Three nodes, one panel, one keg.
The power plant sits outside: panel, MPPT controller, battery, main fuse. One fused 12 V feed enters through a cable gland to a five-slot fuse block. A fault on one node opens one fuse and leaves the other two running. That is the keg-scale version of redundant power distribution in a data hall.
- 0
- 0
- 10
- $1,000
- 2026-10-26
Power path.
100 W monocrystalline panel, doubling as the sunshade, into an MPPT controller with a LiFePO₄ profile, into a 12.8 V nominal, 20 Ah battery with integrated BMS. The main fuse sits within 7 inches of the positive post, per marine wiring convention. Inside the keg, slots B1 to B3 feed three 12 V to 5.1 V buck converters, one per node; B4 feeds the switch; B5 feeds the 3.3 V sensor rail and is the spare.
| Load at the 12 V bus | Idle W | Full W | Grade |
|---|---|---|---|
| Raspberry Pi 5, per node, ×3 | 3.0 | 7.0 | PLAUSIBLE |
| 5-port gigabit switch | 2.0 | 2.0 | PLAUSIBLE |
| Sensor rail and MPPT self-use | 0.5 | 0.5 | PLAUSIBLE |
| Buck conversion efficiency | 0.90 | PLAUSIBLE | |
| Cluster total at bus | 12.8 | 26.1 | derived |
Sixteen weeks, four gates.
- 01
Energy
Panel, MPPT controller, LiFePO₄ battery, the main fuse, the fused 12 V bus, and the first node as sensor host.
Node 1 runs from the fused bus; both INA226 monitors report sensible watts in and watts out.
- 02
Compute
Nodes 2 and 3, the gigabit switch, wired network, and a three-node load test under a stop rule set in advance.
Three-node load test completes under the pre-set stop rule without a brownout.
- 03
Keg
Primer and safety-orange paint, IP68 gland and RJ45 bulkhead, the passive heat path to the keg wall, sensors, desiccant.
Water-only leak test of the empty keg, then a seven-day sealed soak with a flat humidity trend.
- 04
Deploy
Land, tethered water, and a ground-based orbit analog: simulated eclipse power cycles and radiation-only cooling.
Eclipse cycles logged; radiation-only cooling run completes with SoC below throttle.
Bill of materials, estimate against actual.
Prepared September 30, 2026. Only two lines were checked against live listings: the Pi 5 boards and the keg. The rest are estimates. Actual prices are written beside each estimate at purchase, and the gap is the project’s first cost-variance dataset. Assumed on hand: a laptop, a drill, hand tools.
| Item | Qty | Estimate | Price status | Actual | Variance |
|---|---|---|---|---|---|
| 01Energy | |||||
| Digital multimeter | 1 | $25 | Estimate | — | — |
| LiFePO₄ battery, 12 V 20 Ah, BMS | 1 | $80 | Estimate | — | — |
| Solar panel, 100 W 12 V mono | 1 | $80 | Estimate | — | — |
| MPPT charge controller, 10–20 A | 1 | $40 | Estimate | — | — |
| Buck converters, 12 V to 5.1 V 5 A | 4 | $25 | Estimate | — | — |
| Fuses, holders, Powerpole, wire | 1 | $35 | Estimate | — | — |
| Sensors: 2× INA226, 4× DS18B20, BME280 | 1 | $30 | Estimate | — | — |
| 02Compute | |||||
| Raspberry Pi 5, 1 GB | 3 | $135 | Checked 9/2026 | — | — |
| microSD, 32 GB high endurance | 3 | $30 | Estimate | — | — |
| 5-port gigabit switch | 1 | $20 | Estimate | — | — |
| 03Keg | |||||
| 5-gallon ball-lock keg, 304 stainless, new | 1 | $130 | Checked 9/2026 | — | — |
| Self-etching primer, safety-orange enamel | 1 | $30 | Estimate | — | — |
| Heat path: heatsinks, 3 mm Al plate, pads | 1 | $35 | Estimate | — | — |
| IP68 gland, RJ45 bulkhead, sealant, step bit | 1 | $30 | Estimate | — | — |
| Outdoor Cat6 and patch cables | 1 | $30 | Estimate | — | — |
| Desiccant and humidity cards | 1 | $10 | Estimate | — | — |
| 04Deploy | |||||
| Water tether kit | 1 | $25 | Estimate | — | — |
| Panel stand / canopy | 1 | $40 | Estimate | — | — |
| Parts subtotal | $830 | — | — | ||
| Contingency, 20.5%: price swings, shipping, tax, mistakes | $170 | — | — | ||
| Project budget | $1,000 | — | — | ||
Ten claims under test.
Each design claim is paired with the instrument that will test it and the expectation it must beat or replace. Where a value is withheld as a lesson, the builder produces the number before buying the part. Results land in the build log and replace these rows one at a time.
| Claim | Grade | Instrument | Test | Pre-build expectation | Measured |
|---|---|---|---|---|---|
| Panel open-circuit voltage matches its label | PLAUSIBLE | Multimeter, DC V | Full sun, then half-shaded | Near label Vₒₑ; shading moves voltage little, current a lot | not yet |
| Main fuse is sized to the wire | LESSON | Multimeter, continuity | Continuity across installed fuse; fused-lead V equals battery V | Rating per I_wire ≥ I_fuse ≥ 1.25 × I_cont,max; value recorded in log | not yet |
| Buck holds 5.1 V under load | PLAUSIBLE | USB-C inline meter | Output before any Pi, then under load | ≥ 5.0 V at 3 A; no Pi under-voltage flag | not yet |
| Per-node idle and full-load watts | PLAUSIBLE | INA226 #2 | Idle, then 1–4 jobs on one node | 2.5–3.5 W idle; 6–8 W full; second job costs less than the first | not yet |
| Three-node full load at the bus | PLAUSIBLE | INA226 #2 | Three-node stress under stop rule | About 26 W, ±20% | not yet |
| Night idle fits usable battery energy | PLAUSIBLE | INA226 #2, MPPT SoC | One overnight log, sunset to sunrise | ≥ 26 Wh above the 51 Wh reserve at dawn in winter; more in summer | not yet |
| INA226 range covers bus current | LESSON | INA226 #2 vs multimeter | Compare at 1–5 A | Agreement within shunt tolerance; range per 81.92 mV / R_shunt | not yet |
| Shell rise is small; the SoC-to-plate interface dominates | SPECULATIVE | 4× DS18B20 | Three-node load, 60 min, sealed | Shell 5–10 K above ambient; SoC under 80 °C | not yet |
| Keg stays dry when sealed | PLAUSIBLE | BME280 | Seven-day sealed soak, then tethered water | Flat humidity trend; any rise is a leak | not yet |
| Daily harvest scales with peak sun hours | PLAUSIBLE | INA226 #1 | Seven clear and seven cloudy days | About 70 Wh per PSH; derate 0.6–0.8 | not yet |
Rules that gate every step.
Two further constraints follow from the hardware. Pi GPIO pins operate at 3.3 V and are not 5 V tolerant, so the sensor rail is 3.3 V only. Every load test stays above the BMS low-voltage cutoff; a node that browns out has already reset, and there is no recovery step to prove.
Fuse first.
The main fuse goes on the battery's positive terminal before anything else is connected.
Never pressurize the keg with electronics inside.
The pressure-relief valve is never disabled or painted over. Leak tests use water on the empty keg, never gas.
Water tests are tethered only.
A pool, a private pond, or a permitted dock. No open water in v1.
Orbit is a ground analog.
Simulated eclipse power cycles and radiation-only cooling. Nothing launches.