
| 24 V to 5 V 5 A converter, TVS, terminal strip, INA226 on the feed | Power tidy. The box feed is one small branch on its own auto-reset breaker at the breaker landing bar (sec 7.6); no blade-fuse block |
| Hailo-8L M.2 accelerator | In the carrier's second slot. Three bought because they were cheap: one flies on Boat 1, the other two are Boat 2 stock, not spares. Nothing aboard carries a second one |
| No secure element chip | The HMAC key is a file on the SSD, outside the source repo, rotated at each refit (deleted, do not re-propose) |
Wall entries (11 + vent + bond stud): main 24 V feed (gland), CAN control drop, CAN house drop, VESC A UART + kill, Iridium A serial, inter-box link (alive discrete only), SBMS0 USB, one Ethernet run to the TSW202 at the aft junction (RJ45 coupler), GPS A coax (SMA), dAISy coax (SMA), 915 MHz whip (SMA). The separate Starlink and Pi B Ethernet entries are gone with the box switch. Not yet in this count: the glands for the science serial leads from the USB hub (three RS-232, one RS-485). Where the hub sits (inside this box, or in its own small dry box nearer the instruments so that one USB cable crosses this wall) is settled at fit-out, and the glands are counted then; the box has room for three more holes under the 14-hole cap (decision open, Pending Amendments v9.7).
| Inside | Note |
|---|---|
| CM5 on carrier + own CAN FD HAT | Stack staked |
| 250 GB NVMe SSD | Mirrored state, in the carrier's M-key slot |
| WD-B watchdog (Nano Every + regulator) | Owns the whole lightning role. Same default-ON rail rule as WD-A |
| GPS B (hiBCTR NEO-7M, bought), IMU B (BNO085) | Position from GPS B; heading from the IMUs and the track |
| Isolator (ADuM1201) for Iridium B serial; optocoupler for the alive discrete from WD-A | The heading UART and its isolator are gone |
| No secure element chip | Same rule as Pi A: the key is a file on the SSD |
| Own DC-DC + TVS referenced to the floating negative |
Wall entries (8 + vent): emergency feed (gland), CAN control drop, CAN house drop, Iridium B serial, inter-box link (alive discrete), detector lead (7-core, under 3 ft), Ethernet to the TSW202 at the aft junction (UTP, plastic coupler; a new long run to the stern), GPS B coax (SMA).
Same cabin, a few feet apart, is fine.
The archive drive is one 4 TB NVMe SSD: a WD Blue 4 TB, bought, arriving 2026-09-21. It sits in one M.2 slot of the CM5 carrier on hand. The Hailo-8L sits in the other (Build Guide v9.6 sec 8.1). Nothing for storage plugs in over USB.
Bench-prove both slots (drive and Hailo, together) before the wiring is committed.
No spinning drives, no mirror, no 8 TB pair. Losing the archive drive loses science audio, not the boat. Highlights go home over Starlink anyway.
The friction regimen still applies to every USB device: cable clamped within 2 in of the plug, plug staked.
| Box | What is in it | Bus | Holes (all glands unless noted) |
|---|---|---|---|
| Central node (near battery/busbars/compute) | The Nucleo plus one MCP23017 I2C expander for the convenience outputs. Release channels: enable relay, MOSFET switch modules. Monitoring: two 1-wire chains, BME280 barometer and humidity on one I2C chain. Steering: the reversing DC drive for the Ewellix CAHB-22 with current sensing, behind the de-energized-closed power-cut relay (sec 5.6). The 8-channel relay module (Compute Bay Plan v9.6 sec 11 owns the channel map: five switched-load channels plus the charge bypass). A Taidacent DIN-rail surge module on the CAN drop. Inputs: the two Water Witch states, the post wind sensor (Calypso UART) and encoder A on one trunk from the aft junction (likely; count at fit-out). Release runs aft leave on two trunks to the transom terminal box | Control | ~13 + vent (count at fit-out) |
| Release Node B (stern, near the transom boxes) | Second minimal release chain: Nucleo on screw carrier + CAN transceiver with a Taidacent DIN-rail surge module on the drop, 24 V to 5 V converter, fuse, enable relay, dual-MOSFET modules (2 live channels: drogue lid B-horns, boxes 1 and 2), mini DIN strip. No opto board, no short relays | House | 4 + vent |
| BMS box (the box as bought, beside the battery box, never inside it; "the watchman box") | Electrodacus SBMS0 + JKBMS JK-B2A8S balancer, its shunt, the 1k 1 W resistor and a small terminal strip where the five MPPT remote on/off cables parallel on EXTIO4 (the charge pause, sec 6.7) | None (USB to Pi A) | 4 + vent |
| Aft junction (stern, passive, no Nucleo) | DIN terminals for the post cables (one auto-reset breaker per post feed). The Teltonika TSW202, the boat's only network switch, always on from the house branch, with its Mean Well DDR-120B-48 supply at 54 V. Hardware photocell relay for the nav lights, motoring-white relay. The switched-load feeds arriving from the central node on five 16/2 duplex runs. Encoder A landing | House bus end (the switch has its own always-on 24 V branch) | ~21 + vent, 10 RJ45 couplers; measure the box before the loom |
| Iridium A box, Iridium B box | RockBLOCK 9603 carrier + tiny 5 V converter | None (serial to its Pi) | 3 each + vent |
| Detector mini-box | AS3935 breakout only | None (7-core lead to WD-B) | 1 + vent |
| VESC A box (VESC B box MOVED TO BOAT 2 at v9.5, GPIO review, Mike 2026-09-05: Boat 1 sails with one VESC) | Purchased sealed enclosure; Flipsky 75200 Pro V2.0 (sec A7) | Control | 9 |
| Battery box | Per Battery Box Building Guide v9.6. No brain inside. Five holes: two PG16 for the 2 AWG mains, one PG13.5 for the combined signal bundle (nine taps shared by the SBMS0 and the JK, and the DS18B20 chain), the Gore vent, and the 2 in relief stub in the forward wall. No heater feeds and no gas-sensor leads (both deleted, do not re-propose) | None | 3 glands + vent + relief stub |
| Transom terminal box (stern, passive, box already on hand) | DIN strip only: the central node's release trunks land and fan out to the drogue boxes and JSD actuators (the aux-rudder and pull-pin runs are deleted at v9.4 with the single-actuator steering decision; the bridle tension switch entry is DELETED at v9.5) | None | 6 + vent |
| Starlink enclosure | Deck; sacrificial (enclosure heater DELETED at v9.5) | None | 2 |
| Native devices, no drilling | em-trak (house, channel fails to ON); the wing charging coil pair (a 24 V inductive coil pair bought on Amazon 2026-09-15, part number and rating to be logged at the next mail audit; its cable takes the can-wall gland; the NearFi pair and its M12 cordsets are deleted). The Ewellix steering actuator is a plain DC device on the central node's drive, not a bus device (A14) | Tee + drop each |
Box count (v9.4). Plastic: central node 1, Release Node B 1, transom terminal 1, BMS 1, junction 1, Iridium 2, detector 1 = 8 plastic boxes, plus the two aluminum compute boxes. (The FSESC box is deleted with the backup rudder actuator.)
Every plastic box carries a membrane vent (Gore-type, M12 screw-in, 1/2 in hole), so it breathes through a membrane instead of pumping moist air through its gland threads.
The circuit is the release recipe translated to screw-terminal boards.
