Methodology

Last updated July 12, 2026

Every number on this site is either sourced (a certified spec we pulled from a public registry, cited to it) or computed (a figure we derived with a formula we publish below). We never blur the two, and we never invent a spec — a missing value is shown as "not published," not estimated.

How we source performance specs

Efficiency ratings (SEER2, EER2, HSPF2, AFUE) and capacity figures (cooling at 95°F; heating at 47/17/5°F) come from two public sources, joined on the AHRI reference number:

• AHRI Directory — the industry certification database. AHRI's data carries redistribution terms, so we publish our computed layer and cite the AHRI reference number for each unit rather than republishing their table verbatim.
• ENERGY STAR certification data — supplies the ENERGY STAR, Most Efficient, and Cold Climate designations, and carries the AHRI reference number that lets us join the two.

Where a registry doesn't disclose a field, that field renders as "not published." We don't interpolate.

The cold-climate truth: COP and capacity at 5°F

A heat pump's marketing will say it "works down to −15°F." What actually matters is two certified numbers at 5°F: the COP (how much heat it moves per unit of electricity) and the heating capacity (how many BTU/h it can still deliver when it's frigid). We surface both on every heat-pump page and rank our cold-climate guide on them.

How we compute operating cost ($/MMBtu)

Our headline metric is the cost to deliver one million BTU (one MMBtu) of heat. It's the "cost per calorie" of home heating, and it's what makes an efficient heat pump the cheapest box in the house to run.

Constants: 1 kWh = 3,412 BTU; 1 therm = 100,000 BTU. Default prices are EIA U.S. averages (EIA US avg 2026: $0.1883/kWh, $1.68/therm) — the operating-cost calculator lets you enter your own. At U.S. averages, an efficient heat pump delivers heat for about $10.76/MMBtu versus $55.19/MMBtu for electric resistance.

Savings vs. electric resistance is simply (1 − heat-pump $/MMBtu ÷ resistance $/MMBtu) × 100. These are steady-state estimates; real bills vary with weather, defrost cycles, and backup-heat use.

How we estimate the size you need

The sizing calculator gives a ballpark, not a Manual J. Cooling load ≈ conditioned square footage × a per-square-foot BTU factor for your IECC climate zone (roughly 22 BTU/sq ft in hot zones down to 15 in the coldest, reflecting a modern, reasonably insulated home); heating load uses a larger factor that rises with cold (about 18 BTU/sq ft in hot zones up to ~58 in the coldest). Both are scaled by an insulation/air-sealing factor (0.85–1.2) and a ceiling-height factor, then cooling is rounded to the nearest half-ton and shown as a ±8% range. These rule-of-thumb factors deliberately cannot see your windows and their orientation, real air-sealing, duct leakage, shading, or occupancy — which is exactly what a licensed ACCA Manual J load calculation measures. Use our number to sanity-check a contractor's, not to place an order.

How we compile installed-price ranges

Installed-cost ranges on product pages and the costs page are national figures for a standard like-for-like replacement (equipment + labor + permit, reusing existing ductwork and electrical), reconciled from published 2026 market data — EnergySage's real-transaction averages, Modernize's homeowner-quote data, National Air Warehouse and HomeGuide's AFUE-tier furnace figures, This Old House, and Trane's own branded pricing, among others. Two rules keep them honest: (1) where sources disagreed materially we widened the range rather than averaged; (2) several lead-gen sites publish "installed" numbers that omit labor or permits and run about half of real quotes — those under-scoped lows are trimmed out, not blended in. No source publishes a clean size-by-efficiency price matrix, so a unit's exact size (tonnage) or efficiency (AFUE) places it within a range; we never present a range as a firm quote, and when a spec needed to place a unit isn't published we show the full span and say so.

