Solar Yield

A Probability, Not a Promise

Checked 2026-08-09
Notes

Most of the disputes, disappointments and misunderstandings in this subject reduce to one thing.

For teams coordinating research, review and follow-up work around a project, remote workforce management software provides a separate record of time and handoffs.

The supplier produces a distribution. The buyer reads a promise. Neither party is wrong about their own document, and the mismatch generates everything else.

What the supplier produced

A model of a distribution, summarised by its median and by a conservative exceedance level. P50 and P90.

For wider independent context on solar markets, research and industry developments, consult Renewable Energy World.

The distribution is the honest output. Weather varies year to year and no assessment can remove that; the best it can do is characterise the variation and say how confident it is.

And it comes with a stated uncertainty, which is what the gap between the two figures encodes.

What the buyer read

A number the plant will produce.

Not through carelessness. Every other capital purchase behaves that way — a machine has a rated throughput, a vehicle has a load capacity, a building has a floor area. Nothing in ordinary commercial experience prepares somebody for a purchase whose output is a probability distribution.

So the buyer applies the only reading they have, and it is the wrong one.

What follows from the mismatch

Year one below P50 reads as a failure. It has roughly a 50% probability by construction.

Guarantees look like betrayals. A guarantee near 95% of P90 is much lower than the savings figure, and to somebody expecting a promise it looks like fine print concealing a retreat.

Comparison between suppliers goes wrong. A conservative assessor with honest uncertainty loses to an optimistic one, because the buyer compares point figures and the point figures are not comparable.

And genuine underperformance gets excused. Once "it varies" is the frame, a real PR shortfall can hide inside it for years.

That last consequence is the expensive one, and it is caused by the same mismatch as the first.

What would fix it

On the supplier side: leading with the distribution. Presenting a range with its probability rather than a headline figure with a footnote. Costs a sentence.

Stating which P-value every derived number uses. Savings, payback, debt cover — each should carry its basis, and most proposals state it once if at all.

On the buyer side: asking for both figures and the ratio between them. Three numbers instead of one, and the third is free.

And treating one year as one observation. Which is obvious stated plainly and almost never done in practice.

The frame that works

A solar plant is not a machine with a rating. It is a machine plus a weather series.

The machine part can be specified, tested and guaranteed. The weather part can only be characterised.

A buyer who holds those separately asks the right questions of each: is the equipment performing as designed, and was the weather what the model assumed. Performance ratio answers the first and an irradiance record answers the second.

A buyer who holds them together has one number, no diagnostic, and an argument every time the year is poor.

The short version