No soldering iron anywhere, and no push-on jumper pins anywhere in release logic. Every logic line between a Nucleo and a relay, MOSFET, or opto board lands under a screw terminal or a latched connector. Each Nucleo is staked in its screw carrier after burn-in.
The base:
Release channels. 24 V feed, fuse, enable relay, release bus, then per channel a MOSFET switch module to the solenoid (flyback diode under the terminal screws) or burn wire.
In-service exercise schedule:
| Cadence | Test | What it proves |
|---|---|---|
| Monthly | Exercise the enable relay with release-bus voltage readback | Releases nothing, because every MOSFET stays off |
Monitoring. Two DS18B20 chains: chain 1 is the six battery probes, chain 2 is compartment air. A BME280 barometer breakout on the I2C chain (the box breathes through its membrane vent, so it reads outside pressure). A humidity breakout. Nothing else: no branch current meters (pump running is known from the Water Witch 24 V output), no MOV or SPD status contacts (every clamp is checked with a meter at each refit and at the post-strike check), no VE.Direct reader (the five MPPT VE.Direct ports carry the charge-pause cable, sec 6.7). No high-water sensor, anywhere. No gas sensors and no gas breakout (deleted, do not re-propose).
The cell-temperature trip on chain 1 is the battery trip. 150 F on any pack probe, or a 10 F rise in five minutes, isolates the bank, HIBERNATE, alert. Release after one hour with every probe below 120 F, with a warning-only alert at 131 F (System Settings Sheet v9.6 sec 4; set by Mike 2026-09-15, to confirm).
Steering drive and power-cut relay. The Ewellix CAHB-22 is a plain 24 V DC actuator with no bus, so the central node drives it (sec 5.6): a reversing DC drive built from the two 24 V 30 A relay modules on hand as a reversing pair, or a 24 V H-bridge, with a current-sense module so the node can see a stall.
Ahead of the drive sits the power-cut relay: a 24 V 30 A relay module in series with the actuator feed, wired de-energized-closed. With no coil power, whether the node is dead, the driver is dead, or a wire is broken, the actuator stays fed. Cutting a jammed or runaway actuator takes a live node. The screw is self-locking, so the rudder holds where it was left with no power on.
Switched loads. A 24 V 8-channel relay module, driven through the MCP23017 expander. Compute Bay Plan v9.6 sec 11 owns the channel map; this plan follows it. Five switched-load channels are used: 1 Starlink power, 2 science group, 4 em-trak, 6 bilge pump manual-on (primary pump only; the backup pump stays on its Water Witch alone), 7 motoring whites coil. Channel 8 is the charge-continue bypass (sec 6.7). Channels 3 and 5 are spare. The network switch is NOT on a relay channel: the TSW202 is always on from the house branch, because shedding it would blind Starlink and cut the Pi-to-Pi Ethernet. Camera power saving is per-port PoE control on the switch (sec 6.1). The photocell relay for the nav lights is hardware only, with no Pi backup channel.
The em-trak channel is wired through its normally closed contact, so the transponder fails to ON. A dead node leaves the boat answering ships. The other channels de-energize off, and nothing that keeps the boat alive runs through them.
Inputs. The two Water Witch states, the post wind sensor and rudder encoder A (PWM, sec 5.6) come in from the aft junction, and the bilge manual-on feed goes out the same way. One 12-core trunk is likely enough; count at fit-out. The post wind sensor is the Calypso ULP CMI1018 (UART/I2C), wired straight to a Nucleo UART; it needs no level chip. (Allocation: the CMI1018 UART/I2C unit on the post, the CMI1017 RS485 unit at the wing tip.)
The second, physically separate release chain: its own fuse off a house-side branch, its own enable relay, its own MOSFET pairs, the same two-conditions interlock, commanded over the house bus.
Each drogue box's two release channels split across the nodes, one triggered from the central node and one from Release Node B, through separate glands on separate routes. Either channel alone cuts the strap.
Release Node B's live channels are exactly the two drogue B-horns. Contents: enable relay B, MOSFET channels for horn B of boxes 1 and 2, CAN with a Taidacent DIN-rail surge module on the drop. Five Nucleo pins. No opto board, no short relays. Box entries: CAN, 24 V, drogue 1, drogue 2, vent.
One box death can no longer remove every release on the boat.
Node addresses, message IDs, release channels, probe serials, I2C addresses, and opto inputs are printed on each lid. See Node Boxes Explained v9.6 for the drawings.
This table is the authoritative release-channel map (owned here since v9.4 - four documents previously pointed at each other and none owned it). Node Boxes Explained, the Box Wiring Sheets, the Compute Bay Plan, and lid labels and firmware message names all follow this table. Command vocabulary (renamed at v9.4): the two-step is ENABLE then RELEASE; status wording is "channel N released" (the "confirmed" suffix of v9.4 went with the confirm switches at v9.5; deployment is confirmed by the speed drop and the aft camera).
| Channel | Action | Central node | Release Node B |
|---|---|---|---|
| 1 | Drogue box 1 lid strap | Horn A (3 coils) | Horn B (3 coils) |
| 2 | Drogue box 2 lid strap | Horn A (3 coils) | Horn B (3 coils) |
| 3-6 | The four JSD release actuators (T8 fid pulls) | All four | - |
| 7-9 | RESERVED for Boat 2 (aux-rudder release: solenoid 1, solenoid 2, burn wire). UNWIRED on Boat 1 - the aux rudder is deleted from Boat 1 with the 2026-08-27 single-actuator steering decision; Boat 2's redundancy is the aux rudder, so the numbers are held | (Boat 2) | - |
| 10-15 | RETIRED at v9.4 (were the six steering pull-pin paths - the pull-pin stations are deleted with the single-actuator decision: no backup actuator, no pins, nothing to release; numbers retired, not reused) | - | - |
| 16 | RETIRED at v9.4 (was the trim-linkage decouple pulse - a wire to a device never designed; every wing-trim failure already ends in feather by passive means, so the channel and its gland are deleted; the number is retired, not reused) | - | - |
The drogue-lid channels keep the physical split: each box's two horns reach their nodes by separate glands on separate routes. On Boat 1, Release Node B's live channels are exactly the two drogue B-horns.
NMEA 2000 Micro-C (M12 A-coded 5-pin), 120 ohm, 22 AWG power pair.
CONTROL BUS: power tee [END] -- [4-port: Pi A, Pi B] -- VESC A -- central node [END]
HOUSE BUS: power tee [END] -- [4-port: Pi A, Pi B] -- em-trak tee
-- Release Node B tee -- aft junction [END]
(VESC B drop and the Electrak tee: Boat 2 only)
24 V, injected once per bus from its own busbar branch, through a 5 A automatic-reset breaker (12/24/48 V rated) at a tee next to the multiport.
Load is about 1 A. Micro-C parts are 60 V rated. The Nucleos need the 5 V converter. VESC A set to 250 kbps. The TSW202 at the aft junction is not fed from the pair; it has its own always-on 24 V branch.
| Device | Bus | Why |
|---|---|---|
| VESC A, central node (which drives the Ewellix steering actuator) | Control | Steering, release channels, motor |
| em-trak, Release Node B, aft junction (bus end) | House | So no single bus failure strands both release nodes (VESC B is Boat 2 only) |
| Pi A, Pi B | Both | Only the Pis touch both |
Island rule. No emergency-bank device on either bus except through an isolated transceiver, which is Pi B's HAT.
