SouthPing · the probe's power source

Battery Pack Wiring

Six 18650 cells become one 20.7 Ah [C] battery, and that battery reaches the rest of the probe through two wires and one plug. This is what joins what, in what order, and which of those joints carries the whole probe's current.

Companion to buoy-parts-book.html, which carries the same power source from the plug onward. Topology from ADR 0005 and architecture-power-and-wiring.md; geometry from the 2026-09-03 integrated-platform note.

Before anything else

SoC-match the six cells to within ~50 mV of each other before you join them. Nothing in this design limits the balancing inrush between mismatched neighbours — the moment two cells at different states of charge are bridged, the difference is dumped through whatever joins them, at whatever current the copper allows. Measure all six, charge or rest them together until they agree, and only then fit the bus.

Every joint below is soldered to a contact, never to a cell. The cells are protected button-tops that press against spring and plate contacts — nothing is soldered or welded to a battery anywhere in this build.

The pack, and where its two wires go

Battery_Platform — six printed bays bays at 21.0 pitch · Ø18.6 cell in a 19.0 bay [C] NCR18650GA · protected NCR18650GA · protected NCR18650GA · protected NCR18650GA · protected NCR18650GA · protected NCR18650GA · protected − + − bus bare solid 18–20 AWG in the 1.100 × 111.2 slot six joints, one per bay + bus same bar, other end wall 20 AWG silicone — red +, black − F 1.6 A both leads lace through a base hole BEFORE the joint ⚠ polarity is NOT keyed by the shell JST-PH 2-pin — the fuel gauge’s ports dictate it + − to the fuel gauge’s LEFT port the parts book has the rest Two thermistors ride here too NTC #1 ‖ 200 kΩ → charger THERM pad cold gate as built (ADR 0018): stops at −2.3 to −0.4 °C, resumes just above 0 [C] NTC #2 → carrier J7 (IO13, IO39) the report, in software separate parts, by design All six in parallel 6 × 3450 mAh [D] → 20.7 Ah [C], 3.0–4.2 V the pack voltage IS the cell voltage: nothing is in series, so no balance lead and no BMS to fit protection = the cells’ own PCBs Why 20 AWG these two wires carry everything the probe draws — the modem’s 0.45 A [M] burst included — and all charge current back inward worst legitimate draw ~0.6 A [C]; the fuse is 1.6 A
Six cells, twelve contacts, two bus bars, two wires, one plug. The only current path out of the battery is the pair at the bottom right — which is why the fuse sits in it, why it is 20 AWG, and why both leads are laced through the platform before their joints.

Build order

Steps 1 and 2 cannot be swapped. Everything after the bus is soldered is difficult to undo without heating a joint next to a cell.

  1. Match the cells. Measure all six open-circuit. If any pair differs by more than ~50 mV, rest or charge them together until they agree. A mismatched pair balances through the bus the instant you close it, and nothing in this design limits that current.
  2. Fit the twelve contacts — one spring and one flat plate per bay, uxcell 16.5 × 16 mm nickel-plated. They are fitted by hand after printing; the end wall backs each one over 12.000 mm unbroken, and each stands 1.801 mm proud of the wall into open air [C].
  3. Lay each bus bar in its slot — bare solid 18–20 AWG copper down the 1.100 × 111.2 mm channel in the end wall, and solder it to all six contacts on that side. Do this with the bays empty: the cells go in last, and a soldering iron does not belong next to a lithium cell.
  4. Solder the two leads — 20 AWG silicone, red to the positive bus, black to the negative — after lacing each one through its hole in the platform base. Any later pull then lands on plastic instead of on the joint.
  5. Put the fuse in the positive leg, 1.6 A, in series. Not the negative leg: a fuse in the return leaves the positive rail live against the hull and everything in it after the fuse has blown. How it is metered, soldered and read is the fuse rule under Soldering the joints (FW-28).
  6. Solder the leads to the JST-PH pigtail — red to the pin the gauge treats as positive. There is nothing to crimp: the pigtail is pre-made. The shell does not key polarity for you; confirm it against the gauge's own marking before the first power-up, and meter the plug for sign before it mates (the connector section has the rule).
  7. Drop the cells in, button-tops toward the plate contacts, and fit the 3 mm foam pad over them.
  8. Check before power: pack voltage at the plug (3.0–4.2 V, never negative), continuity from each cell to its bus, and no continuity between the two buses. The audit of 2026-10-15 puts the bus-to-bus and lead-polarity checks before the cells go in, since with the cells seated an ohms reading across the buses reads the cells; that waits on the project lead’s go-ahead (FW-25).

