Circuit Sauna · rev A.1 · 2026
Field electronics are rated to −40 °F (−40 °C); winters aren't. Circuit Sauna keeps cold-exposed boards inside their operating window on 12/24 V DC — and reports temperature, heater current, duty cycle, and faults over WiFi/MQTT, so a dead heater is a 2 am alert instead of a spring-time discovery.
Every incumbent enclosure heater — STEGO, DBK, the panel-shop house brands, including STEGO's low-voltage DC line — is thermostat-only: no telemetry, no fault reporting, no way to know it's dead until the site is. Circuit Sauna measures its own heater current, so a burned-out element, a stuck switch, or a detached probe is a distinct alarm within minutes. For unattended sites, proof of life is the product.
Power flows fuse → element → two independent thermal cutouts → MOSFET → current shunt. The logic side taps off through reverse-polarity protection into a 9–72 V-rated regulator, so a charging or equalizing 24 V bank (28.8–31 V) is inside spec, not past it.
1. A 185 °F (85 °C) bimetal cutout on the element mount — works with the MCU dead or mid-flash. 2. A one-shot 219 °F (104 °C) thermal fuse that latches open on true runaway: a failed switch means safe-and-dead (and reported), never weeks of oscillating heat. 3. A 41 °F (5 °C) thermostat on the MOSFET's gate: below 41 °F the element heats open-loop even when it's too cold for the ESP32 to boot — the classic smart-heater failure is that the brain freezes before it can turn the heat on — yet the contacts carry microamps and a live firmware always retains authority. 4. Watchdog, sensor-sanity window, 80% duty ceiling, current-based fault discrimination — every fault published over MQTT.
Service-first physical design: a gloved tech at −4 °F (−20 °C) swaps the controller (sockets), the fuse (spare in its own holder), the element, either thermostat, or the sensor — each in under five minutes, none with a soldering iron, and the two look-alike thermostats can't be confused because their roles are silkscreened at their plugs. Diagnostic LEDs face the enclosure door with blink codes printed on the board itself.
Every line verified orderable against distributor listings (July 2026) — including the two parts the first draft got wrong: the original MOSFET pick turned out to be EOL and under-rated against the TVS clamp, which is exactly why the parts list gets adversarially checked before anything ships.
| Role | Part | Source |
|---|---|---|
| Controller | Espressif ESP32-DevKitC-32E | Digi-Key ~$10 |
| Switch | Infineon IRLZ44NPBF (55 V logic-level) | Digi-Key / Mouser |
| Sensor | Analog Devices TMP36GT9Z | Mouser / Digi-Key |
| 3.3 V rail | RECOM R-78HB3.3-0.5 (9–72 V in) | Digi-Key ~$10 |
| Overtemp | Sensata-Airpax 67F085 · Cantherm DF104S TCO | Digi-Key |
| Protection | Littelfuse P6KE39A TVS · 1N5819 | Digi-Key / Mouser |
| Current sense | Ohmite 13FR050E 0.05 Ω 3 W | RS / element14 |
| Field wiring | Phoenix Contact MSTB 2,5 plugs · Keystone 3557 ATO | Digi-Key |
| Element | Nichrome 80 (TEMCo / Pelican Wire / Omega NI80) | direct |
| Enclosure | Bud NBF (NEMA 4X, −40…158 °F) | Digi-Key / Mouser |
| PCB / assembly | JLCPCB protos · MacroFab or AdvancedPCB (US) · AISLER, Eurocircuits (EU) | direct |
Parts-count prediction (Telcordia-style FIT budget, 4× outdoor-environment factor): 4980 FIT → ~23-year field MTBF, honestly quoted as a 8–69 year band. Top contributors:
| Contributor | FIT (field) | Share | Note |
|---|---|---|---|
| ESP32-DevKitC module | 1600 | 32% | complex sub-assembly: MCU module + USB-UART + LDO + 38 pins |
| Connectors (plugs/sockets/fuseholders) | 800 | 16% | 5 separable interfaces; top field-failure class |
| Buck (RECOM R-78HB3.3) | 600 | 12% | qualified module, ceramic caps; clone LM2596HV boards would be ~300+ |
| TMP36 + remote probe lead | 480 | 10% | die 20 + lead/termination in thermal cycling 100 |
| Nichrome element+mount | 400 | 8% | low watt-density; terminations dominate, wire itself robust |
| PCB + THT solder joints | 400 | 8% | 2-layer, big thermal swings; THT joints beat SMT in cycling |
Every top-five failure mode fails safe and reported: heater off, MQTT silent or alarming — silence is itself an alarm. The one mode that could command heat (shorted switch) sits under 1% with three independent backstops. Full model in the repo.
Launch price $129. Loaded COGS $42.08 at qty 100 — 67% gross margin, with the reliability-grade regulator and pluggable service wiring already in the number. One avoided truck roll pays for the unit.
| Qty | Parts | Labor | Loaded COGS | GM @ $129 |
|---|---|---|---|---|
| 10 | $43.30 | $14.58 | $59.62 | 54% |
| 100 | $32.10 | $8.75 | $42.08 | 67% |
| 1,000 | $24.37 | $4.67 | $29.91 | 77% |
North America first: WISPs and land-mobile-radio site operators (mountaintop relay cabinets — WISPA's 800-member vendor channel; firms like Day Wireless with 75+ owned radio sites), solar O&M fleets (NovaSource-class utility-scale down to residential truck-roll operations), state DOT road-weather stations and the mesonets (Montana's network is growing from ~216 to ~540 stations by 2028), cold-climate EV charging operations, rail signal maintainers, and drone-dock installers. Europe next (CE via the Radio Equipment Directive — the WiFi radio, not voltage, is what pulls the product in): Nordic fast-charge installer networks, Alpine snowmaking control, agri-weather resellers, and the road-weather OEMs as a long-game component sale.
Gas dispensers themselves are Class I hazardous locations (NFPA 30A) — this product does not go inside them, full stop. But most forecourt electronics live outside the classified zones: price signs, canopy comms cabinets, car-wash controllers, kiosk POS — even tank-gauge consoles are required by their own manuals to sit in non-hazardous locations. That's sellable today through the same petro-equipment distributors; a Class I Div 2 variant is the someday-unlock for the dispenser cabinet itself, and it waits for revenue.
Design rev A.1 (July 2026), generalized from a unit already deployed and surviving winters, then hardened by a 20-agent adversarial design review (16 confirmed findings, all fixed) and a 49-item sourcing/market verification sweep. Phases: KiCad layout and five protos now → cold-chamber (incl. a 41 °F ↔ 5 °F bootstrap-hysteresis oscillation profile) and ten pilot builds by October → at least three instrumented pilot units in real sites through winter 2026–27 → rev B, certification go/no-go, and the first 100-unit run in spring 2027 off a season of MQTT data. Mains variants and hazardous-location ratings wait for revenue.