Why every installer's year-1 production number is 8% high.

The first-year kWh estimate anchors everything downstream: the bill-savings calculation, the payback math, and — crucially — the financed amount on the loan. So when we tell homeowners the median estimate in our 312-proposal audit set ran 8.4% high, the first question is always the same: are installers cheating?

Mostly, no. They're using defaults.

What the number models.

The kWh estimate is the output of a physics simulation. PVWatts is the most common — a NREL-built model that takes a system spec (DC kW, tilt, azimuth, module type) and a weather file (typical meteorological year for the location) and returns expected hourly production. The simulation is good. The defaults are bad.

Three places where the defaults run high:

Soiling. The PVWatts default is 2%. Real soiling losses in dusty climates — Arizona, Texas, the Central Valley — are 4–7%. Coastal Florida, with regular rainfall, is closer to the default. Almost no proposal we've audited used a region-specific soiling figure.

Snow loss. Defaults to zero in many tools. New Jersey and Massachusetts roofs lose 3–6% annually to snow cover. The proposal-tool default is being used in Boston.

TMY weather files. Built from 30-year historical climatology. The trend in cloud cover and aerosol optical depth means recent years are consistently dimmer than the typical year for some regions, and brighter for others. Recent California summers, for example, are 4–8% sunnier than the TMY would predict.

"The estimate is a default that nobody updated, multiplied by an installer with no incentive to update it."

What we found.

For each of the 312 proposals, we had monitored year-one production from the customer's inverter. We computed proposal-vs-actual for first-year kWh and recorded the system's location, tilt, azimuth, soiling region, and snow zone.

Median overstatement: 8.4%. P10: −1.6% (a small set of pessimistic proposals). P90: +14.7%. The distribution is skewed positive — a long right tail, a short left tail. Random measurement noise would land symmetric around zero; this skew points at something structural. That asymmetry is the tell.

The skew comes mostly from the structural defaults. Snow-zone systems missed by an average of 4.8% on the snow line alone. Dusty-climate systems missed by 3.1% on soiling. Together those two factors explain about 60% of the variance.

What an honest estimate looks like.

We've started defaulting to PVGIS for a sanity check on PVWatts in our own audits. Different weather methodology, different default loss stack, similar physics core. When PVGIS and PVWatts disagree by more than 4% on first-year kWh, something is wrong with one of them — usually the soiling, snow, or shading assumption — and we kick the proposal back for a manual review of the loss stack.

The fix is mundane: regional defaults plus a second model as a sanity check. That's table stakes for any planning tool that wants to call its number a forecast rather than a quote. Most installer software is still running national defaults from 2014. The number on your proposal is exactly as accurate as the lowest-effort version of that math allows.

If you have a quote, ask which soiling factor, which snow factor, and which TMY vintage. A salesperson who can't answer is telling you the number came straight from the software defaults — which is exactly the problem.