How we compute water heater running cost (UEF)

Water heaters aren't rated on SEER2/HSPF2/AFUE — they're rated on UEF (Uniform Energy Factor), the DOE's current test metric (it replaced the older "Energy Factor" in 2017). UEF is only comparable within the same fuel type — a heat pump water heater's UEF (~2.8–4.5) and a gas tankless unit's UEF (~0.8–0.98) are measured against different DOE test procedures, so we rank each fuel type against its own field, never across fuel types.

13.12 MMBtu/year is the DOE's own fixed medium-usage draw pattern (64.3 gal/day, 67°F rise) — the same baseline UEF itself is defined against, applied identically across every fuel type so the numbers are apples-to-apples (this mirrors how DOE's own EnergyGuide label cost estimate is built). Dual-fuel (natural gas/propane) units are costed on the natural-gas basis, disclosed rather than silently assumed. No U.S.-average heating-oil price exists in our data, so oil-fired units show no annual-cost figure rather than a guessed one.

At these prices, a heat-pump water heater runs for about $196/year in our corpus's median, versus $787/year for plain electric resistance — data current as of July 18, 2026. Every water heater's diy_buyable flag is sourced per brand from evidence-cited spot-checks (a brand's own retail-channel statement or its presence/absence across major DIY retailers) — brands we haven't spot-checked resolve to "not yet researched," never a guessed yes or no.

Where water heater dimensions come from

Physical dimensions (height & diameter for storage tanks; height, width & depth for tankless units) come from the ENERGY STAR certification record matched to each AHRI unit — not AHRI itself, which doesn't publish physical size. Only the ES-matched slice of the corpus carries a dimension: 985 of the active corpus (20.4%) as of this build. Everything else shows "not published" — we don't estimate a tank's size from its capacity or fuel type.

A plausibility guard rejects values outside a realistic real-world range for that unit type (0 rejected this build) so a data-entry artifact (a stray "1" placeholder, a decimal-place slip) never gets published as a real measurement. Separately, a live spot-check against manufacturer spec pages caught one confirmed cross-source conflict: ENERGY STAR's published dimensions for Rinnai's REHP65 family (6 models) contradict Rinnai's own current spec page by about 19.5 inches of height — a wrong-but-plausible value that the guard above can't catch on its own. That family's dimensions are nulled rather than shown with a caveat (6 units affected) — a footnote doesn't stop a wrong number from being read at face value; only nulling does. Where a dimension is published, we also show one physically-true fit comparison — whether the unit's width or diameter clears a standard 32-inch interior doorway — never a broader, unverifiable claim like "fits most homes."

How we compile water-heater installed-cost ranges

Installed-cost ranges on water-heater product pages and the costs page are national ranges for a standard like-for-like replacement (equipment + labor + permit, same fuel and location/venting), reconciled from 4+ published 2026 cost-guide sources per segment (National, like-for-like-replacement installed-cost ranges (equipment + labor + standard permit) reconciled across 4+ published 2026 cost-guide sources per major segment. Where sources disagreed materially, ranges were WIDENED to the outer bounds rather than averaged. Figures that were clearly unit-only, labor-only, or otherwise under-scoped relative to a source's own fuller breakdown were trimmed and are logged in SOURCES.md, not folded into the low/high bounds. All figures assume standard/like-for-like replacement (same fuel, same location/venting) unless a segment is explicitly labeled a conversion/upgrade scenario. Regional, local-permit, and home-specific variation is real and unquantified in free-tier published data; ranges are national. No figure here is a quote for any specific home. Federal 25C Energy Efficient Home Improvement Tax Credit expired 12/31/2025 per OBBBA (signed 2025-07-04) and was cross-checked live this session (DSIRE, RewiringAmerica.org/federal-incentives/25c-heat-pump-water-heater-tax-credits, hvacbase.org) — several source pages used for cost figures still present the expired 25C credit as active (EnergySage heat-pump-water-heater-guide dated Sept 2024, Rewiring America project page dated 2023, Today's Homeowner heat-pump-water-heater-cost dated April 2025, pieceofmindplumbing.com tankless guide). Those incentive claims were excluded entirely; only the cost figures from those pages were retained.). Same two honesty rules as the HVAC ranges: where sources disagreed materially we widened the range rather than averaged, and under-scoped figures that were clearly unit-only or labor-only (not the full installed job) were trimmed out and logged, not blended in. Most segments have no single defensible "typical" figure in the source data — we show the honest wide range rather than inventing a midpoint; the two segments that do carry an explicit typical (gas tankless replacement, heat-pump water heaters) say so plainly, not as the number to expect.

