Your solar is producing less than promised: what to check

The estimate has error bars, and they are published

Almost every residential solar estimate traces back to a free modelling tool from a national laboratory, either directly or through software built on top of it. It is a genuinely good tool. It also documents its own limits with a candour that never survives into a sales proposal.

From the manual, verbatim: "the results should be interpreted as being a representative estimate for a similar actual system operating in a year with typical weather. The errors may be as high as ±10% for annual energy totals and ±30% for monthly totals for weather data representing long-term historical typical conditions. Actual performance in a specific year may deviate from the long-term average up to ±20% for annual and ±40% for monthly values."

Sit with the monthly figures. A single month coming in 30% below the projection is within the model's stated error for typical weather, and a specific year can deviate by up to 40% in a month. Judging a system on one month is judging it against a number the model itself does not claim to that precision.

The manual is also explicit that this is the low-fidelity option by design: it "hides much of the complexity of accurately modeling PV systems from the user by making several assumptions", and while useful for a quick estimate, "several more sophisticated tools are available for making more accurate predictions". The laboratory names its own more detailed model as the alternative.

None of that means your system is fine. It means the comparison you are making needs a tolerance attached before it can tell you anything, and most people are comparing a real measurement against a number they believe is exact.

The default losses your estimate probably used

Underneath the headline number is a stack of assumed losses. The tool ships with defaults, and unless whoever produced your estimate changed them, those defaults are what your projection assumed about your roof.

Three of them deserve attention, and two are zeros.

Shading is defaulted to 3%, and it is not the shading you are thinking of. The documentation is explicit that this figure represents an unshaded horizon, and that it does not account for shading from the homeowner's own trees or nearby buildings. If you have a tree, the default assumed you did not.

Snow is defaulted to 0%. Not a small number, zero. Set that against a national laboratory's published statement that "snow-related power losses can exceed 30% of annual production" at northern latitudes, and it becomes the largest silent assumption in any northern estimate. That statement is a bounding claim rather than a typical value, and the gap between a bound like that and a default of zero is where a northern homeowner's disappointment lives.

Ageing is defaulted to 0%. The model assumes no degradation over the period. Panels degrade, and separate research puts the median field rate around half a percent a year with a mean higher than that. Over a long projection that assumption compounds.

What the shape of a shortfall tells you

Most diagnosis effort goes into the size of the gap. The shape of it is more diagnostic and your monitoring already shows it.

Uniformly low across every hour and every month points at something systemic: the system is smaller than you think, the model was optimistic in a way that does not vary, or something is derating everything equally. It rarely points at shading, which is time-specific by nature.

Fine at midday, poor in the morning or afternoon is the signature of shading, and which end of the day it falls on tells you which direction the obstruction is. This is the pattern trees produce, and it worsens over years as they grow, which is one reason a system that met expectations in year one may not in year eight.

Fine most days, occasional near-zero days is usually weather or an outage rather than a defect. Cross-check the near-zero days against local weather and against any grid interruption.

One string or one area consistently below its neighbours is the most actionable pattern of all, because it localises the problem to hardware in a specific place. If your monitoring reports at panel level, this is where that capability earns its cost.

A flat top on sunny days is not a fault at all. It is clipping: the array producing more DC than the inverter can convert, which is a deliberate design choice in many systems. Worth confirming rather than reporting.

What a realistic expectation looks like

There is one more piece of context worth having, from a large fleet study rather than from a model.

Work examining a portfolio of well over a hundred plants totalling around nine gigawatts found actual generation running at roughly 81 to 82% of ideal expected energy, with inverter downtime around 2 to 3% against roughly 1% typically assumed in financing. That is utility-scale rather than residential, and it is not directly transferable to a house.

What it does establish is directional and useful: across a large professionally operated fleet, real output sits meaningfully below ideal expectation, and one of the larger contributors is availability rather than the panels underperforming. Residential systems are not monitored or maintained to that standard.

One more asymmetry is worth naming, because it explains why this argument is so often unresolvable. The installer holds the model and its assumptions; you hold the meter. Neither of you can settle the question alone, and the assumptions are the half that never gets written down. Asking what the estimate assumed is not an accusation, it is a request for the other half of the evidence.

So the realistic frame is that a well-functioning residential system produces somewhat less than an idealised model says, for reasons that are mostly not defects. The question worth asking is not "is it hitting the number" but "is the gap explained by things I can name". If every item on the list above accounts for part of it, the system is probably fine. If a large part of the gap is unexplained after you have worked through them, that is when to escalate.