How long modules really last
Two different promises share the word warranty. The product warranty covers the module itself - the frame, the glass, the junction box - and is the one that matters if something breaks. The performance warranty covers output, and guarantees only that the module will still make some stated fraction of its rating after a stated number of years.
They rarely match. The median product warranty is 15 years and the median performance warranty 30, so for the second half of a module's guaranteed output life the manufacturer is no longer promising the hardware. Both are ordinary contract terms rather than physics: a 40-year product warranty from a firm that folds is worth nothing.
What the distribution shows
Warranty terms are round numbers because they are commercial decisions: 12, 15, 25 and 30 years carry almost the whole market. 2,835 modules carry 25 years or more of product cover, which is 20.6% of those declaring a term.
Declared annual degradation runs from 0.18% to 1.00%. The difference looks small and compounds: over 24 years after the first, 0.18% of annual loss leaves 95.8% of the rating, and 1.00% leaves 78.6%.
Nearly every datasheet quotes a larger loss for year one than for the years after it - light-induced degradation happens once, early. The median first-year figure is 1.00% against 0.40% a year afterwards, and applying the annual rate to year one instead is how a 25-year projection ends up roughly a percentage point optimistic. A median module modelled properly holds 89.9% of its rating at 25 years.
The longest-backed modules
Two lists: what the contract says, and what the module is expected to do.
By manufacturer and architecture
n-type cells avoid the boron-oxygen defect behind light-induced degradation in p-type silicon, and their declared rates are lower for it. The median across the whole set is 0.40%. Lower is better.
The typical guarantee is a floor at the end of the term: the median declared end-of-warranty output across panels stating one is 87.4% of the original rating. It is a floor, not a projection - a claim needs the module measured below it under test conditions, at the owner's cost in most contracts.
It is also the manufacturer's own arithmetic. The degradation curve, the term and the end figure are chosen together, which is why comparing curves across brands says more about legal caution than about silicon.
How these numbers are built
Every figure comes from a manufacturer datasheet ingested into the panel database. Nothing is modelled, estimated, or supplied by a brand outside its own published specification.
Aggregates recompute as new panels arrive. A panel that does not declare a value is left out of that average rather than filled in, so every figure here is backed by modules that actually publish it.
No paid placements, and no figure here reads which shops pay us a commission. Each power rating counts as the module it is. Bifacial datasheets that quote one module at several assumed rear-side gains are counted once, on their front-side rating.