Editorial policy

The mathematical conclusion is never adjusted to make a product look better. Three findings currently on the site argue against a purchase:

• A home battery recovers its cost on bill savings inside its warranty in 0 of 51 states. In 5 states it has no bill-saving mechanism at all.
• A heat pump is cheaper to run than gas in 17 of 51 states, not all of them.
• Solar recovers its cost within ten years in 12 of 51 states.

Those numbers are recomputed from the calculation engine every time the site builds. If a data refresh changes them, this page changes with it — they are not typed in.

Every figure published anywhere on the site — state pages, category hubs, rankings, the homepage — is produced by the same calculation engine. No number is ever hand-copied between pages, so a data update moves all of them at once and they cannot silently disagree.

Models are documented in full on the methodology page, including the formulas and every constant. The constants shown there are imported from the calculation code rather than restated, so the documentation cannot drift from the implementation.

We prefer primary public datasets — EIA, NOAA, AAA, PVGIS — and we say which one every field comes from, with its vintage and retrieval date, on the data sources page.

Where a field is an informed estimate rather than a sourced measurement, we list it as an estimate. We do not attach a plausible-looking citation to a number we did not actually take from that source, and we do not substitute model-generated values for public data.

Every result is a modelled estimate built from state averages, not a quote for a specific property. We state the assumptions rather than burying them, expose the ones that matter most as adjustable inputs, and name what each model excludes.

Where an assumption cuts against a product we say so plainly — for example, heat-pump efficiency falling with climate severity, or the round-trip losses that reduce battery arbitrage.

When we find an error we fix it and record it. The changelog documents both data refreshes and changes to the models themselves, including the direction and size of the effect.

A worked example: in August 2026 we found the solar production model was running about 9% high because a flat loss factor could not capture temperature and angle-of-incidence losses separately. Fixing it lengthened every solar payback by roughly a year. That is recorded in the changelog rather than quietly corrected.

To report a suspected error, cite the page and the figure. Corrections that change a published number are logged.

At the time of writing Payback Math carries no advertising, no affiliate links, no sponsored placements and no lead-generation partnerships, and takes no payment from any manufacturer, installer or utility.

We are not going to claim permanent independence as a marketing line. If that changes, the change will be disclosed on this page before it goes live, the commercial relationship will be labelled where it appears, and the rule that calculations are never adjusted to favour a product will continue to apply. Any monetisation that would require softening an unfavourable conclusion is incompatible with the point of the site.

What is changing, and what is not. We intend to earn money by referring readers to installers when — and only when — the maths says a purchase is worth investigating. The machinery for that is now built and is switched off: there is no approved partner, no affiliate link anywhere on the site, no tracking script, no cookie and no analytics provider. When a partner is approved, any paid link will be labelled as such at the point it appears, and this paragraph will be updated before it does.

The part worth holding us to is the separation. Commercial configuration cannot reach a calculation: every payback, rank, verdict and headline figure is computed before any partner is consulted, and an automated check runs all four calculators across all 51 jurisdictions with the commercial layer active and inactive, failing the build if a single financial output moves. If a partner one day pays more for California, California does not become a better place to buy solar.

The visible consequence is that we decline to sell on a bad result. A state where a heat pump costs more to run than a furnace gets a plain statement that we would not recommend switching, and no quote button — currently 33 of 51 states. Where a purchase misses the mark on price alone, we publish the price it would have to hit instead, which is useful whether or not you ever click anything.

The calculations are deterministic code, not model output. We do not generate filler articles, spin up near-duplicate pages, or publish statistics that were not produced by the calculation engine.

We also do not create pages purely for search engines: crawlers and people are served exactly the same content, and we do not build geographic pages unless the underlying data genuinely produces a different answer for that location.