NMEA 2000 (em-trak), VESC's own frames, and ours from the two nodes. They coexist, and the Pi's software handles each. There is no J1939 on the hull buses: the Ewellix steering actuator is not a bus device. The only J1939 aboard is the wing pod's own short CAN link between its Nucleo-G431 and the Thomson Electrak MD trim actuator, and that link never touches the hull buses (Compute Bay Plan v9.6 sec 10.2). (Boat 2's Thomson Electrak HD would add J1939 to the control bus; give the VESC a controller ID clear of any J1939 address so the two boats share one software build.)
The primary rudder. The steering actuator is the Ewellix CAHB-22-F4E-3000529-BAA0PT-000, bought 2026-09-10. It is a plain 24 V DC actuator: 10,000 N (about 2,250 lbf), 11.8 in stroke, 0.4 to 0.5 in/s, a self-locking screw (it holds position unpowered, no brake needed), built-in position feedback, no CAN.
Pi A sends the rudder target to the central node over the control bus. The node drives the actuator with a reversing 24 V DC drive: the two 30 A relay modules on hand wired as a reversing pair, or a 24 V H-bridge. The drive carries current sensing.
Rudder angle is read two ways and cross-checked: encoder A on the stock (an AS5048A magnetic encoder; the node reads its PWM output on one timer pin), and the actuator's own position feedback. Disagreement flags a sheared-linkage or encoder fault for shore. Boat 1 steering is single string, accepted deliberately.
Jam logic (commanded effort with no encoder motion, or motor current at stall) cuts drive power within seconds. A stalled actuator burns its windings in 10 to 25 minutes, so this is a protective reflex, not a convenience. The de-energized-closed power-cut relay in the feed is the second cut. No J1939 anywhere on the steering.
Rod-travel life. Ewellix rates the CAHB-22 for about 3.1 miles (5 km) of double strokes. The voyage estimate was about 7 miles (11.2 km) of rod travel on the 190,000-move budget, so the move budget is halved: widen the deadband and lengthen the minimum interval between moves. The numbers go in the System Settings Sheet v9.6 sec 3. Confirm the rated-load basis with Ewellix.
Placement. The foredeck fisheye on a bonded G10 pad: a weatherproof (IP66) 180 degree PoE fisheye, Reolink Duo 3 PoE or Duo 2 PoE class, TO BUY (about $150 to $200). The Reolink FE-P bought for the spot is an INDOOR unit with no weather rating; it goes back or serves indoors. Forward and aft cameras high on the post on bonded pads, tabbed with two layers of biax: two Reolink RLC-810A, not yet bought. An interior dome watching the compute boxes, battery box and bilge. And the UNDERWATER camera: an Axis P1245 Mk II modular sensor head (received) in an angled epoxy-glass through-hull tube low on the hull, fused-silica window, 275 nm UV-C keep-clear ring, aimed at the prop hub and rudder; it also trends fouling and watches the shaft collar anode. The head is replaceable afloat through the inboard cap. Its dry main unit sits inside the boat and is what takes the PoE. The camera is electrically ISOLATED, never bonded: plain UTP Cat6, shield broken at the camera end, kept off the anode circuit. Build detail: the Underwater Camera Port Guide v9.6.
One network switch on the boat: the Teltonika TSW202 at the aft junction (bought 2026-09-07, $155.64). Eight PoE ports and two SFP slots. Every camera, both Pis, Starlink and the interior light driver hang on it. The Netgear GS308EP and GS305EP are out (returns or the shelf), and the small switch inside the Pi A box is gone.
| Port | Device | Power |
|---|---|---|
| 1 to 5 | The five cameras | PoE |
| 6 | The PoE light driver, which runs both interior disc lights (sec 6.4) | PoE |
| 7 | Pi A | data only |
| 8 | Pi B | data only |
| SFP 9 | Starlink, on a copper SFP module (10GTEK SFP-T, two bought) | data only |
| SFP 10 | Spare |
The switch is always on, from the house branch. It is never on a relay channel. With one switch on the boat, shedding it would kill the cameras, Starlink's data path and the Pi-to-Pi Ethernet at once; Starlink would sit there powered and blind. Always-on costs about 72 Wh a day, nothing against the bank.
Camera power saving is per-port PoE control on the switch. Pi A turns a port off or on through Teltonika's documented Web API (the endpoint /api/ports_settings/config), then reads the port's state back over SNMP (the standard RFC 3621 MIB, read-only on this switch). Command and confirmation travel by two different paths, the same discipline as the rudder. Turning off ports 1 to 5 is the zero-camera state; one camera can watch while the others sleep. Pin the firmware version: Teltonika renamed the port setting between firmware 1.01 and 1.05, and code written for one silently stops working on the other. Test on the exact firmware that flies and record the working request. (The switch runs TSWOS, not RutOS.)
PoE budget. The Mean Well DDR-120B-48 supply stays (no upgrade): about 112 W available, cameras about 60 W set aside, plus the light driver at 25 W in 802.3at or 12 W in 802.3af. Both fit, with 22 W or 36 W spare. The 24 V feed current stays about 5.5 A.
Bench checks before the loom (Launch Punch List item 29): 1) turn a port off and on through the Web API, recording the exact request and the firmware version; 2) read the same port back over SNMP; 3) read watts per port and record the command and units; 4) find out whether a per-port PoE power limit or priority exists; 5) power all seven PoE devices at once and read the total.
Two new long Cat6 runs go to the stern with this change, Pi B and Starlink, each through IP67 couplers (to buy). The aft junction grows to 10 RJ45 couplers and about 21 entries; measure the box before the loom (A6).
Hardening rules, all pass/fail:
Post mounting per the Instrument Post Guide: nothing through the tube wall, one bond spur per metal body, cables in their own glands with drip loops.
Outdoor Cat6 on the deck legs, plain Cat6 inside. IR LEDs off in settings.
Two independent receive chains (was three; the RTL-SDR second listener and its whip are DELETED at v9.5, Mike 2026-09-05):
Each ear has its own antenna, so no single antenna fault deafens both.
The em-trak transponder is on the house bus over N2K. Its power channel fails to ON (normally closed contact): software can switch it off to save power when the sea is empty, but a dead node leaves the boat visible.
The transmit whip lives under the panel-free deck patch, stood as near vertical as headroom allows. No AIS antenna above deck.
Ethernet to the TSW202 at the aft junction (SFP port 9 through a copper SFP module; a new long Cat6 run to the stern), power on relay channel 1, declared sacrificial (no arrestor). No enclosure heater (deleted, do not re-propose: the dish has its own snow-melt heater).
Deck penetration: the instrument post's 1 in PVC deck stub. Since 2026-09-15 the GPS, Starlink and Iridium A coaxes come below through it with the post bundle, and the four separate 1 in standpipes and their four compression deck seals are deleted (A13; Instrument Post Guide v9.6 sec 8). Iridium B keeps its own small penetration under its foredeck blister, 18 ft forward, so the two satellite chains share no gland (Pending Amendments v9.7 item 111).
Red and green side lights at the deck edges, plus a white stern light on the post, all on the hardware photocell relay in the aft junction. The two all-round whites on the post come on additionally when motoring.
There is no current sense on the light circuit and no Pi backup channel: both are DELETED at v9.5 (GPIO review, Mike 2026-09-05). Nothing could be done about a dark light at sea, the cameras show the lights, and the photocell relay is hardware only, with no backup.
Fixtures are USCG-marked sealed LED, about 30 Wh per night.
