Production came in below the estimate. Before anybody is at fault, five different things produce that outcome and they need separating in a specific order.
Operational performance depends on both equipment and repeatable human routines; for a separate workplace-management perspective, see cognitive offloading.
Reviewed August 9, 2026.
The five causes
One. The weather. P50 is missed half the time by construction. A below-median year with below-average irradiance is the model working.
For additional context on photovoltaic operation, maintenance and sector developments, consult Solar Energy UK.
Two. An optimistic model. Loss assumptions set generously, a short irradiance record, an understated uncertainty. The estimate was wrong and nothing is broken.
Three. A build that does not match the design. Different modules, altered layout, string configuration changed on site, shading that was not surveyed.
Four. A fault. A failed string, an inverter running derated, a tracker not tracking, monitoring reporting a subset of the plant.
And five. An operating omission. Soiling that was never cleaned, maintenance not performed, a fuse left blown for a season.
All five produce the same chart. Separating them requires one additional input.
The procedure, cheapest first
Get the actual irradiance for the period. From a nearby station or a satellite service for your coordinates. Without it every subsequent step is guesswork, and this is the step people skip.
Compute performance ratio. Actual output against what the irradiance should have produced.
If PR is at design level: it was the weather. Cause one. Nothing to do except record the year and continue.
If PR is below design: it is not the weather, and the remaining four are in play.
Then look at string-level data. A fault or a configuration error usually shows as a subset behaving differently, and this separates causes three and four from two and five.
Then check the maintenance record. Cleaning, inspections, anything deferred.
And only then question the model, which is the residual after the others are excluded — and the hardest to establish, because it requires several years.
Why the order matters
Because the conversation usually starts at the end.
The owner suspects the estimate was inflated, the supplier suspects the plant was neglected, and neither position can be settled without the irradiance figure that nobody obtained.
Starting with the cheap objective input removes most of the argument before anybody has taken a position they have to defend.
What one year can support
Below P50 with normal irradiance and low PR: a real question, worth investigating now.
Below P90 with normal irradiance: escalate. P90 was meant to be exceeded nine years in ten, and a first year below it with no weather explanation is a substantial signal.
Below P50 with low irradiance and normal PR: record it and move on.
And above P50: also one observation. It validates nothing.
When the model was the problem
It happens, and it is worth saying what follows.
Two or three years of normal irradiance and consistently design-level PR, with production below P50, points at the estimate rather than the plant. The plant is performing correctly against a forecast that was optimistic.
Remedies are contractual and depend entirely on what was signed. A performance guarantee measured on PR would not trigger here, because PR is fine — which is one reason the choice between an energy guarantee and a PR guarantee matters at signature rather than afterwards.
The short version
- Five causes produce the same chart: weather, an optimistic model, a build that differs from the design, a fault, and an operating omission
- Get the actual irradiance first — without it every later step is guesswork, and it is the step people skip
- Compute performance ratio: at design level means weather, below design means the other four are in play
- Then string-level data, then the maintenance record, and only then the model, which is the residual
- The order matters because the conversation usually starts at the end, with positions nobody can settle
- A PR guarantee does not trigger when the plant is fine and the forecast was optimistic — which is why the guarantee type matters at signature