17 reviewsAll reviews →
Method

How we research home backup power

As an Amazon Associate I earn from qualifying purchases.

In short

Porchlight tests nothing itself. We read what makers promise against what labs and owners measured, file every fact against the exact US model, and say 'not measured' where nobody measured.

Every power station box makes a promise. Hardly any of them tells you how it was measured. That gap is where Porchlight works. We haven't run a single one of these machines, and we won't pretend we have. What we do is set each maker's promise against what independent labs measured and owners saw, file every fact against the exact model sold in the US, and tell you plainly what it adds up to. Every number on a page has a name and a date behind it, and where nobody has measured something, we say so.

Who writes this, and why an engine builder?

Who is David King?

Porchlight is edited by David King, and the buck for every page stops with him. He spent his working life in aerospace manufacturing, most recently as a Production Leader at Rolls-Royce running the build of jet engine sections for Airbus and Boeing aircraft, and before that as a Production Manager at WEC Group (his profile on LinkedIn (opens in a new tab)). On an engine line, a part doesn't ship because the supplier says it's good. It ships with a measurement, a tolerance and a trail back to whoever checked it.

He is not a power station expert and doesn't claim to be. He's an editor and researcher who still asks the question he put to suppliers for years: measured at what load, and what happens when it fails? I read a runtime banner the way I'd read a spec sheet with no test report stapled to it: as a wish with a part number. There's more about him on our About page.

What we never do

Has anyone here tested these products?

No. Nobody at Porchlight has used, tested or owned the products we cover. We don't run a lab, and we turn down free units from makers. Nor is a spot on any page for sale. When a page says a station kept a fridge cold through the night, that's a lab's or an owner's finding, and the page credits it to them. Our opinions are ours, and we'll tell you straight what we'd buy and what we'd skip. The experience always belongs to someone else, and it's named as theirs.

How the research is done

Is this written by AI?

The research is AI-assisted, and we'd rather say so than have you guess. AI models read makers' product pages and US manuals, independent labs' published tests, reviewers' write-ups and videos, and owners' reports on Reddit and YouTube. They behave like a research assistant who won't take a quote on trust. Each lead is chased to where it started, and we keep a complete copy of every web page the research leans on. Reddit threads are the exception: those read before November 2, 2026 were saved only as a summary, so they're tallied on their own and never counted among the sources kept whole.

The model that drafts a page sees that page's evidence and nothing else. It stays offline from first line to last, which keeps a claim picked up from somewhere else out of the page.

As of November 13, 2026, our files hold 27 exact models and 974 facts about them: 352 measurements, 61 things someone saw happen, and 561 makers' claims, drawn from 589 sources kept in full. Notice which pile is biggest. Makers have said far more than anyone has checked, which is exactly why this work needs doing.

How do you know a number belongs to the model I'm buying?

Every fact is filed against the exact model it covers, along with where it came from and when. It's also tagged for its kind, so a maker's promise never sits in the file looking like a lab's reading. In backup power this matters more than almost anywhere. A DELTA 3 Classic is not a DELTA 3 Plus. A Generac 8945 is not the 8947 built for California's rules. A bundle with an extra battery is not the bare unit. A newer generation, a regional version and the same hardware under a new name each get their own file, because a result for one is not a result for another.

Where a page leans on a sibling model's figure, it says so in the same sentence. Where the only figure was measured on a different model altogether, we leave it out. The only reading we found for one Anker station's neutral-ground bond came from a different Anker unit. So we dropped it, and the record for that station now says the maker hasn't stated it.

What happens when two labs disagree?

We show both. Labs test differently, and most gaps between their numbers come from how they tested, not from who's right. A lab that drains a battery under a heavy load gets more out of it than your house will. A cycling fridge and a router make a light load, and at light loads the inverter wastes a bigger share of the energy. Another lab counts the energy lost charging the battery back up, so its figure always looks lower. Average the two and you get a number nobody measured.

So each lab's figure stays beside the test it came from, and our conclusion goes on the page in plain words. Every buying page closes with "The evidence in full", where every lab's figure sits with the method behind it, for anyone who wants to check our working.

Do you trust sponsored reviews?

We read them, and we flag them. The Solar Lab, which buys many of its own units, says a large share of power station content online is sponsored one way or another, and that some creators are asked to highlight features and avoid criticism. In our own files, a YouTube furnace runtime test ran on stations the maker sent, and every unit on Wired's power station list came from the maker or its PR firm. None of that makes their numbers wrong. It means we note who supplied the unit, and a measurement counts for more with us than praise.

What if nobody has measured it?

Then the page says "not measured". Not a guess, not a blank. You'll see it more than you might expect. Idle draw, the power a station burns just keeping itself switched on, decides how long it lasts through a multi-day outage, and almost no big publisher reports it. The home backup power calculator on Porchlight keeps to the same rule. Each figure in it is credited and dated. Where no measured capacity exists, the arithmetic uses the maker's rated one and labels the result to match.

How a page gets written and checked

Before a page is published, scripts comb through it for the mistakes we refuse to make. They reject any price figure, and any line that puts a lab's or an owner's experience in our mouth. They require every number with a unit to match a fact we hold for the model named beside it, and every listed source to be one the page truly relied on. Each product write-up finishes on a plain call (buy, skip or wait) and who it suits. A skip always names the product that suits that reader better, because "don't buy this" with nothing to buy instead is half an answer.

Prices and keeping pages current

Why don't you show prices?

Because they change daily, and a price frozen on a page is wrong by the weekend. Pages give a price band instead, and Amazon shows the current price when you click through. Product links are checked whenever the site is updated, a model that is no longer sold points to its replacement, and each page carries the date of its most recent update.

How Porchlight earns money

As an Amazon Associate I earn from qualifying purchases.

When you buy through a link on one of our pages, Amazon pays us a commission. It costs you nothing extra.

Does the commission change your picks?

No. The commission never decides a call. A product that doesn't suit a reader gets a skip whatever it pays, and nothing here is for sale: not a place on a page, not a kinder verdict. A maker that wants a better write-up from us has one route, and it runs through a better product.

↑ ↓ to move↵ to open