Interior lights (new at v9.6). The PoE Texas AT-LED-24V kit, bought 2026-09-07 ($49.99): one PoE-powered driver and two 4 in LED disc lights, on TSW202 port 6. The driver is a bare board, so it lives in a small dry box of its own. It has no current limiter of its own: do not add a third light to it. The disc fixtures are dry-location parts; mount them where condensate cannot collect and no drip can land in them. They light the interior dome camera for colour, which infrared cannot give. KEPT.
Modem A near Pi A (main bank, always on); modem B near Pi B (emergency bank, always on). Each in a small vented plastic box, serial to its Pi, coax straight to its quarter-wave stub arrestor.
The stub is the only coax protection, because a stub cannot wear out.
Antenna domes: clear acrylic over each stub antenna on a white base plate, rubber gasket, oversize holes and snug screws, and a 1/16 in weep hole at the flange low point.
No arrestor and no bias tee on either GPS coax (Compute Bay Plan v9.6 sec 14.1 rules). A plain stub cannot pass the DC the receiver sends up to its active antenna. A DC-pass gas tube can fail shorted and silently kill the antenna for the rest of the voyage while looking healthy, which is the one failure the surge doctrine forbids on a channel the boat cannot inspect.
The protection is the two independent chains (different receivers, antennas and boxes), the ferrite at the wall, and the low cost of the receiver behind a Starlink GNSS cross-check.
Two boards in the BMS box as bought, next to the battery box, neither of them inside it.
Electrodacus SBMS0, the watchman. It reads:
It watches and has one hand: the charge pause. The SBMS0's EXTIO4 output drives five Victron VE.Direct non-inverting remote on/off cables (ASS030550320, on hand), wired in parallel.
Wiring:
Above about 6 V the chargers run; below 1 V or floating they stop. The pause opens on a high cell, on a pack below 32 F (the SBMS0's low-temperature charge stop, which is the whole cold plan; there are no heaters), or on SBMS0 power loss.
A dead SBMS0 pauses charging for at most 10 minutes; then charging continues. The central node's charge-continue bypass (relay channel 8 through the expander; battery + through a second 1k 1 W resistor onto the yellow bus in the watchman box) closes when the watchman stream has been silent 10 minutes and the battery probes read above 32 F. It opens again when the stream returns. It controls no load, and the Pi controls no charging.
Using the VE.Direct port for the pause cable rules out a GX device, a dongle or a data cable on that controller. Bluetooth is off too (power draw), so there is no per-string data and no VE.Direct reader board. The five MPPTs' own lithium profiles (absorption 27.6 V, float 27.0 V, temperature compensation off, RX pin set to "Remote on/off"; System Settings Sheet v9.6) stay as the backstop if a plug shakes loose.
Bench test: unplug the SBMS0, all five chargers must stop; plug it back, they resume. CONFIRM (C1): the SBMS0 firmware has the low-temperature stop and the EXTIO4 polarity is as assumed.
Its USB serial stream to Pi A is required, not optional (settled 2026-09-16). The load-shedding code takes its state-of-charge figure from this stream, because the SBMS0's shunt is the only current sensor aboard. The SBMS0 base board has no data port; the stream needs the Electrodacus WiFi/USB add-on board (ordered), through the USB isolator into Pi A (entry 7, on the isolated hub). No CAN, no RS-485 from the BMS. The serial reader and the stream-gone-quiet alert stay required software (BMS Software Change Notes v9.6): a quiet stream means eyes lost, and the "no charge for a solar day" alert catches a dead watchman too.
The watchman box and the signal bundle. Both boards sit in the BMS box as bought, beside the battery box (A4). The battery box's one PG13.5 signal bundle carries the nine taps shared by the SBMS0 and the JK, and the DS18B20 chain (A9). The taps and the shunt leads stop at the watchman box; the DS18B20 chain carries on to the central node. There are no gas-sensor leads (both gas sensors deleted, do not re-propose). Software rule (BMS Software Change Notes v9.6 sec 4): 150 F on any pack probe, or a 10 F rise in five minutes, isolates the bank, HIBERNATE, alert; release after one hour with every probe below 120 F; warning-only alert at 131 F.
JKBMS JK-B2A8S, the balancer. 2 A active balancing through the nine taps it shares with the SBMS0 (one bundle, one gland; Mike 2026-09-05). It talks to nothing and controls nothing.
What software does with the stream, the four alerts, and the pause behavior live in the BMS Software Change Notes v9.6.
Battery box heaters: deleted, do not re-propose. The sea keeps the pack above freezing on every leg, winter or summer; the SBMS0 32 F low-temperature charge stop is the whole cold plan. (The wing pod heaters are a different matter and STAY: Wing Power System v9.6.)
Every box entry is a nylon IP68 gland.
| Gland | For | Drill |
|---|---|---|
| PG7 | One cable of 1/8 to 1/4 in | 1/2 in |
| PG9 | Small bundles | 5/8 in |
| PG11 | Actuator bundles | 3/4 in |
| PG13.5 | Paired harness entries; the switched-load duplex runs (five 16/2, three and two per gland, A3 entry 16 and A6 entry 10); the battery box signal bundle | 13/16 in |
| PG16 | 2 AWG (the VESC feed and the battery mains) | 7/8 in |
All glands are nylon IP68, as bought. No brass, no nickel-brass.
Nothing needs to unplug at a box, so no M12 receptacles anywhere. The M12 joints that exist are the tees on the CAN backbones and the molded couplings mid-cable on point-to-point runs.
Two small same-route cables may share one gland a size up, with the voids between the jackets filled with sealant. Fast cure is fine; a filled gland cures before the salt-spray test.
Grouping is allowed only for cables that run the same route to the same box and would be serviced together. Never for release circuits, mains, or cables arriving from different directions.
Current grouped entries: the central node's two 1-wire chains, the BMS box's shared tap bundle, and the five 16/2 switched-load duplex runs, three and two per PG13.5 gland at each end (A3 entry 16, A6 entry 10), sealant-filled.
The M12 disconnect is on the cable, outside the box: at a tee for CAN, and at the molded male-female coupling of a cut cable for point-to-point runs.
Female on the box side of the joint, male on the cable, boat wide.
One genuine mix-31 snap-on per entry, sized to the cable.
| Cable | Core ID |
|---|---|
| Micro-C, Cat6, coax | 0.30 in |
| 12 to 8 AWG feeds | 0.5 in |
| The three bundled phase leads at the VESC | 1.0 in |
A ferrite must never hang on a cable. Bond the core to the bulkhead or settee structure just below the box, with the cable passing through with a little slack on each side. 3M 4200 is allowed for this bond (Compute Bay Plan v9.6 sec 9.4); it is one of only two places 4200 is used on the boat (App B).
Keep them off the hull skin and away from the lightning straps. A cable clamp on a bonded pad a few inches below each box supports the drip loop.
The main fuse is a 300 A Class T at the bank (JIAKUPA kit, bought). The motor branch takes a 250 A MEGA fuse under it. The main bus is two generic 400 A busbars with four 3/8 in studs each (bought); each main branch leaves through a MEGA fuse in its own holder (four holders bought). There is no Victron Lynx anywhere (deleted, do not re-propose).
Every small branch that can be an auto-reset breaker is one: 24 V stud-type auto-reset breakers (bought 2026-09-07), landed on 150 A bar pairs. That covers the CAN power tees (5 A), both bilge pumps (5 A), the box feeds, the post feeds and the relay-board loads. Blade-fuse blocks and most inline holders are out.
