How we research
Who makes this site?
RuntimeWise editors. We build comparison data. We do not run a test lab, and no page claims hands-on testing.
Where do the numbers come from?
- Specs: each manufacturer's official US spec page, store listing or manual. Every model page links its source and the date we checked it.
- Specs a manufacturer doesn't publish are shown as "Not published", never estimated. Where official pages disagree, the spec carries a short note.
- Appliance wattages: generator-maker wattage guides (Honda, Generac, Champion), DOE/Energy Saver guidance, ENERGY STAR and appliance manuals, cited on each appliance page.
- Which questions we answer: questions buyers post on Reddit, Stack Exchange and in YouTube review titles, which we grouped by appliance and use.
- Context figures: outage data, measured appliance draw, battery lab tests, and air-travel, campground and grid rules come from 79 primary documents (EIA, CPSC, FAA, NREL, Sandia, the National Park Service, ENERGY STAR and manufacturer manuals). We checked each figure against the document itself, and every page that uses one names the publisher, year and file type.
How are runtimes calculated?
Runtime = rated capacity × 85% inverter efficiency ÷ (appliance watts × duty cycle). For example, a 1,024 Wh station running a 200 W fridge that is on a third of the time: 1,024 × 0.85 ÷ 66 ≈ 13 hours.
- A load counts as runnable only if its running watts fit the continuous rating and its start-up surge fits the surge rating, or the continuous rating when no surge is published.
- "Best for" lists are filters, with the requirement shown at the top of each page.
Related: how to size a power station and surge vs running watts.
Why build it this way?
Most buyer questions are "will it run X, and for how long". That depends on numbers, so we show the numbers and how we got them.
Who, how and why: what should you know about this site?
RuntimeWise editors build this site with our own scripts, from checked manufacturer data, to help you decide whether a power station will do a specific job. Google asks publishers to answer who made a page, how and why, and here are our answers in one place.
- Who. RuntimeWise editors, who run RuntimeWise. We don't list personal credentials, and no page asks you to rely on anyone's expertise. Each figure carries its source.
- How. Specs go into a database after they are checked against each manufacturer's official US page. Our scripts compute runtimes, fit checks and lists from that database when the site is built. Prose is drafted with AI assistance under written rules.
- Why. To help a buyer finish a job, such as running a fridge through an outage, and pick the right size. Pages exist because buyers ask the question, not to cover keywords.
How are the pages produced?
Every page is generated by scripts from one verified spec database, so a number appears the same way everywhere it's used. The database holds 14 spec fields for each of 56 models from 6 brands.
- Collect. Specs are read from the manufacturer's official US spec page, store listing or manual. The source link and check date are stored with the model.
- Check. Each value is compared with the official page. Values the manufacturer doesn't publish stay blank and display as "Not published".
- Compute. At build time the scripts calculate runtimes with the formula above, run the running-watt and surge checks, and apply each list's filter.
- Write. Explanatory sections are drafted, then checked by a script that confirms every number placeholder resolves to the data and every internal link exists.
- Build. The whole site is regenerated, so a corrected spec updates every page that uses it in one pass.
How is AI used on this site?
AI assistance is used to draft prose and to support research, and it is never the source of a number. We state this because you should know how the words you're reading were produced.
- Drafting. Explanatory sections were drafted with AI assistance under written rules. The rules say every number must come from the spec data, a basic electrical identity or a cited source document. They also ban testing claims and brand claims the data doesn't support.
- Numbers. Specs, runtimes and counts are inserted from the database at build time. The drafting step writes a placeholder, and the build fills in the value.
- Outside statistics. The 79 context figures were checked against the source documents by an AI-assisted research process, and each one carries its caveats.
- Review. We don't claim a human expert reviewed every page. Automated checks catch unresolved numbers and broken links. Reader reports catch the rest.
How are specs kept current?
Product data is re-checked before we change a page and whenever a brand updates its own pages. Model specs were last checked on 2026-10-12, and every model page shows its own check date.
Brands revise spec pages without notice. A new firmware version can raise an output rating, a product can be renamed, or a store can drop a model. When we find a change, we update the database from the official page and rebuild. List prices are shown with the date we saw them, because they move far more often than specs do.
How are lists and rankings ordered?
Lists are ordered by spec metrics only, after each model passes or fails the page's stated requirement. No brand pays for placement, and commissions don't change an order.
A "best for" page first filters out every model that misses the requirement printed at its top. The models that pass are then sorted by a spec measure suited to that job. The code that sorts them has no input for commission or affiliate status. Comparison pages put two models side by side using the same fields, so neither gets a field the other lacks. See the best-for lists and the comparisons to see the stated requirements in use.
How do we correct mistakes?
We correct a mistake by re-checking the official source and updating the data, which then updates every page at the next build. You can report one through the contact page.
A useful report names the model, the page and the number, with a link to the manufacturer's page that shows the right value. If the official page agrees with us, the spec stays as it is. Where official pages disagree with each other, the spec carries a short note saying so.
What are the limits of this method?
Our numbers are calculated planning figures, not measurements. They are as good as the published specs and the assumptions behind them.
- Efficiency. We assume 85% inverter efficiency for every model. Real losses vary with the load size and the unit.
- Appliance draw. We use cited typical wattages and duty cycles. Your appliance can draw more or less, so enter your label in the runtime calculator.
- Conditions. Cold, heat and battery age reduce usable capacity. We don't model them.
- Prices. We show the manufacturer's list price with the date we saw it. Sale prices and retailer stock change too often to track, so check them before you buy.
- Coverage. We track 6 brands. A model or brand outside that set isn't a judgment on its quality.
Read surge vs running watts to see why the start-up check matters as much as the runtime.
What do official documents say about our assumptions?
NREL's 2025 battery cost study assumes 85% round-trip efficiency for grid batteries. That is a different measure from our 85% one-way inverter figure, but it shows how much energy real systems lose in conversion.
Cost Projections for Utility-Scale Battery Storage: 2025 Update (NREL/TP-6A40-93281)
NREL's PVWatts model assumes a fixed solar array loses 14% of its rated output before the inverter: dirt (2%), shading (3%), mismatch (2%), wiring (2%) and other small losses. Our 75% assumption for portable panels allows for those losses plus heat and less-than-ideal angles.