Example project: a soil probe from first board to fleet
Example (fictional): Loamwick is not a real company and this is not a client project. It shows, week by week, how one engagement could take a battery-powered sensor from first boards to signed updates on a pilot fleet.
Key characteristics
- Status
- Fictional example, not a client project
- Device
- LW-1 soil probe and solar gateway
- Pilot fleet
- 40 probes and 4 gateways
- Firmware
- Zephyr, MCUboot, two 448 KB slots
- Rollout
- Waves of 4, 12 and 24 probes
Example (fictional). Loamwick is not a real company, the LW-1 is not a real product and this is not a client project. It is a worked example of how we run an engagement, and the same names and numbers appear in our tools and on other pages so the site tells one consistent story.
Loamwick is a two-founder startup in Denver with one contract hardware engineer. Its product, the LW-1 soil probe, measures soil moisture and temperature every 5 minutes. Each probe runs on two AA lithium cells and talks Bluetooth LE to a solar-powered gateway, which forwards readings over LTE-M cellular to Loamwick’s cloud. The probe uses a Bluetooth LE microcontroller with 1 MB of internal flash. Rev C boards are the prototypes; Rev D boards will be the trial units.
The goal: 40 probes and 4 gateways running on test plots on Colorado’s Eastern Plains, eight weeks after the first firmware review, with a way to fix the firmware once the probes are in the ground.
The plan at a glance
| Week | Stage | What happens |
|---|---|---|
| 1 | Review | Firmware review and written plan |
| 1 to 2 | Bring-up | Rev C prototype boards brought up and measured |
| 2 to 6 | Build | Application firmware on Zephyr |
| 4 to 6 | Ship | Update pipeline: signing, MCUboot, two slots, staged waves through the gateway |
| 8 | Ship | Field trial starts on v1.0.0 |
| During the trial | Sustain | Maintenance release v1.0.2, then v1.1.0 in three waves |
| End | Sustain | Team handover to Loamwick’s first firmware engineer |
Week 1: firmware review
Loamwick sends the Rev C schematics, a few boards and the demo firmware its hardware engineer wrote. The demo reads both sensors and talks to the gateway. It has no bootloader, no image signing and no way to roll back, and one image fills the flash. Probes shipped like that could never be fixed in the field.
The written plan ranks four risks: no update path; little flash for two image slots; an unmeasured battery requirement; and an eight-week schedule with no room for a second board spin. It recommends a bring-up sprint on Rev C, a product firmware build, and update pipeline setup in parallel from week 4.
The requirements go into the firmware requirements worksheet, each with a way to verify it:
| ID | Requirement | How it is verified |
|---|---|---|
| FW-001 | Read soil moisture and temperature every 5 minutes | Bench test against a reference probe for 72 hours |
| FW-002 | Run 3 years on two AA lithium cells | Measured average current on 5 boards, battery life calculator |
| FW-003 | Accept signed updates from the gateway and roll back on a failed self-test | Pull power at 10 points in the transfer |
| FW-004 | Refuse images not signed with the production key | Try an image signed with a test key |
| FW-005 | Reconnect to the gateway within 10 minutes after it restarts | Restart the gateway 20 times and log reconnect times |
| FW-006 | Factory test finishes in under 60 seconds per probe | Time 50 boards on the test fixture |
The RTOS is chosen with the RTOS selection worksheet: Zephyr scores 113 of 125, FreeRTOS 93 and bare metal 75. The weights favor bootloader and update support, a maintained Bluetooth LE stack and long-term security fixes.
Weeks 1 to 2: bring-up of Rev C boards
The Rev C boards go through the board bring-up checklist: every rail measured, the debug probe reading the chip ID, clocks measured, every sensor answering at its address, flash erase, write and read verified across the whole part, the radio working at short range. Problems go to Loamwick’s hardware engineer with measurements attached, in time to fix them on Rev D. The pin map and clock tree go into Loamwick’s repository.
Weeks 2 to 6: application firmware
The firmware wakes every 5 minutes, reads both sensors, sends the readings to the gateway and goes back to sleep. Power is measured on five boards: 4 µA asleep, 3.2 mA for 150 ms at each of the 12 sensor wake-ups an hour, and 6 mA for 2.5 ms at each of 3,600 radio events an hour. The battery life calculator turns that into an average of 24.0 µA and an estimate of 12.1 years, assuming 85% of a 3,000 mAh rating is usable and 1% self-discharge a year.
The weekly status reports the currents and the estimate, and says plainly that it is not a promise: cells lose capacity in the cold, and winter on the plains is cold. The 3-year requirement has a wide margin on paper, and the field will check it.
Weeks 4 to 6: the update path
The flash layout, checked with the flash budget calculator:
| Region | Size | Holds |
|---|---|---|
| MCUboot bootloader | 48 KB | Signature check, swap and revert logic |
| Image slot 1 | 448 KB | The running firmware |
| Image slot 2 | 448 KB | A downloaded update waiting to be tested |
| Settings | 32 KB | Calibration and gateway pairing |
| Logs | 48 KB | Reset reasons and update history |
| Total | 1,024 KB |
The 412 KB application image fills 92.8% of the 444 KB a slot can hold after the bootloader’s header and trailer. 25% growth would make it 515 KB, 71 KB too big, so the decision log records a plan: Rev E moves the second slot to an external flash chip.