Bench trip-test every breaker at about twice its rating, and hold rated current for minutes without tripping. Mount high, terminals down under the covers, with drip loops.
Still fused, on purpose: the release channels (emphatically), the Iridium feeds, the watchdog rails, and the steering actuator feed ahead of the power-cut relay. A fuse there is a deliberate one-shot: a release channel or a voice channel must never cycle on its own.
Surge parts, as bought (Compute Bay Plan v9.6 sec 14.1 rules):
Existing orders stand as recorded in Appendix B. The live order sheet is the BOM Order Status workbook. New or changed since v9.1, corrected to v9.6:
Everything else in Appendix B stands.
Every entry is a nylon IP68 gland (as bought; no brass) unless it says RJ45 coupler or SMA. CAN drops are the male half of a cut Micro-C cable (plugged into the tee outside), bare end through a PG7 gland; point-to-point runs are one Micro-C cable cut in the middle and glanded at both ends. Hole sizes: PG7 1/2 in, PG9 5/8 in, PG11 3/4 in, PG13.5 13/16 in, PG16 7/8 in, SMA 1/4 in, RJ45 coupler ~7/8 in (per part), vent 1/2 in, bond stud 1/4 in. One mix-31 ferrite outside every entry. Compute box bottom face is 11.8 in: up to 14 holes on 1.2 in staggered centres fit with room for vent and stud.
Where: inboard aft of the settee, bonded to the hull below the waterline. Bus: both. Contents: CM5 on the carrier on hand, which seats both the 4 TB NVMe SSD (WD Blue, bought) and the Hailo-8L (three bought); eMMC boot. 2-CH CAN FD HAT (staked). WD-A Nano Every + regulator (default-ON rail switch). GPS A, hiBCTR NEO-7M (bought; position only). IMU A BNO085. LoRa bonnet. A metal screw-mount powered USB hub with a captive cable, the USB isolator on hand on its link, carrying the dAISy, the SBMS0 lead (via the WiFi/USB add-on board), three FTDI USB-to-RS-232 cables and one FTDI USB-to-RS-485 adapter for the science instruments; every cable clamped within 2 in of its plug and the plug staked; FTDI only, udev rules by serial number. ADuM1201 isolator for Iridium A serial; optocoupler for the inter-box alive discrete. 24 V to 5 V 5 A converter, TVS, terminal strip, INA226 on the feed. Vent, desiccant, humidity breakout, bond stud. No box switch (Pi B and Starlink run to the TSW202), no VE.Direct-to-USB cables, no secure element.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | Main 24 V feed, own MEGA fuse at the busbar | PG9 | 5/8 in |
| 2 | CAN control drop (shield drain to bond stud) | PG7 | 1/2 in |
| 3 | CAN house drop | PG7 | 1/2 in |
| 4 | VESC A UART throttle + GPIO kill | PG7 | 1/2 in |
| 5 | Iridium A serial | PG7 | 1/2 in |
| 6 | Inter-box link to Pi B: the main-island-alive discrete only (optocoupler inside) | PG7 | 1/2 in |
| 7 | SBMS0 USB (watchman telemetry, optional) | PG7 | 1/2 in |
| 8 | Ethernet to the TSW202 at the aft junction (switch port 7) | RJ45 coupler | per part |
| 9 | (not drilled at v9.6: the aft uplink and the Starlink run merged into entry 8; Starlink now runs to the TSW202) | - | - |
| 10 | (not drilled at v9.6: Pi B now runs to the TSW202) | - | - |
| 11 | GPS A coax (no arrestor, no bias tee) | SMA F-F | 1/4 in |
| 12 | dAISy AIS coax (below-deck receive antenna) | SMA F-F | 1/4 in |
| 13 | 915 MHz whip | SMA F-F | 1/4 in |
| 14 | (not drilled: the RTL-SDR and its whip are gone. The box is 11 entries + vent + bond stud) | - | - |
| S | Science serial leads from the USB hub (three RS-232, one RS-485): glands to be added here only if the hub lives in this box; count at fit-out, up to three more holes fit under the 14-hole cap (decision open, Pending Amendments v9.7) | PG7 or PG9 | 1/2 or 5/8 in |
| V | Vent | 1/2 in | |
| B | Bond stud 1/4-20 | 1/4 in |
Where: a few feet from Pi A on the settee panel. Bus: both, through the isolated HAT. Contents: CM5 + carrier + CAN FD HAT (staked), 250 GB NVMe SSD in the carrier's M-key slot, WD-B Nano Every + regulator (default-ON rail switch), GPS B (hiBCTR NEO-7M, bought; position only), IMU B (BNO085), ADuM1201 isolator for Iridium B serial, optocoupler for the alive discrete, own DC-DC + TVS to the floating negative, fuse, terminal strip, vent, desiccant, humidity breakout. No bond stud. No secure element. Plastic couplers and glands only.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | Emergency bank feed | PG7 | 1/2 in |
| 2 | CAN control drop | PG7 | 1/2 in |
| 3 | CAN house drop | PG7 | 1/2 in |
| 4 | Iridium B serial | PG7 | 1/2 in |
| 5 | Inter-box link from Pi A: the alive discrete only (optocoupler inside) | PG7 | 1/2 in |
| 6 | AS3935 detector lead, 7-core, under 3 ft, to WD-B | PG7 | 1/2 in |
| 7 | (blank; spare) | ||
| 8 | Ethernet to the TSW202 at the aft junction (switch port 8), UTP, a long run to the stern | RJ45 coupler (plastic) | per part |
| 9 | GPS B coax (no arrestor, no bias tee) | SMA F-F | 1/4 in |
| V | Vent | 1/2 in |
Bus: control. Powered from: main bank branch, own fuse. Contents: NUCLEO-G431 (KB or RB) on Nano screw carrier (staked) + CAN transceiver breakout, with a Taidacent RS-12V/2S DIN-rail surge module on the CAN drop. MCP23017 I2C expander, screw-terminal board, with pull-down resistors on its outputs (convenience outputs only, sec 4.1). Potted 24 V to 5 V converter. DIN rail with DK2.5N blocks. ONE 8-channel relay module: the enable relay and the switched loads per Compute Bay Plan v9.6 sec 11 (five load channels, the charge bypass on channel 8, em-trak on its normally closed contact; see the fit-out flag there on which physical relay the enable function uses). Steering power-cut 30 A relay (de-energized-closed) and the Ewellix reversing drive (the two 30 A relay modules on hand as a pair, or a 24 V H-bridge) with a current-sense module. Dual-MOSFET switch modules (six release channels). ONE 8-channel opto input board (the two Water Witch states). 1N5408 diodes across each solenoid pair. DS18B20 chains x2. BME280 barometer breakout. Humidity breakout. Every logic line under a screw or latched connector. No short relays, no branch current meters, no VE.Direct reader, no gas breakout.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | CAN control drop | PG7 | 1/2 in |
| 2 | 24 V feed | PG9 | 5/8 in |
| 3, 4 | Release trunks to the transom terminal box (drogue node-A channels and the four JSD actuators, actuator power only; the confirms and bridle switches are DELETED at v9.5, so fewer cores; fire pairs 12-14 AWG, 22 AWG for what remains, per Mike 2026-09-05; aux-rudder and pull-pin runs deleted at v9.4; count the cores at loom time) | PG11 x2 | 3/4 in |
| 12, 13 | Steering actuator feed in from its breaker (12 AWG), and the actuator cable out to the Ewellix: the motor pair from the reversing drive plus its position-feedback leads (count the feedback cores from the Ewellix datasheet) | PG11 x2 | 3/4 in |
| 14, 15 | Aft trunk from the aft junction (12-core): two Water Witch states, bilge manual-on, encoder A, wind (Calypso CMI1018 UART). One trunk is likely enough; entry 15 is drilled only if the count at fit-out needs it | PG11 (x2 if needed) | 3/4 in |
