Data sources

The question worth answering is not how many of our fields we can call “sourced”. It is whether you can tell, for any number on this site, where it came from and how much weight to put on it. So every input carries one of these labels:

Sourced data

An externally observed value from a named dataset.Marked on the site as Sourced

Modeled assumption

A baseline we chose because no single objectively correct value exists.Marked on the site as Assumption

Derived value

Calculated by Payback Math from the inputs above.Marked on the site as Calculated

Source unresolved

Presented as fact but not yet traced to a dataset we have verified. Treat with caution.Marked on the site as Source unresolved

Current tally: 9 sourced, 27 modeled assumptions, 12 calculated outputs, and 2 with unresolved provenance. That last figure is published rather than quietly reclassified. Driving it to zero by calling unresolved fields “assumptions” would be the easiest change on this page and the most dishonest.

9 inputs come from 8 external datasets. Vintage is the period the data describes; retrieved is when we fetched it. A 2023 dataset pulled in 2026 is not stale if 2023 is the latest complete release — but you should be able to see that for yourself.

These 2 inputs are presented on state pages as though they were measured, but we have not traced them to a dataset we have actually verified. They are the weakest numbers on the site. Each one says what is missing and why we could not close it:

• Annual outage durationSource unresolved

  Currently 2 – 13 hours/yr · used in battery

  Should come from EIA-861 reliability (SAIDI, with major event days stated explicitly). The EIA-861 archive we tried returns a stub. Note SAIDI is average interruption duration per customer — not the same as 'hours without power' for any individual home. Used only to frame backup value, which is deliberately excluded from payback.

• Solar export compensation policySource unresolvedmaterially affects the result

  Currently none / partial / full · used in solar, battery · editable in the calculator

  The regulatory structure — whether a state credits exports at the retail rate, runs net billing, or leaves it to each utility. Seven jurisdictions are now verified against a primary regulator or DSIRE's published policy notes (California, Arizona, Texas, Idaho, Vermont, New York, Oregon); the other 44 carry the site's earlier three-bucket assumption and are openly flagged as unverified rather than promoted. DSIRE's per-state records sit behind a licensed API and its summary maps encode states as unnamed polygons, so reading 51 classifications out of map colours would be inferring data from a picture. Full per-jurisdiction records in src/data/exportPolicy.js.

These 27 values were chosen by us. In most cases no objectively correct number exists — installed prices depend on the installer, tariffs depend on the utility, and a house is not a state average. Publishing a chosen value as a state fact would be false precision, so each is labelled and, where it matters most, made editable so you can substitute your own.

Everything else on the site is arithmetic. A calculated value needs no third-party source — it needs a formula, named inputs and a version number, all of which are on the methodology page. The main published outputs and what each is computed from:

• Solar system size — from monthlyKwh, kwhPerKwYear
• First-year solar generation — from systemKw, kwhPerKwYear
• Payback period — from netCost, annual savings, degradation, price escalation
• Lifetime net gain — from annual savings, analysis period
• Cost per mile — from rateCentsPerKwh, EV_MILES_PER_KWH, CHARGING_EFFICIENCY
• Break-even gasoline price — from EV running cost, gallons
• Seasonal COP — from hdd
• Annual heating demand — from hdd, DEFAULT_HOME_SQFT, HEAT_LOSS_PER_SQFT
• Battery annual bill saving — from touSpreadCents, the jurisdiction's export compensation, ROUND_TRIP_EFFICIENCY
• State ranking — from the calculation engine across all 51 states
• National median — from the calculation engine across all 51 states
• Financial verdict — from payback period, product lifetime or warranty

A worked example of that tracing, for one calculator across all 51 jurisdictions, is published as heat pump vs gas furnace economics by state, which carries its own data, assumptions and limitations alongside a downloadable snapshot.

Because every calculated value traces back to the tables above, a calculated number is exactly as trustworthy as its weakest input — not more. A payback figure built partly on an unresolved input inherits that uncertainty, which is why the unresolved list is published rather than buried.

One input on this site is a legal fact rather than a measurement, and it can change the day a bill is signed. It is therefore recorded with the provision, the authority and the date we last checked it, and it is keyed to the model year rather than treated as permanent.

Tax eligibility depends on individual circumstances. This is the baseline federal treatment for a typical homeowner-owned installation and is not tax advice.

Roughly 60% of what a rooftop system generates is exported, so what an exported kilowatt-hour earns drives close to half of every solar result on this site. It is also the input where a tidy state-level number is most tempting and least honest, so the two questions behind it are recorded separately.

The solar and battery calculators now read the same record. They price the same transaction — one exported rooftop kilowatt-hour under the state's rules — once as income to a solar owner and once as the income a battery owner gives up by storing it instead. Until recently each calculator carried its own copy of the assumption and they disagreed in 22 states; that duplication has been deleted, so there is one export value per jurisdiction and one place to change it.

Where compensation is not at retail we model it in cents per kilowatt-hour rather than as a share of retail — avoided cost follows wholesale generation value and does not rise with a state's retail rate. The 6¢/kWh baseline is anchored on Arizona's published Resource Comparison Proxy export tariffs, the only residential export rates we confirmed with a regulator.

DSIRE's May 2026 national summary records 28 states and DC with net metering, 16 with net billing or other compensation, and 1 with none. Its per-state records sit behind a licensed API and its summary maps encode states as unlabelled map shapes, so we did not try to read 51 classifications out of map colours — that would be inferring data from a picture. The unverified rows above say so plainly instead.

• Gasoline is a spot price. AAA state averages on the retrieval date, not an annual mean. Pump prices move 30–50¢ seasonally, so the EV comparison drifts between refreshes.
• Electricity is a partial-year average. Year-to-date through May, so summer-peaking states may run slightly higher across a full year.
• Solar resource is reanalysis, not satellite. PVGIS-ERA5 is roughly 30 km resolution — coarser than NREL's satellite-derived NSRDB, which is unreachable from our build environment. Optimal tilt also flatters a real roof: east/west facing or shaded arrays yield less.
• Mileage is per licensed driver. FHWA vehicle-miles divided by licensed drivers. That matches our one-driver EV comparison, but it is not miles per household, and states carrying heavy through-traffic on interstates read high.
• Washington DC has no NOAA HDD series. Maryland is used as the proxy.

Figures on this site are free to cite with attribution to Payback Math and a link to the page the number came from. Where a figure originates with EIA, PVGIS (EU JRC), AAA, NOAA CPC, FHWA, IRS, please cite them as the underlying source — our contribution is the modelling, not the raw measurement.

The models themselves are documented on the methodology page, and every change to data or formulas is recorded in the changelog.