Methodology and data sources

Every generator page on BackupAtlas is generated from a structured dataset with a documented origin. This page explains where each number comes from, how it is checked, the exact formulas used in calculations, and — just as important — what the data cannot tell you.

Why this dataset is hand-curated, not scraped from a government source

Unlike ENERGY STAR water heaters or EV chargers, there is no public certification database for home backup generators — no single government agency publishes running watts, transfer-switch ratings, or fuel consumption in one place. Every record in this database was built by hand from manufacturer product pages, spec-sheet PDFs, and owner's manuals, one model at a time. That is slower and smaller in scale than an API-driven dataset, and it's why the database currently covers 86 models across 12 brands rather than thousands.

Every generator page links to its source or sources in a Sources section, so you can verify any figure against the document it came from rather than taking our word for it.

The verification flag

All 86 models have been verified against manufacturer documentation, so no record currently carries the internal needs-verification status. That flag marks a record built from the best available public source but not yet cross-checked against a second source or a manufacturer spec sheet; every record has now cleared it. If you are relying on a figure for a purchase or installation decision, still treat it as a starting point to confirm against current manufacturer documentation, since specifications change between model years.

Sourcing mix

Most records link directly to a manufacturer's own domain (Generac, Rehlko/Kohler, Cummins, Briggs & Stratton, Champion, Westinghouse, DuroMax, Honda, Firman, WEN, Pulsar, and others). A smaller share link to an authorized retailer's product page or spec-sheet PDF where the manufacturer's own listing was incomplete or unavailable — those pages are used only when they reproduce the manufacturer's published specifications, not retailer marketing copy.

Update frequency

The dataset is refreshed by hand as new models are added or existing listings change. Each entity stores the date its record was retrieved; the sitemap's lastmod and this page reflect the newest retrieval in the current build. Current data snapshot: 2026-07-24 (86 generators across 12 brands).

Data cleaning

• Model names and part numbers are normalized to match how retailers list them — a manufacturer's internal SKU is shown alongside the marketing name where both exist.
• Wattage figures are recorded per fuel exactly as published: runningWatts / startingWatts are the headline rating (gasoline for portables, propane for standby units, which have no gasoline mode), with separate *Lp and *Ng fields for dual- and tri-fuel derated ratings. A model with no published figure for a given fuel simply has no value for that field — nothing is estimated or backfilled.
• A small set of manual overrides corrects attributes against updated manufacturer documentation as errors are found. Overrides are reviewed, not automated.

Formulas

These are the exact formulas behind every computed number on the site:

• Whole-house coverage verdict: banded from the highest published running watts across fuels — under 5,000 W is essentials-only, 5,000–9,500 W partial-home, 9,500–15,000 W most-circuits, 15,000 W+ whole-house. These bands follow common sizing guidance (fridge + furnace + lights ≈ 3–5 kW; add central air and a typical home needs 9–15 kW; a 200 A whole-house service wants 15 kW+) and are a planning estimate, not a load calculation for your specific panel.
• Runtime at 25% load: uses the manufacturer's published figure when available; otherwise estimated as the published 50%-load runtime × 1.6, reflecting that fuel burn does not scale linearly with load — engine overhead means a quarter-load burn is roughly 60–65% of the half-load burn, not half.
• Estimated daily fuel cost (24 hours at ~50% load): gasoline portables use tank size ÷ published half-load runtime; natural-gas standby units use published half-load ft³/h × 24 (≈97 ft³ per therm); propane units use published half-load gal/h × 24. National average rates used as defaults: $3.2/gal gasoline, $2.7/gal propane, $1.45/therm natural gas (EIA averages) — override them with local prices on the fuel cost tool.
• Recommended inlet size: uses the published transfer-switch-ready outlet amperage when available; otherwise inferred from running watts and output voltage (240 V units above 7,200 W need a 50 A path; 240 V units at or under 7,200 W fit a 30 A path; 120 V-only units at or under 3,600 W fit the 120 V 30 A path; above that, no recommendation is shown, since a 120 V-only unit has no standard higher-amperage inlet convention).

Limitations

• Manufacturer-published numbers, not independent lab tests. Running watts, noise ratings, and runtime figures are the manufacturer's own published specifications; BackupAtlas does not run generators or verify performance independently.
• Curated, not comprehensive. The database covers a curated selection of current models across major brands, not every generator sold in the United States. Absence here is not a judgment on a model's quality.
• No installation-cost modeling. Transfer-switch and interlock installation, gas-line work, concrete pads, and permit costs vary by home and by region and dominate the real project cost for standby systems; the guides explain what to check with a licensed professional.
• Planning estimates, not a substitute for a load calculation. Sizing bands, inlet recommendations, and cost estimates are deliberately simple and stated plainly so you can see exactly what they assume; a licensed electrician's assessment of your actual panel and loads always takes precedence.

Data license

The raw specification rows in this database — as published on entity pages and in llms-full.txt — are available under CC BY 4.0: reuse them freely with attribution to BackupAtlas (a link to this page is ideal).

Corrections

If any specification disagrees with the manufacturer's current documentation, contact us with a link to the authoritative source. Verified corrections ship in the next build; systematic ones become curation overrides so they survive dataset refreshes.