A soil moisture sensor built for a community allotment in Aarhus worked on the bench for a fortnight and died in the ground in four days. The board was fine. The battery was fine. The sensor's exposed probes had corroded, because they were held at a constant voltage in wet soil, which is an electrolysis cell rather than a measurement. Powering the probe only for the twenty milliseconds of each reading extended it to a season.
That fault is not a programming fault and not really a component fault. It is the thing this topic exists to teach: a circuit is a physical object in a physical place, and the difference between a bench and a field is where most projects are lost.
The physics you need, which is less than you fear
Almost nobody working at this level needs the mathematics of electronics. What is needed is a handful of relationships used constantly and a habit of checking them.
Ohm's law: volts equal amps times ohms, and it sizes every resistor you will place. A light emitting diode wanting 20 milliamps at 2 volts, on a 5 volt supply, needs to drop 3 volts, which is 150 ohms. That single calculation covers most of the resistors a beginner places, and doing it once properly is worth more than reading about it three times.
The second is the one that explains the allotment failure and a great many others. A microcontroller pin can source only tens of milliamps, and a motor wants amps. Anything with a coil or a heater in it is driven through a transistor or a relay rather than directly, and connecting it directly is the most common way a first board is destroyed. The pin does not warn, and it does not fail gracefully.
The third is about measurement rather than power. A floating input reads whatever is nearby, so an unused pin needs a pull up or a pull down. An input connected to nothing is not zero. It is an antenna, and it will read as a button being pressed at random, which produces a project that works when you are touching it and misbehaves when you are not, and that symptom sends people looking for a software bug that does not exist.
The instrument that answers most questions
A multimeter is the whole starter kit, and three of its settings do nearly everything.
- Continuity: it beeps when two points are connected, which finds a broken joint in seconds
- DC volts: measured across a component, with the circuit powered
- DC amps: measured in series, with the circuit broken and the meter inserted into it
The third is where beginners destroy their meter, because measuring current requires breaking the circuit and putting the meter in the path, and putting a meter set to amps across a supply is a short circuit through a fuse. Getting that distinction wrong once is nearly universal, which is why it is worth reading before the first purchase rather than after.
The order, and the two directions people arrive from
The order that works is the small amount of physics, then a circuit on a breadboard that does something, then a microcontroller, then whatever the project actually is. The reason it matters is that the two upper layers hide the two lower ones, and a fault that lands in the bottom two is unreadable to somebody who started at the top.
People arrive from two directions and stall in opposite ways. Programmers arrive at a small computer on a board, in your project with the code already easy and no instinct for the physical layer, and they get the allotment failure: correct software, corroded probe. People arriving from the physical side build a working circuit and stall the first time it has to make a decision. Both are missing exactly one of the two halves, and both are far closer than they feel.
The allotment sensor sits at the top of that ladder and failed there, which is the ordinary place to fail. Its code was correct on the first day and on the last, its circuit was correct throughout, and what defeated it was four days of wet soil against two exposed contacts. Nothing on the bench could have reported that, and nothing in the machines, the networks between them, and keeping both standing is more consistently underestimated than the gap between a thing that works and a thing that keeps working unattended.
Bench to field
How far a working prototype sits from something that survives outside
Placed by how much of the remaining work is physical rather than electronic. A judgement, made here, from one soil sensor.
Judged by what stopped the Aarhus sensor and what would have to change for each stage. Indicative rather than measured.
- Reads correctly on a deskWhere a project feels finished, and where a fortnight of testing said it was
- Runs on its own batteryThe power budget arrives, and every constant draw becomes a deadline
- Sealed and connectorisedDamp, and the joints that were fine indoors
- Survives four months of soilThe probe powered only during a reading, which is 20 milliseconds a minute
It says nothing about difficulty of the electronics, which barely changes across the four. What changes is that the questions stop being electrical and start being mechanical and chemical.
Where people stall
The stall specific to this topic is that a circuit gives no error message. Software fails loudly and in a place; a board that does not work presents as nothing happening, with no line number and no log, and the instinct built by every other subject on this site is useless. The replacement instinct is to measure rather than to reason: check the supply arrives, check the ground is common, check continuity at each joint, and do it in that order every time. That method is the content of building one that works on a bench and it converts a mysterious board into three quick questions.
The second stall is the jump from bench to deployment, which is the allotment story exactly: connectors, damp, temperature, vibration and power budget are a second subject, and they are the one that decides whether a project survives, especially once boards start reporting to something else in boards that talk to something else.
What to open, in 2026
An ESP32 board is the sensible default for anything that needs to report a reading somewhere, costs a few euros, and includes wireless, which is what displaced the earlier boards that needed a separate module for it. Arduino remains the better first board for learning, because its ecosystem assumes nothing and its failure modes are gentler, and it is free and open source in both hardware and software.
A multimeter at around thirty euros is the correct first purchase and a more expensive one buys accuracy that nobody at this level needs. It replaced guessing, which is not a joke: the majority of the time lost on a first project is lost to not measuring.
For designing a board rather than wiring one, KiCad is free and open source and has replaced the free tiers of commercial packages that limited board size, and it is what the physics you need and no more leads to once a circuit is worth making permanent. What none of this covers is the point where the thing being controlled is a factory rather than a plant pot, which raises questions about safety and certification that a hobby project never has to answer.