A soil sensor that has to prove itself first

A reading of 62 is not a watering instruction.

It is a number produced by a particular probe, at a particular depth, in a particular growing medium. Until those details mean something, adding a pump mostly gives an uncertain measurement more influence.

This version stops at observation. A separate test pot, a compatible low-voltage reader, a scale, and a notebook are enough to ask whether the probe is tracking a useful change. The plant you care about does not have to be the test instrument.

Pick one question the setup can answer

"How wet is the soil?" sounds simple but hides several different questions. How much water was added? How much remains in the pot? How is it distributed? Is the region beside the probe representative of the roots? What does this particular plant need?

The first project question should be narrower: does the raw reading change consistently as this test pot becomes wetter and then drier? That is something a modest setup can investigate without pretending to know the right watering schedule for every plant.

Use a documented probe and a compatible reader. Follow the manufacturer's voltage, connection, insertion, and exposure limits. Some probe assemblies tolerate soil contact while their connector and electronics do not. Do not assume that a green circuit board is waterproof.

Keep the reader and connections above the pot and protected from spills. Use low-voltage equipment suitable for the arrangement. Disconnect power before changing connections or moving wet components near electronics.

Make a pot that can be compared with itself

Use the same growing medium as the intended plant pot, but begin with a separate test container. Record the container, drainage arrangement, amount of medium, and probe position. Mark insertion depth and orientation so removing and replacing the probe is a deliberate change, not an unnoticed reset of the experiment.

Decide how the pot will be weighed. Does the saucer go on the scale with it? Is drained water removed before weighing? A measurement that sometimes includes the saucer and sometimes does not can create an impressive but meaningless trend.

Keep the weighing routine consistent. Record watering times and the amount added where practical. Allow the setup to settle according to the probe and medium's behavior before deciding that a momentary jump represents the whole pot.

The scale is not a direct calibration of the probe. Pot mass can change for reasons beyond the small region the sensor measures. It is an independent observation that helps reveal whether the sensor's story makes sense alongside the rest of the setup.

Save the raw values

A percentage is tempting because it looks finished. Avoid it initially. A scale from zero to one hundred can conceal an arbitrary mapping just as easily as it can communicate a useful measurement.

Record the raw reading, units or converter range where known, and the reading method. If several samples are combined, retain enough information to understand that choice. A median of a short group and a single instantaneous reading are not the same measurement procedure.

Here is a blank record structure, not a set of invented results:

Time:
Pot mass and weighing arrangement:
Raw probe readings:
Probe depth / orientation:
Water added since previous reading:
Visible drainage:
Room or placement changes:
Notes:

Do not smooth away inconvenient jumps before checking them. A sudden change may be noise, a loose connector, a shifted probe, or a genuine local change. The original values help distinguish those possibilities later.

Follow more than one cycle

One afternoon can suggest a relationship. Several wet-to-dry cycles can show whether it repeats.

Keep the probe in place and observe the test pot through ordinary changes. Do not force a valued plant through extreme conditions for the sake of a neat graph. A separate, plant-free test container keeps the early investigation focused on the instrument and medium.

Look for direction first. Does the reading generally move in a consistent way as the measured mass falls? Does adding water produce a repeatable response? Does the reading return near a previous value under apparently similar conditions, or does it drift between cycles?

Avoid choosing a threshold from the most convenient pair of readings. A useful threshold would need to survive the variation you actually observe, and it still would not automatically establish the plant's needs. That is a separate biological question.

If the probe reacts strongly to being moved a short distance, record that. The result may mean placement is central to the project. It does not necessarily mean the sensor is broken.

Give uncertainty somewhere to appear

A display can show a raw value, a recent trend, and a last-read time without claiming "water now." It can also say that the reading is missing or outside an expected range. Those are useful states, not cosmetic failures.

Test a disconnected probe using the manufacturer's safe procedure. What does the reader report? A plausible-looking constant number is more troublesome than an obvious error. The software should not present a failed read as a fresh measurement.

Repeat the test after a restart. Check whether configuration, timestamps, or baseline values were silently reset. A sensor station that forgets its scale after a power interruption can look reassuringly normal while changing the meaning of every number.

Keep an observation log separate from decisions. "Reading fell while pot mass fell" is an observation. "This plant needs water at this value" is a proposed rule. The gap between them is where the rest of the work belongs.

Leave the pump in the drawer

Automatic watering is another project, with overflow, dry-running, stuck-output, and unattended-operation questions of its own. None of those disappear because a dashboard has a green indicator.

For now, use the sensor as a witness rather than a controller. Compare it with ordinary inspection and the plant's established care guidance. If it does not add useful information, revise the placement or the measurement method before adding automation.

A modest result is entirely worthwhile: this probe, in this medium, at this position, produces a trend that can be interpreted. That is a much stronger foundation than a percentage that has never had to explain itself.