How a release reaches a probe:
- CI builds the release from a tagged commit and stamps the version into the image header. Version strings follow the version scheme builder, so a release candidate reads like
1.2.0-rc.2+b217.g3f9c2a1.revd. - A signing service in Loamwick’s own cloud account signs the image. The production key was created there, never leaves it, and Fumbar never held a copy. A separate development key signs bench builds. The key ceremony record goes into the handover documents.
- Loamwick’s cloud offers the release to the gateways serving the current wave.
- Each gateway downloads the image once over LTE-M and sends it to its probes over Bluetooth LE. The Bluetooth LE update time calculator estimates 21.6 seconds of transfer and 36.6 seconds including verify, swap and reboot, to be confirmed on the real gateway.
- The probe stores the image in slot 2 and reboots. MCUboot checks the signature and version, then swaps the images.
- The new image runs its self-test: sensors in range, gateway connected, watchdog quiet. If it passes, it confirms itself. If not, MCUboot reverts at the next reset.
- The probe reports its version, and Loamwick’s cloud counts updates and rollbacks per wave.
Relaying through the gateway also keeps cellular data small. With the over-the-air data cost calculator’s default overhead and retry assumptions, one download is 0.456 MB, so a rollout to 4 gateways uses about 1.8 MB, against about 18 MB if all 40 probes had their own cellular link.
Before the trial, the path must pass: power pulled at 10 points during the transfer without leaving a probe unable to boot; images signed with a test key refused; an older version refused; an image built for another hardware revision refused; a deliberately broken image reverted.
Before shipping, the Rev D trial units are programmed with the bootloader and a signed v1.0.0, each probe gets its own identity, and the debug port is locked. The factory test is timed against the 60-second target.
Week 8 and the trial
The trial starts on v1.0.0. The weekly status now adds fleet signals: probes reporting, reconnects and resets. The first maintenance release, v1.0.2, is the first time the pipeline runs on the real fleet.
After 30 days, moisture readings in wet clay drift upward. Version 1.1.0 fixes it. Its notes, from the release notes builder:
Loamwick LW-1 soil probe firmware 1.1.0
Released 2026-11-20, replaces 1.0.2
Hardware: Rev C, Rev D
Security
- The probe now refuses images signed with the retired development key
Added
- Battery voltage in every report
Changed
- Sensor wake-up interval is now set from the gateway
Fixed
- Moisture readings drifted upward after 30 days in wet clay
Known issues
- Probes on Rev C hardware report temperature 0.5 °C high above 40 °C
Installing
- Rollout: Staged in waves
- If the update fails: Device rolls back automatically to the previous version
The rollout follows the plan written in the update policy worksheet, with 48 hours of soak between waves:
| Wave | Probes | Starts | Which probes |
|---|---|---|---|
| 1 | 4 | Day 0 | Loamwick’s own bench units |
| 2 | 12 | Day 2 | Trial probes |
| 3 | 24 | Day 4 | The rest of the trial fleet |
The halt rule is written before wave 1: if any probe rolls back or goes silent, the rollout stops until someone explains why. The staged rollout planner does the same arithmetic for larger fleets.
Example (fictional). The v1.1.0 release report to Loamwick’s founders: 40 of 40 probes updated and confirmed; 0 rollbacks; no probe went silent; halt rule not triggered. Battery voltage now arrives with every reading and sets the baseline for the battery check. Not yet known: whether the drift fix holds past 30 days in wet clay; the next report compares those plots with the reference probe. Open: the Rev C temperature offset, and the Rev E flash change.
The handover
Near the end, Loamwick hires its first firmware engineer, and the engagement closes with a team handover against the handover checklist. The repository has been Loamwick’s since week 1, so the engineer starts with a clean build from a fresh checkout and CI under Loamwick’s account. Walkthroughs of the architecture, the release procedure and the hardware notes are recorded. Then the engineer builds, signs through Loamwick’s signing service and rolls out one release while we watch. After that, contractor access is removed from every system and shared credentials are rotated.
Loamwick ends up with the firmware and its history, the tests, the signing setup with keys it always held, the rollout runbook, the release reports and a firmware engineer who has already shipped a release.
The numbers and the tools behind them
| Number in this story | Tool |
|---|---|
| Zephyr 113, FreeRTOS 93, bare metal 75 of 125 | RTOS selection worksheet |
| 24.0 µA average, 12.1-year estimate | Battery life calculator |
| 448 KB slots, 92.8% full, 71 KB short with 25% growth | Flash budget calculator |
| 21.6 s transfer, 36.6 s with swap and reboot | Bluetooth LE update time calculator |
| 0.456 MB per cellular download | Over-the-air data cost calculator |
| Release notes for 1.1.0 | Release notes builder |
To plan your own path, see how we work and the engagement types, or request a firmware review.
Send us your board and your plan.
A firmware review reads your schematics, code, flash layout and update plan, and ends with a written plan: what to build first, what could go wrong in the field and how updates will reach your devices.
Or email sales@fumbar.com, or call (720) 675-4202.