What each joint carries

JointCarriesWhy it is what it is
cell → contactup to ⅙ of the loadA pressure contact, not a joint. Nothing is soldered or welded to a cell anywhere in this build
contact → busup to ⅙ of the loadSoldered. The printed channel holds the bar straight and takes the mechanical load off the joint
bus → leadeverythingThe single point where the whole pack's current leaves. Laced for strain relief before soldering
the two leadseverything20 AWG silicone. ~0.6 A [C] worst legitimate draw, the modem's 0.45 A [M] burst included; all charge current flows back the same way
fuse, + legeverything1.6 A, ~2.7× headroom over 0.6 A [C]. Re-rated by ADR 0014 from 2.5–3 A — a 3 A fuse in front of a 0.6 A load lets a 2 A fault [C] run indefinitely inside a sealed hull
JST-PH plugeverythingNot a choice — the fuel gauge's ports dictate the connector. The gauge sits inline, on the charger's side of the fuse: pack → left port, charger → right

The parts, by name

What is settled, what is owned, and what nobody has picked yet. The fuse and its holder were decided on 2026-10-13 (the fuse part itself is in question, FW-28) and the connector is not a choice; only the bus bar and the pack leads have no product chosen.

PartSpecStock
bus barbare solid 18–20 AWG copper, laid in the 1.100 × 111.2 mm channel in each end wallnot on any list — no product chosen
pack leads20 AWG silicone — sized for the current each carries, silicone because it stays flexible coldthe BOM carries one ~$20 line for "JST leads, silicone wire" — no product chosen
fuse1.6 A slow-blow, a square radial-lead micro fuse soldered inline in the + lead — decided 2026-10-13 (uxcell T1.6A). Re-rated from 2.5–3 A by ADR 0014: the old figure assumed a 1.3 A peak the supply never sees; 1.6 A is the standard step nearest the 1.5 A that ADR 0014 sized it to3 needed, 20 owned (Amazon B0F9L6YSTC, $5.26, ordered as 15, counted 20 [M]); the first fitted to the bench pack 2026-10-18. Five of the twenty are bad — three found open (2026-10-25; 10-28 under a reversed plug; 11-03 at no load, though heat-sunk), two resistive (pack 3, 11-02). The part itself is in question and the spares want a recount, 14 by the tally and 12 by the events (FW-28)
fuse holdernone — the fuse's 19 mm legs are soldered into the + lead and covered with heat-shrink; a glass fuse in a clip holder, and a PTC, were rejected for a soldered partdecided 2026-10-13; FW-28 reopens it — a different fuse, or a holder
the connectorJST-PH 2-pin, 100 mm pigtail — not a choice, the fuel gauge's ports dictate the family10 owned, one in each of the three charging chains; listed as "battery/panel leads" — the panel use is gone, and the combiner spec has the panels soldered to the combiner. As built, every panel lead ends in an XH plug into a socket on the combiner's pad pair, on all three combiner boards; the second set's three plugs were crimped red-to-minus and caught by the sign check. Accept or solder is open (FW-29)

Because the pigtails are pre-made, there is no crimping in this build. You cut one to length and solder its two wires to the pack leads. No crimp tool, no loose contacts, no housing to load.

⚠ The wire rule and the connector cannot both be obeyed

The wiring rules say 20 AWG silicone for the pack. The fuel gauge dictates JST-PH. JST's own documentation gives the PH series a wire range of AWG 32–24 and a rating of 2 A at AWG 24 [D].

20 AWG does not fit a JST-PH contact. It is four gauges over the largest wire the crimp is specified for. So one of two things has to be true, and the tree has never said which:

One thing does line up: the 1.6 A fuse sits below the connector's 2 A rating [D], so a fault the fuse tolerates is one the contacts can carry (the audit of 2026-10-15 doubts it, since this part's time-current curve is unrecorded — FW-25). That ordering is worth keeping whichever way the gauge goes. Tracked as PROC-19.

The same contradiction holds at the carrier end: its J5 is a JST-XH port, and an XH contact takes AWG 22–28. The harness leads bought for the carrier are pre-crimped 22 AWG.

Soldering the joints

Four kinds of joint, in the order you make them, and the rule that governs all of them: the iron never comes near a cell. Every joint below is made with the bays empty or on a lead that is not yet in the hull.

  1. Contact to bus bar — the six joints down each end wall. Tin the bar and the contact tab separately, then bring them together and heat the joint until the solder flows, not the solder itself. The printed channel holds the bar straight, so you are not fighting alignment while the iron is on. Nickel-plated contacts take solder readily; the plating is there for exactly this.
  2. Bus bar to pack lead — one joint per leg, and the one that carries everything. Lace the lead through its hole in the platform base first. Strip about 5 mm, tin the strands so they cannot splay, hook the wire over the bar, and flow it. Sleeve it in heatshrink afterwards: this joint sits inside a sealed hull for months and nothing else insulates it.
  3. The fuse's two legs, into the + lead — the fuse rule since 2026-11-02 (FW-28). Meter the fuse first on the lowest ohms range (2k on the Klein MM325, where a good one reads 0.000). Grip each leg with pliers between the body and the joint as a heat sink, give each joint a few seconds of iron, then heat-shrink it. Before the hull closes, read the drop across the fuse with the board drawing: 0.00 V, or the fuse comes out. Soldering heat is the suspect in the fuses that failed, but a heat-sunk one also went open, so this is a rule, not a cure.
  4. Pack lead to pigtail — if you step the gauge down at the connector. Stagger the two joints along the cable so the red and black splices do not sit side by side, tin both ends, and heatshrink each separately before a common outer sleeve.

