A sensor you haven't validated is a sensor you're trusting on faith. Faith is not a measurement.
This is the unglamorous half of a monitoring system, and it decides whether the other half is worth anything. The day to discover that a pH probe reads 0.4 high is before a crop depends on it — not three weeks in, when the plants look wrong and you can't work out why. Commissioning is the discipline of proving every sensor tells the truth before you rely on it. Maintenance is the discipline of keeping it true after. Skip either, and the most sophisticated cascade detection in the world is just acting confidently on bad numbers.
Commissioning — prove it before you trust it
Before any crop enters the facility, three things get done and documented.
Validate every sensor against a known reference. Each reading is checked against something whose value you already trust: temperature against a reference thermometer; pH against fresh 4.0 and 7.0 buffers; EC against calibration standards; dissolved oxygen against air-saturated water at a known temperature; light against a quantum sensor calibrated within the last year; CO₂ against a gas of known concentration. Anything that disagrees with its reference by more than the sensor's rated accuracy is rejected and replaced — not noted and tolerated. Write down every result; that document is your baseline.
Map the airflow. Divide the canopy into 2 m × 2 m cells and measure air velocity at canopy height in every cell. Every cell must read at least ~0.3 m/s. Any cell below that is a dead zone — reposition fans or add circulation until it isn't. This is the one time airflow gets measured this thoroughly (quarterly spot-checks after), so make the map and keep it. Those dead zones are exactly where the boundary-layer gap on the Plant's Eye View opens up — a stagnant cell is a pocket where your room sensors quietly stop describing the plant.
Tune the control loops. For each thing the system actually controls — temperature, VPD, CO₂, pH, EC — check three things when you change its target: how fast it responds, how far it overshoots, and how accurately it settles. Re-tune any loop that overshoots more than ~10% or takes more than ~5 minutes to respond on a critical variable. A loop that hunts and overshoots isn't "working" — it's stressing the crop on its way to the setpoint, twice an hour, forever.
Maintenance — keep it true
Every sensor drifts, fouls, or wears out on its own schedule. Keeping the system honest means keeping that schedule:
- pH probes: two-point calibration weekly; replace every 6–12 months.
- EC sensors: calibrate monthly; clean salt buildup weekly.
- Dissolved oxygen (optical): verify monthly; clean biofilm monthly.
- Humidity / temperature: cross-check quarterly; wipe dust and residue monthly.
- CO₂ (NDIR): calibrate against span gas semi-annually; clean the optical path quarterly.
- Light (quantum) sensors: factory recalibration yearly; wipe the diffuser weekly.
- Air stones / diffusers: acid-soak monthly; replace every few months.
The trick is to make the system maintain itself as far as it can. The same platform that watches the crop can watch its own upkeep: track days-since-calibration for every probe and fire a maintenance reminder through the normal alarm pipeline when one comes due; log every calibration's slope and offset so a probe's slow death shows up as a trend long before it fails outright; track equipment runtime to catch a chiller or blower losing efficiency before it loses the room.
This page is what makes every keep it honest line in the instrumentation matrix actually happen — on a schedule, by design, instead of whenever someone remembers. An uncalibrated probe lies with authority; commissioning is how you start from truth, and maintenance is how you stay there.
A monitoring system is only as trustworthy as its least-honest sensor. Commission it so you begin with truth. Maintain it so you keep it.