| 16 | Switched loads out to the aft junction: five 16/2 duplex runs, three and two per gland, sealant-filled | PG13.5 x2 | 13/16 in |
| 17 | (DELETED at v9.5 - INA226 I2C chain to the branch shunts; gland plugged) | PG7, plugged | 1/2 in |
| 18 | 1-wire chains x2 (two cables, one gland, sealant-filled) | PG9 | 5/8 in |
| 20 | (DELETED at v9.5 - MOV ladder + PV SPD status contacts, not monitored en route; gland plugged) | PG11, plugged | 3/4 in |
| 21 | (not drilled: no gas sensor lead; gas sensors deleted, do not re-propose) | - | - |
| 22 | Charge-continue bypass wire to the watchman box: two cores from relay channel 8 (battery + through the 1k 1 W resistor, and return) | PG7 | 1/2 in |
| 23 to 24 | (DELETED at v9.5 - the VE.Direct JST cables and their PG16 two-hole glands; the five MPPT VE.Direct ports carry the charge-pause cable from the watchman box instead, sec 6.7; holes not drilled) | - | - |
| 25 | Spare | PG9, plugged | 5/8 in |
| 26 | (not drilled: no cabin gas sensor lead) | - | - |
| V | Vent | 1/2 in |
Bus: none. Powered from: the pack, through the SBMS0's own supply lead. Contents: Electrodacus SBMS0, JKBMS JK-B2A8S, the SBMS0 shunt, the 1k 1 W resistor on EXTIO4 and a small terminal strip where the five VE.Direct remote on/off cables parallel (yellow signal wires to EXTIO4 minus, blacks to battery negative; the charge pause, sec 6.7), plus the charge-continue bypass feed from the central node's relay channel 8 landing on the same yellow strip through its own 1k 1 W resistor (a dead SBMS0 pauses charging for at most 10 minutes; then the bypass closes and charging continues). No humidity breakout (no reader in this box; the temperature chain passes through). No gas sensors. Harnesses laced to the plate at each plug; dielectric grease in the plugs.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | The signal bundle from the battery box: nine tap wires shared by the SBMS0 and the JK, the XT1 thermistor and the DS18B20 chain. One bundle, one gland, sealant-filled. No gas-sensor leads | PG13.5 | 13/16 in |
| 2 | Five VE.Direct remote on/off cables ASS030550320 out to the MPPTs (one bundle, sealant-filled) | PG9 | 5/8 in |
| 3 | Shunt / supply leads | PG9 | 5/8 in |
| 4 | USB lead to Pi A (via the WiFi/USB add-on board; telemetry only) | PG7 | 1/2 in |
| 5 | Charge-continue bypass from the central node's relay channel 8, two cores, landing on the yellow terminal strip through its own 1k 1 W resistor | PG7 | 1/2 in |
| V | Vent | 1/2 in |
(Deleted from Boat 1 at v9.4 with the backup rudder actuator; kept here for Boat 2. Not counted in the Boat 1 totals.) Bus: house. Powered from: main bank branch, own fuse. Contents: Flipsky Mini FSESC 4.20, phase-short relay (coil on the box feed), AS5048A encoder B input, potted 5 V converter if the relay logic needs it. Humidity breakout DELETED at v9.2 (no reader in this box). Board conformal-coated.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | CAN house drop | PG7 | 1/2 in |
| 2 | 24 V feed | PG9 | 5/8 in |
| 3 | Motor phases (3 x 14 AWG - this is the Boat 2 FSESC box driving the NEMA-23 backup rudder motor, not the ME1616; the ME1616's 2 AWG phases are in box A7. The change list's "make the 14 AWG row agree with 2 AWG" note was aimed at A7, which already agrees) | PG11 | 3/4 in |
| 4 | Encoder B lead | PG7 | 1/2 in |
| V | Vent | 1/2 in |
Bus: house bus end. Contents: DIN rail with DK2.5N blocks (one auto-reset breaker per post feed). The Teltonika TSW202 switch (8 PoE ports + 2 SFP slots) with the Mean Well DDR-120B-48 on the DIN rail at 54.0 V, sealed; the switch is always on from the house branch, never on a relay. Hardware photocell relay, motoring-white relay. Switched-load feeds from the central node on five 16/2 duplex runs. Encoder A landing. One 12-core sensor trunk out (likely; count at fit-out). Port map: 1 to 5 cameras, 6 PoE light driver, 7 Pi A, 8 Pi B, SFP 9 Starlink on a copper SFP module, SFP 10 spare. Ten RJ45 couplers (nine in use) and about 21 entries.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 to 4 | Post cables: forward camera Cat6 (port 1), aft camera Cat6 (port 2), wind sensor (Calypso CMI1018 UART, through to the trunk), lights | RJ45 coupler x2 + PG7 x2 | per part / 1/2 in |
| 5 | (plugged; was the weather pod) | PG7, plugged | 1/2 in |
| 6, 7 | Camera Cat6: foredeck fisheye (port 3), interior dome (port 4) | RJ45 coupler x2 | per part |
| 8 | Ethernet to Pi A (port 7) | RJ45 coupler | per part |
| 9 | 24 V feed, always on (the switch and its 54 V supply hang on it) | PG9 | 5/8 in |
| 10 | Switched loads in from the central node: five 16/2 duplex runs, three and two per gland, sealant-filled | PG13.5 x2 | 13/16 in |
| 11 to 13 | Load bundles out: lights group, Starlink power, em-trak + science | PG11 x3 | 3/4 in |
| 14, 15 | Trunk to the central node (one likely enough; entry 15 drilled only if the count at fit-out needs it) | PG11 (x2 if needed) | 3/4 in |
| 16 | Bilge leads: the two Water Witch 230-24 pump-running outputs (24 V present = pump running) + the manual-on feed to the primary pump (relay channel 6). There is no high-water sensor | PG9 | 5/8 in |
| 17 | (plugged; was the shaft sensors) | PG9, plugged | 5/8 in |
| 18 | Rudder encoder A lead (PWM output) | PG7 | 1/2 in |
| 19 | Side-light feeds (port and starboard deck edge) | PG7 x2 | 1/2 in |
| 20 | Underwater camera Cat6 from the Axis P1245 main unit inboard (port 5) | RJ45 coupler | per part |
| 21 | Ethernet to Pi B (port 8), UTP, plastic coupler; a long run from the settee (ADDED v9.6) | RJ45 coupler (plastic) | per part |
| 22 | Ethernet to Starlink (SFP port 9 through the copper SFP module); a long run from the deck (ADDED v9.6) | RJ45 coupler | per part |
| 23 | Cat6 to the PoE light driver in its small dry box (port 6) (ADDED v9.6) | RJ45 coupler | per part |
| 24 | (SFP port 10, spare: a coupler only if wanted; not drilled) | - | - |
| V | Vent | 1/2 in |
A7, VESC A (control bus, 9 holes: CAN, UART + kill, battery + and - at PG16, phases x3 at PG16 with the 1.0 in mix-31 clamp outside, sensor lead). The controller is the Flipsky 75200 Pro V2.0, aluminum PCB, power-button variant, $154, ORDERED by PayPal (v9.5, Mike 2026-09-05). Its switch is a maintained on/off in the logic supply, no latch and no firmware, so the board runs whenever battery is present: the button is replaced by a crimped shorted plug. The board has no spark protection, so a precharge resistor (a resistor that lets the controller's capacitors fill slowly before the main contact closes) goes on the 2 AWG feed. COMM (JST-PH 8-pin) and CAN pigtails from Flipsky; JST-PH 8-pin connector pairs x10 on order. The ME1616's SinCos sensor harness is built to the Sep 2 spec (six wires, two connectors, optional divider crimped into the cable, potted, ratios on the label; crimp, no solder). Bench test: power-up with the shorted plug.
