Monitoring only helps once you know what a dull week looks like

A number without a baseline is not information

The app says the array made eleven units yesterday. On its own that fact tells you almost nothing, because nothing in it establishes whether eleven was a good day or a poor one for that roof, in that month, under that particular sky. A great many owners spend their first year quietly worrying about figures that were entirely normal.

Output varies more than most people expect before they own a system. A clear midsummer day at a temperate latitude can produce several times what a clear midwinter day produces from the same panels, and heavy overcast can drop either to a fraction of the clear-sky figure. Day to day, the swing is enormous. Year to year it is surprisingly stable.

So the question worth asking is never how much the system made. It is how much it made relative to what this array does under conditions like these. Everything useful in monitoring follows from having something to compare against.

What the platform is actually measuring

The figures in a monitoring app almost always come from the inverter rather than from a meter. It senses voltage and current on its inputs and output, samples them frequently, and reports an average. Those sensors are built for control rather than for billing, so a small discrepancy against the utility meter is ordinary.

What varies between systems is granularity. A plain string inverter reports what each of its inputs is doing, which on a two-string system lets you compare one half of the roof against the other. A system with optimisers or microinverters reports per panel, which resolves faults a whole-system figure hides for years — a failed bypass diode, one module with a cracked cell, a panel that a new aerial clips at four in the afternoon.

It is equally worth knowing what the platform cannot see. Unless current transformers were fitted around the incoming supply, the system has no idea how much of its output you consumed and how much went to the grid, and without an irradiance sensor it doesn’t know how much sunlight was available to work with.

Compare the array against itself

The strongest baseline you will ever have is your own history. After a full year, every month has a predecessor recorded under the same orientation and the same shading obstructions, which removes most of the variables at a stroke. Weather still differs between one March and the next, but a real fault is usually bigger than that difference.

Better still is comparing two strings against each other on the same day. They saw identical weather and use identical hardware, so if the panel counts match they should track one another closely. A persistent ratio between them — one input consistently making four fifths of what its twin makes, in every light condition — is one of the few things a single day of data can tell you.

Professionals formalise this as a performance ratio: measured output set against what the nameplate rating and the measured irradiance say should have been possible. You can’t calculate it properly at home without an irradiance figure, but the idea still helps — stop judging the system by absolute units and start judging it by consistency.

Faults leave recognisable shapes

The most useful distinction is between a step and a slope. Degradation is a slope, and a very shallow one. Anything that appears as a step — output that was one thing last week and is plainly another this week, across all weather — is a fault worth investigating immediately.

Different faults produce different signatures. A string dropping offline halves the output of a two-string system in an instant, which is unmistakable. A failed bypass diode removes a third of one panel, which on a sixteen-panel array is a loss no whole-system view will ever surface. New shading arrives as a bite taken out of the daily curve at a consistent clock time, and it deepens as the season pushes the sun lower.

Some shapes aren’t faults. A flat plateau across the top of a bright summer day usually means the inverter is clipping, which is normally a design choice. Soiling shows up as a gradual sag that recovers after heavy rain, and a system that shuts down briefly on a very hot afternoon is often derating itself to protect its own electronics.

Knowing when to call somebody

Most of the time the explanation is the weather, and a week is far too short a sample to conclude anything else. A fortnight of grey skies in a maritime climate produces numbers that look alarming next to the same fortnight last year, with no fault to find. Patience is cheaper than a call-out.

Some things do deserve a prompt call. Output that has stopped entirely, an inverter logging repeated errors, one string sitting persistently below its twin, a protective device that trips whenever the system starts, any smell of hot plastic, or discolouration around a connector. The last of those is an electrical safety matter rather than a yield question.

Other things do not. A few per cent below the estimate in a sales document, a disappointing month, a small disagreement between the app and the meter — none of these are worth paying an engineer to look at, and an honest one will say so after taking the fee.

The failure that actually costs money is not the dramatic one. It is the quiet one that runs for eleven months while nobody looks, so set the platform to send an email when generation is zero on a day it should not be, and then stop watching the graphs daily. Watching a solar system closely is a hobby rather than maintenance, and it won’t make the roof produce more.