A cold joint here is the failure you will not find. It reads fine on a meter at the bench, then goes high-resistance under the modem's 0.45 A [M] burst months later, inside a hull nobody can open. A good joint is shiny and has wetted the copper; a dull, grainy or blobbed one gets reheated with fresh flux, not left alone.

The connector, and which way round it goes

A JST-PH 2-pin pigtail, 100 mm — you own ten, one in each charging chain. It is not a choice: the fuel gauge's two ports are JST-PH, so the pack side is fixed by the gauge.

Since PROC-18 closed, the pigtail does not go to the charger. It goes to the fuel gauge's left JST-PH port; a second lead leaves the gauge's right port for the charger's LiPo Batt socket. The gauge sits inline, on the charger's side of the fuse, so it reads the pack only while the fuse is good: with a resistive fuse, board 3's gauge read the charger's 4.23 V over a 3.54 V cell [M]. Two gauges have answered with the right ID and a fixed, plausible reading against a cell at another voltage [M]. A gauge must be seen to move: read it against a meter, and before and after the modem rail comes on (a live one falls 0.1–0.19 V within a second [M]), and treat a reading that does not move as no reading (FW-26).

PinWireGoes to
1red, + the positive bus, through the 1.6 A fuse
2black, −the negative bus, direct

⚠ The shell does not key polarity for you. A JST-PH housing mates only one way round mechanically (whether it also latches is unrecorded, and retention is undecided: FW-25), but nothing stops the wires being soldered to the wrong ends of it — and a pre-made pigtail's red and black are only a convention, not a guarantee. Ring the pigtail out with a meter against the gauge's own marking before the first power-up. Then, every time, meter the plug for sign before it mates: red probe on position 1, black on the ground position, a plus sign expected, on the unmated plug, before it goes into the gauge, the charger or J5. One pack plugged into a gauge with + and − swapped opened the fuse and killed a MAX17048 and both owned bq24074s, which have no reverse protection. The bench cable's XH end has no key against J5; a pack lead that mates one way only is wanted (FW-27).

The pack is one node, not a supply

The cells, the charger's LIPO pad and the carrier's J5 are all one electrical point. Nothing draws through the charger — it only pushes current into that node when the sun is up. So these two wires are not "the charging wires": they are the probe's entire power path, in both directions.

The fuse in them is not the primary protection, and the record is explicit about that since the pack went back to protected cells: each cell's own PCB cuts at about 8 A (no datasheet for that figure is in the repo), and a short pulls through all six at once and trips them together. The fuse is redundancy — kept because it is cheap, not because it is the thing standing between a 20.7 Ah [C] bank and a harness fault — and the audit of 2026-10-15 disputes this (FW-25): below the ~48 A [C] at which all six cell boards trip together, the 1.6 A fuse is the only protection. The first uxcell was fitted to the bench pack on 2026-10-18; the fuse fitted on 2026-10-15 was a Littelfuse 1.5 A very-fast, a different part.

On the carrier board the node arrives at J5, a 5-way JST-XH port: battery positive on pin 1, ground on pin 5, the three between vacant. Behind pin 1 sits Q1, a MOSFET that blocks a reversed pack — but it protects only the carrier. The gauge and the charger sit ahead of it and have no such guard, so the sign check above, before every mate, still decides whether they survive. The carrier's own 1.1 A [D] polyfuse protects only the 3.3 V regulator's branch; the modem's supply taps the pack before it, so this pack fuse is the only fuse on the modem's path. The firmware's low-voltage lockout at about 3.2 V at rest protects the mission, not the cells: at 3.0 V the loaded modem, ESP32 and DS18B20 are already below their minimums, while each cell's own board fires at ~2.5 V [D]. About 18,540 mAh of the 20.7 Ah are usable down to 3.2 V at +25 °C, 17,700 at +2 °C [C]. FW-05 is blocked and FW-20 (the TX threshold against the voltage chain) is open, so 3.2 V is a plan, not built.

Open, and not to be guessed at the bench

HULL-23
The twelve contacts have no retention design. They are fitted by hand after printing and there is no generator feature holding them. How they stay put through a Drake Passage winter is undecided.
HULL-27
Which end takes the spring. The note records one spring and one plate per bay but not which goes where. The drawing above puts the spring on the flat negative end, where it takes up cell-length tolerance — that is the usual convention, not something this tree has decided.
HULL-27
The bus-bar channel is a reading, not a specification. The 1.100 × 111.2 mm slot is unexplained by the generator; the 2026-09-03 note infers its purpose from the fact that 18 AWG solid copper is 1.024 mm. The spec models no bus bar at all.
HULL-24
Bay clearance is 0.200 mm a side — tight for ASA before elephant's foot — and the bays leave 3.900 mm over a 3 mm foam pad [C].

Part of SouthPing — southping.org. This page is licensed CC BY 4.0: reuse it, adapt it, publish it, with credit to SouthPing and a link back. Source code in the project is Apache-2.0. Manufacturers' datasheets and product photographs remain theirs and are not covered by this grant.