A8, VESC B: MOVED TO BOAT 2 at v9.5 (GPIO review, Mike 2026-09-05). Boat 1 sails with one VESC; propulsion loss is a schedule delay because the wing sails the boat (Routing Guide 11.5). The box, its 8 holes (CAN, battery + and - at PG13.5, phases to the contactor x3 at PG16, contactor coil set/reset + sense at PG7), the phase-transfer contactor and its coil driver all go to the Boat 2 column and are not counted in the Boat 1 totals.
Bus: none. Five holes:
| # | Entry | Through | Hole |
|---|---|---|---|
| 1, 2 | 2 AWG mains, + and -, in line with the terminal posts at measured post height, cables straight over the aft G10 plate (aft wall) | PG16 x2 | 7/8 in |
| 3 | Combined signal bundle: nine taps shared by the SBMS0 and the JK (each fused at the terminal with a 30 AWG tap-wire link, sleeved, laced) and the DS18B20 chain (aft wall) | PG13.5 | 13/16 in |
| V | Gore vent, the breathing inlet (aft wall) | M12 | 1/2 in |
| R | Relief stub: 1-1/2 in Sch 40 PVC glassed through the FORWARD wall, high, with a Dometic/Sealand 1-1/2 in toilet duckbill (OEM 385310076) inside it, lips out, flange clamped by a PVC coupling, discharge aimed forward away from the electronics. No screen on the stub: the duckbill is shut except when the box vents, the stub points up at 45 degrees into open cabin space away from the electronics, and the cell-temperature trip is the protection. A bug cap and a fine flame mesh were both considered and dropped. | 2 in hole saw | 2 in |
Deleted, do not re-propose: heaters, gas sensors (in the box or at the cabin deckhead), the high-water sensor, the overboard relief duct with its transom hole and burst membrane. The six DS18B20 pack probes and the cell-temperature trip are the protection. Build detail per Battery Box Building Guide v9.6.
The team-built 8x EVE LF50 + JBD 60 A bank is replaced by a Dyness 24V 100Ah LiFePO4: 25.6 V, 100 Ah, 2,560 Wh (the emergency bank is now 2.56 kWh, was a nominal 1 kWh - Build Guide secs 4.1/4.3 restated), 20.9 x 8.11 x 8.54 in, 42.6 lb, plastic case, built-in 100 A BMS with low-temperature cutoff, ~$360. The unit is its own enclosure: NO box is built, NO hole schedule and NO gland list exist for it, and this appendix entry stops being a placeholder.
What A10 now covers, all four items: 1. Restraint - the real work. 42.6 lb at 20.9 in long must hold in a knockdown and a full inversion: chocked shelf, two straps over the top. 2. Terminals and fuse. M8 posts under a plastic cover, not a gland plate. Insulating boots on both posts, the compartment treated as the enclosure, and an inline 30 A ANL within a few inches of the positive post (protects the cable with room over these tiny loads). 3. Case bond: nothing to do, by design. Plastic case, so no metal can accidentally tie the floating emergency-island negative into the strike path (the Lightning Ground System Guide requires this island's negative to float). 4. One maker question, answer recorded here when it lands: does the built-in low-temperature cutoff block charging? That is the one that matters, because charging a cold LiFePO4 cell is what ruins it. Mike's call, 2026-08-27: the draw side is not a concern (a discharge cutoff sits near -20 F on these packs, far below anything Boat 1's route reaches, and A10 already noted it as academic here; revisit for Boat 2's colder route), and standby current is not a concern either, since the pack floats off the main bus through the Orion-Tr.
Circuit connections unchanged: feed to Pi B, feed to Iridium B, SmartSolar 75/10 in, Orion-Tr float in - now landed at the posts and a small fused DIN strip beside the battery rather than through box glands.
Contents: RockBLOCK 9603 carrier, potted 5 V converter, TVS. Entries: serial to its Pi (PG7), coax straight to the stub arrestor (PG7), always-on power (PG7), vent. A on the main bank near Pi A; B on the emergency bank near Pi B.
Contents: SparkFun AS3935 breakout only. Entries: 7-core lead to WD-B (PG7), vent.
Entries: Ethernet to the TSW202 at the aft junction (RJ45 coupler; SFP port 9 through the copper SFP module), power (PG9). No heater entry. Deck: the instrument post's 1 in PVC deck stub (Iridium B keeps its own forward penetration) since 2026-09-15. The post bundle and all four antenna coaxes (GPS, Iridium A, Iridium B, Starlink) go through that one stub and its 1 in liquid-tight flexible conduit, which runs from the stub up the outside of the post about 10 in and into a side face, with the seal round the wires at the stub (Instrument Post Guide v9.6 secs 7 and 8). The four separate standpipes and their compression deck seals are deleted. Lay the whole bundle out on the bench and check it passes before the stub is glassed in, and make the coax connectors up after the cables are through.
em-trak transponder (house, tee + N2K drop; channel fails to ON). Wing charging coil pair: a 24 V inductive coil pair bought on Amazon 2026-09-15, part number and rating to be logged at the next mail audit, feeding the wing charge controller like a panel; its cable takes the can-wall gland. The pod needs about 20 W if both heater pads run from the coupler, so either it carries the pad load or the pads run from the pod packs (Wing Power System v9.6 secs 4.2 and 8.3). The NearFi 200 pair, its M12 cordsets and its G10 standoffs are deleted. Wing pod / fairing / nose locker (potted glands per Wing Power System v9.6); inside the pod, the Thomson Electrak MD trim actuator (MD24A025-0050CNO) hangs on the pod Nucleo-G431's own CAN through the same transceiver breakout as the central node, 120 ohm terminators at each end, no switched feed (Compute Bay Plan v9.6 sec 10.2); that short bus never touches the hull buses. The Ewellix CAHB-22 steering actuator has no bus and no box of ours: its cable runs to the central node (A3 entries 12, 13), where the reversing drive, the current sense and the power-cut relay live. The Electrak HD tees on the control backbone are Boat 2 only.
