Project a PV module's retained power against a guaranteed retention curve: a P50 curve, a p05–p95 band, the probability of staying inside the guarantee at the end of the term, and the loss rate above which a breach occurs. Use your own rate, or a population prior drawn from published studies. It runs in your browser. The templates are illustrative and are not any manufacturer's warranty.
Pick an illustrative template, or enter the guarantee points from your own warranty document. Then take the loss rate either from one system (your assumed or measured rate) or from the population prior. The projection needs JavaScript; the rest of this page reads without it.
| Year | Guaranteed % | P50 % | p05 % | p95 % |
|---|
A projection, not a claim by any manufacturer. No brand is named or rated. The band shows the spread of the chosen source, not a measurement, and every probability is a model output. The warranty you signed always governs.
Templates are illustrative. Published medians, reference rates and the pooled spread are marked VERIFIED. The single-system spread and the log-SD are our assumptions and are marked ESTIMATED. Anything you change becomes your assumption, and the page says so.
Based on the single-figure example in the Victorian government's solar guidance, section 8, "Panel warranty from manufacturer" (fact warranty-02, CONFIRMED). The average loss it implies, about 0.6 %/yr, is our derivation (warranty-03, ESTIMATED). Breach above — %/yr, at year 25.
The stepped example in the same guidance section, "Panel warranty from manufacturer" (warranty-01, CONFIRMED; the source gives no frequency). The 80 % at year 25 matches the figure the source gives for a high-quality panel (warranty-05, CONFIRMED). Breach above — %/yr, set at year 25.
The year-by-year points you enter from your own warranty. The projection uses them as entered, and the breach threshold is the tightest year among them. Nothing is checked against a manufacturer's data.
The median, 0.5 %/yr, of 1,920 reported rates for flat-plate modules and systems, in Jordan and Kurtz (2013), Sec. 3.1 and Fig. 2(a) (fact deg-jk2013-median-all). The crystalline-Si subset, 1,751 rates, has the same median (Fig. 2(b), fact deg-jk2013-median-si).
0.94 %/yr is the all-technology median in Straub-Mück et al. (2025), Fig. 1, from 80 studies (deg-sm2025-overall). The c-Si range 0.43–1.41 %/yr is the meta-regression prediction under best-practice assumptions, by climate and mounting (deg-sm2025-predicted-csi).
Our assumption for the standard deviation of one system's rate around its own value. It applies only when the source is one system (assumed or measured). It is not the population spread: the pooled cross-study standard deviation is — %/yr (fact deg-sm2025-spread-sd), and it mixes technologies, climates and methods. The default comes from priors.json.
The median is the all-technology median of 80 studies (fact deg-sm2025-overall). The log-SD, —, is our parameter from priors.json and is ESTIMATED. The lognormal with these two values has a mean of — %/yr and a standard deviation of — %/yr. The source reports a mean of 1.10 and a pooled SD of 0.99 %/yr (deg-sm2025-overall, deg-sm2025-spread-sd). No lognormal with median 0.94 reproduces both, so the page keeps the source's median and shows the implied moments.
The SD is the pooled cross-study value (fact deg-sm2025-spread-sd). The centre uses the same median as the lognormal, so the two priors differ in shape only; that choice is ESTIMATED. A normal distribution allows negative rates. A module that gains power is shown capped at 100 % retained, and it cannot breach.
Example values for the first view. They are not data about any module.
Each run draws one constant loss rate for the module's life, and the loss is linear in time. Retained power is capped at 100 %, so early gains earn no credit. Between stated guarantee years the curve is interpolated linearly; the source does not say how. A breach is counted when the run's rate is above the threshold in any year of the term. Runs use a seeded generator, so the same inputs give the same numbers.
The loss you measured, divided by the years over which you measured it. Use it only when the loss is measured against the module's rated power, under the same test conditions.
Six findings behind the defaults and the limits of this projection. Each cites its source, locator and fact ID.
Median 0.5 %/yr across 1,920 reported rates; mean 0.8 %/yr, so the distribution is skewed toward higher rates. Thin-film rates have a median of 1.0 %/yr. Source: Sec. 3.1, Fig. 2(a)–(c), author manuscript pp. 5–6 (facts deg-jk2013-median-all, deg-jk2013-share-below-1pct, deg-jk2013-median-tf).
Median 0.94 %/yr across 80 primary studies and about 70,000 modules. Thin-film medians (CIGS and CdTe) are 1.5–2 %/yr. Source: Renewable and Sustainable Energy Reviews 216, 115697, Fig. 1–2 and the text (facts deg-sm2025-overall, deg-sm2025-sample, deg-sm2025-tf-median-range).
A five-year study of 834 modules gives a median of −0.58 %/yr (negative = loss, as printed). The Sandia Lab News report of the same study finds the degradation rates highly nonlinear over time. Source: Theristis et al., Prog. Photovolt. Res. Appl. 31(2), abstract (OSTI record; fact deg-theristis2023-median); Sandia National Laboratories, Lab News, 3 Nov 2022, results paragraph (facts warranty-07, warranty-08).
After 15 or more years, nearly all affected module types still meet the manufacturer's power warranty. These are a selected set of types, not a random sample. About 2 % of the fleet, from modules of 21 manufacturers after 8 years in the field, is predicted not to meet the warranty at 11–12 years. Source: Ch. 3, Fig. 3.3–3.4, PDF p. 11 (facts warranty-24, warranty-25).
About 60 % of modules would meet 80 % of their initial power after 35 years, or about 70 % allowing for ±3 % measurement uncertainty. Manufacturers were then considering that 80 % threshold. Source: Virtuani et al., Prog. Photovolt., DOI 10.1002/pip.3104, abstract (fact warranty-11).
The EU JRC study, in draft, finds that manufacturers' degradation claims, and the power guarantees linked to them, lack a standardised basis, and that the claims usually lack an explanation of method (fact warranty-19). Working life is typically defined as ending between 80 % and 70 % of the initial power rating, or at a safety problem (IEA PVPS, Ch. 3, PDF p. 11; fact warranty-06).
The model uses one constant loss rate per module. Field studies disagree on the shape: one five-year study reports highly nonlinear degradation (warranty-08), while a majority of modules in the 2016 compendium show a fairly linear decline (warranty-15). The year-25 value is therefore a model output, not a forecast of the module.
The IEA PVPS definition of a module failure is a power loss that normal operation does not reverse, or a safety issue. A failure is relevant for the warranty when it occurs under conditions the module normally experiences (IEA PVPS, 2017, PDF p. 10; fact warranty-26, CONFIRMED). The model has no failure term, no returns and no claim process.
A high percentage of references take only one or two measurements, typically against the nameplate STC rating (Jordan and Kurtz, 2013, data-quality note; fact deg-jk2013-data-quality). The single-system spread is a handle on this, not a calibrated uncertainty. The population prior describes published modules in general, not yours. A measured rate from your own module is the better basis when you have one.
Read the output as a range. The P50 is the middle run, the band is the 5th to 95th percentile of the runs, and the probability is the share of runs at or above the guarantee at the end of the term. The breach probability is the share of runs whose rate is above the threshold in any year. None of these is a promise about any module.
Send the warranty and the module datasheet. We can run this projection on your clauses and on a rate measured from your own data, and state each assumption next to the result.
Ask about a warranty projectionMeasure the rate first with the PV Health Score · How the evidence is graded