Segment-specific notes: heat-pump water heaters carry an electrical-adder note (a new 240V circuit, and occasionally a full panel upgrade, are usually needed — folded into the range, not extra) plus a 120V-plug-in exception for the specific models that support it (do not assume 120V compatibility without confirming the exact model). Gas tankless shows a second, higher range for converting FROM an existing storage tank (gas-line upsizing and venting changes cost more than a straight tankless-for-tankless swap) — the two are never blended. Propane storage-tank installs are flagged as a thin-source segment: unlike gas or electric, this figure traces to essentially one underlying survey methodology echoed across a few sites, not several independent ones. Dual-fuel (natural gas/propane) units are priced on the natural-gas basis, disclosed rather than silently assumed — the same basis their running-cost figure uses. Oil-fired units show no installed-cost panel; no published range exists for that segment. None of these figures are a quote for any specific home — get 2–3 quotes.

How we source recall history

Recall history on water-heater product pages comes from the U.S. Consumer Product Safety Commission (CPSC, saferproducts.gov), covering recalls from RecallDate >= 2016-01-01 (calendar-year cutoff, ~last 10 years as of pull date; see script docstring for why a calendar-year cutoff is used instead of a rolling day-exact window). Every corpus brand is checked, including brands with zero recalls — a clean record is shown as an explicit positive statement ("No CPSC water-heater recalls on record for [brand], checked July 2026"), not silence, because silence reads as "we didn't check."

Two limits worth knowing: (1) this is a brand-level signal only — CPSC's own recall records don't publish a model number for water heaters, so we can say a brand had a recall, never that this specific model was the one recalled; and (2) complaint counts are not publicly available — no bulk incident/complaint API exists for this category, so we show only formal recalls, never an estimated complaint volume. Brand attribution uses a strict whole-word/phrase match against each recall's own title and text (9 recalls attributed across 8 brands as of this pull) — no fuzzy matching, so a recall is never attributed to a brand it doesn't actually name.

How we source rebates

Incentive data is derived from DSIRE (Database of State Incentives for Renewables & Efficiency). Note the federal 25C tax credit expired 12/31/2025 — the live layer is state, utility, and HEEHRA programs. Rebate rules change constantly, so every rebate page carries a review date.

What we don't do

• No star ratings. We don't run a proprietary score that collapses efficiency, capacity, and price into one undisclosed number. We show the sourced numbers and let you sort by what matters to you.
• No pay-for-placement. Rankings are strictly by a stated, verifiable metric, computed identically for every unit.
• No invented specs. If a registry doesn't publish a value, we mark it "not published" rather than estimate it.

Freshness

Stale data is the incumbent failure mode. Our corpus is refreshed from AHRI, ENERGY STAR, and DOE CCMS, and every page shows the date its data was last refreshed — no vague "recently updated" claims.

Affiliate disclosure

See the full affiliate disclosure. In short: some buy links (Amazon) earn us a commission; they never influence which equipment appears or how it's ranked — ranking is strictly by the stated, verifiable metric.

Corrections

Found a wrong spec, a stale figure, or a mismatched AHRI reference? Email erinrose451@gmail.com with the model number and what's wrong. We fix verified errors and update the "last updated" date on the affected page.

Download the data

The full derived dataset — every certified product, its sourced specs and our computed operating cost — is published machine-readable and licensed CC-BY-4.0: download /dataset.json. No affiliate parameters, no sign-up.