Bus: house. Powered from: house-side branch, own fuse. Contents: NUCLEO-G431 (KB or RB) on Nano screw carrier (staked) + CAN transceiver breakout with a Taidacent RS-12V/2S DIN-rail surge module on the CAN drop, potted 24 V to 5 V converter, fuse, enable relay (24 V 30 A module), dual-MOSFET modules for the two live channels (horn B of drogue boxes 1 and 2), mini DIN strip. Five Nucleo pins. No shorting relays, no opto board: the release bus is dead until the enable relay closes and each channel also needs its MOSFET on, so two conditions remain. No humidity breakout (no reader in this box). Same two-conditions interlock as the central node; all logic under screws.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1 | CAN house drop | PG7 | 1/2 in |
| 2 | 24 V feed | PG9 | 5/8 in |
| 3 | Drogue box 1, node-B channel | PG7 | 1/2 in |
| 4 | Drogue box 2, node-B channel | PG7 | 1/2 in |
| 5 | (DELETED at v9.4 with the pull-pin stations - single-actuator steering, no pins) | - | - |
| 6 | (DELETED at v9.4 with the pull-pin stations) | - | - |
| 7, 8 | (DELETED at v9.5 - the opto sense leads added at v9.4 go with the opto board; back to a 4-hole box plus vent) | - | - |
| V | Vent | 1/2 in |
The dying-gasp reflex of v9.2/v9.3 is DELETED: a power-dead boat with wet lab valve 1 frozen open is an acceptable state. Valve 2 (the duty valve) is auto-return and springs closed on power loss by itself, and that alone is the boat-level protection; valve 1 (stays-put 5-wire, open 24/7) needs no drive and open is safe - the hull is still owned by the seacock and by valve 2. Nothing is wired: no valve close lines to the backup island, no fuses for them, no pin on Pi B or WD-B. The valve types themselves are specified once, in the Wet Lab doc, which this appendix points at.
Bus: none. Contents: DIN strip only, one landing per conductor; no electronics, nothing to fail. Fuses stay at the central node end.
| # | Entry | Through | Hole |
|---|---|---|---|
| 1, 2 | Release trunks in from the central node | PG11 x2 | 3/4 in |
| 3, 4 | Drogue box 1 and 2, node-A channels out | PG9 x2 | 5/8 in |
| 5, 6 | JSD T8 actuator jackets out | PG11 x2 | 3/4 in |
| 7 | (Aux-rudder release out - Boat 2 only; unwired on Boat 1 at v9.4) | - | - |
| 8, 9 | (DELETED at v9.4 - pull-pin runs gone with the single-actuator decision) | - | - |
| 10 | (DELETED at v9.5 - bridle tension switches; drogue deployment is confirmed by the speed drop and the aft camera; gland plugged) | PG9, plugged | 5/8 in |
| V | Vent | 1/2 in |
Totals (recomputed at v9.6 from the entries above; Boat 1 boxes only, A5 and A8 are Boat 2): no M12 receptacles on any box; glands PG7 35 (A1 6, A2 6, A3 4 of which 1 plugged, A4 1, A6 5, A7 3, A11 6, A12 1, A15 3), PG9 14 (A1 1, A3 3 of which 1 plugged, A4 2, A6 3 of which 1 plugged, A13 1, A15 1, A16 3 of which 1 plugged), PG11 14 (A3 6 of which 1 plugged, A6 4, A16 4; two more if the second aft trunk is drilled at fit-out), PG13.5 6 (A3 2, A4 1, A6 2, A9 1), PG16 7 (A7 5, A9 2); RJ45 couplers 12 (A1 1, A2 1, A6 9, A13 1; 13 if the spare SFP port gets one); SMA 4 (A1 3, A2 1); membrane vents 10 across the boxes plus the battery box Gore vent; ferrites ~60 (one per live entry; estimate carried, not recounted). Not yet counted: the science serial glands (A1 row S). Every gland is nylon IP68, as bought. The M12 disconnects that exist are the tees on the two backbones.
The live order sheet is the BOM Order Status workbook; this appendix is the summary as of 2026-09-15. Older order history is in the Changelog.
On hand or ordered: Axis P1245 Mk II (received 2026-09-04); Teltonika TSW202 ($155.64) and two 10GTEK SFP-T copper SFP modules (2026-09-07); PoE Texas AT-LED-24V light kit ($49.99, 2026-09-07); Calypso ULP wind pair CMI1017 + CMI1018 ($999.98, 2026-09-10); Ewellix CAHB-22 steering actuator ($273.70, 2026-09-10); BNO085 x3 and the LoRa bonnet ($141.61, 2026-09-08); Hailo-8L x3 (Mouser, $85 each; one flies on Boat 1, two are Boat 2 stock, not spares); WD Blue 4 TB SSD ($442.99, arriving 2026-09-21); 2 AWG marine wire ($134.99, arriving 2026-09-16) and RIOCAN 2 AWG lugs; 300 A Class T kit, 250 A MEGA, four MEGA holders, two 400 A busbars ($40.20), auto-reset breakers and 150 A bar pairs (2026-09-06/07); Taidacent RS-12V/2S CAN surge modules x2; Fairview FMSP1066 stubs x2; Arduino Nano Every (the watchdog board); FAIR-RITE ferrite clamp; Thermal Grizzly Minus Pad 8; Flipsky 75200 Pro V2.0 ($154, PayPal); JST-PH 8-pin pairs x10; Electrodacus WiFi/USB board (PayPal); five ASS030550320 charge-pause cables; MCP23017 expander, BME280 breakout, 1k 1 W resistors; PG16 x2, PG13.5 x5 and the PG7-19 kit (all nylon IP68); two Belleville packs; the Reolink FE-P (indoor unit, to return or reassign, sec 6.1); the NEO-7M 4-pack (2026-09-09; two fly, two are Boat 2 stock).
Out (deleted or returned; do not re-propose): Netgear GS308EP and GS305EP; the 5-port box switch; Victron Lynx distributors; ZED-F9P moving-base pair and the two extra ADuM1201 isolators; LCJ CV7-OEM wind pair; Thomson Electrak HD24B026 (cancelled 2026-09-11, never charged); GPS bias tees and GPS stub stock; SM712 CAN clamps; secure element breakouts; both gas sensors; battery box heater parts; the RTL-SDR and its whip; the VE.Direct reader board; the Delrin pull-pin stock; the bolt-on hull anode; the Phoenix Contact NearFi 200 coupler pair with its M12 cordsets and G10 standoffs; quad-UART HATs (no serial add-on boards, sec 8); the four antenna standpipes and their four compression deck seals (one deck stub now, sec 6.3 and A13); the two-part polyurethane topcoat and its primer (one-part Rust-Oleum Marine Topside everywhere above the waterline, Build Guide v9.6 sec 3.1).
To buy: FTDI USB-to-RS-232 cables x3, FTDI USB-to-RS-485 adapter x1, metal screw-mount powered USB hub; Cat6 and IP67 couplers for the Pi B and Starlink runs to the stern; a small dry box for the PoE light driver; TVS diodes 1.5KE33CA; the Ewellix reversing drive and current-sense parts (to be chosen); Reolink RLC-810A x2; MOV thermal-disconnect modules x4 and PV SPD modules (plain, no status contacts); nine 30 AWG tap-wire fusible links; membrane vents x9; Tef-Gel; Berkland 4000 UV sealant; tinned 18/4 shielded cable for the wing trailing edge, 50 ft (Amazon B0D9JV83HJ, $36.99), which was never ordered.
Sealant rule: butyl for the compute-box bolt bedding; Berkland 4000 UV for glued boxes and pads; 3M 4200 in exactly two places, the ferrite-core bonds (Compute Bay Plan v9.6 sec 9.4) and the instrument post cusp caps; never 5200 or microballoons under a heat-carrying bond (sec 2.5).
Drawn 2026-08-23 for the v9.1 set; revised 2026-09-15 for the v9.6 set. History, decisions, and the glossary: see the Changelog.