Power Station Data

Abstract ordered ascending steps illustration in amber and pale sand on white — decorative header for How We Compile These Figures

How we compile these figures

Every number on this site is derived from published manufacturer specifications by a repeatable process. This page describes that process, including the parts of it that fail.

Where the data comes from

We collect power station listings from major retail catalogues and extract each specification the manufacturer publishes — capacity, continuous output, surge rating, cell chemistry, solar input, weight and dimensions. Nothing is measured by us, and nothing is estimated.

Normalising units

Manufacturers state the same quantity in different units, sometimes within one listing: capacity appears as “1024Wh” in the title and “1.02 kWh” in the specification table. Everything is converted to a single internal unit before comparison.

This step is where the most damaging errors live, because a missing conversion does not look like an error — it looks like a number. In this corpus the clearest case was amp-hours: a specification table saying “60000 milliamp_hours” was landing in a watt-hour column as 60,000, because an unrecognised unit was being assumed to be the right one. Amp-hours are charge, not energy; without the pack voltage there is no conversion. The extractor now returns nothing rather than a plausible-looking number.

Cleaning the corpus

Search results contain a great deal that is not a power station. From the raw set we removed 7 products — uninterruptible supplies, jump starters, bare solar panels, carry cases.

The uninterruptible supply is the hard case and worth explaining, because it is the one that nearly got through. Many genuine power stations advertise a UPS mode, so excluding anything mentioning “UPS” removes real products. The rule that works is that the first comma-delimited segment names the product: a CyberPower or APC unit says “UPS Battery Backup” in its own name, while a station says “Power Station” and mentions UPS later. Four separate corpora on this network have been damaged by an exclusion pattern matching a word that appears legitimately elsewhere in a title.

Where physics rules a figure out

A portable station holds between 100 and 25,000 watt-hours, outputs between 50 and 8,000 watts, and weighs between 1 and 120 kilograms. Cells outside those ranges are measuring something else — a phone bank, a rack battery, a typo. In those cases the cell is discarded and the rest of the listing kept, because discarding the whole product would lose a real station over one bad figure.

The runtime figure, and exactly what it is

Capacity divided by continuous output, expressed in minutes. It is arithmetic on two published numbers and nothing more.

It is a ceiling, not a promise. Real inverters lose 10 to 15% turning DC into AC, so the usable figure is lower — and not one listing in this corpus states its inverter efficiency, so we do not estimate it. We publish the figure because the alternative is that the two numbers on the box never meet, and the buyer discovers what they imply during an outage.

How baselines are computed

A baseline is the median of a specification within one group. Groups here are output class, capacity band and cell chemistry — each taken on its own.

That is a departure from how a larger corpus would do it, and the reason is arithmetic. 4 output classes by five capacity bands is twenty combinations, and this corpus fills almost none with enough listings for a median to describe anything. Computing them anyway would produce numbers with the appearance of precision and none of the substance.

Capacity bands are: under 500 Wh, 500 – 1,200 Wh, 1,200 – 2,500 Wh, 2,500 – 5,000 Wh, 5,000 Wh and up.

A group needs at least 8 listings before a baseline is published. 11 of the 11 groups clear that bar.

What we will not publish

  • Star ratings and review counts. Not in any form, including paraphrase.
  • Exact prices. Prices change daily; a scraped price is misleading by the time it is read.
  • Estimated specifications. Where a manufacturer has not published a figure, the page says not stated.
  • Test results. We do not test, so we have none.
  • Efficiency-adjusted runtimes. We could assume 90% and it would be a guess wearing a decimal point.

Known limits

AC recharge time, inverter waveform and operating temperature range are too sparsely published to support comparison, so none is offered. The corpus reflects one retail catalogue at one point in time. Figures were last computed on 2026-08-27.