Research log

Building the model

The first step was deriving the breakeven relationship from first principles rather than adopting a published rule of thumb. Running cost for a combustion furnace is the useful heat divided by AFUE, priced per unit of fuel. For a heat pump it is the useful heat divided by seasonal COP, priced per kWh. Setting the two equal and rearranging gives a threshold that depends only on effective COP divided by AFUE.

That result was checked two ways. Working in dollars per kWh of chemical energy and working in dollars per delivered BTU produce the same threshold, which they should. The derived figure for an HSPF2 11.0 unit against a 95% AFUE furnace came out at 3.39, close to the 3.5 that circulates in the trade press, which suggested the published rule of thumb quietly assumes that specific pairing.

The sanity check that mattered

After the first data pull, spot checking raw prices showed Georgia residential gas at $33.28 per thousand cubic feet and Texas at $30.48. Typical heating season figures sit in the $10 to $18 range. That gap was large enough to indicate a systematic problem rather than an unusual month.

The cause was the summer revenue over volume distortion described in the corrections log. The check that surfaced it was simple: print the raw numbers and compare them against a remembered range before trusting anything derived from them.

What was verified

• Unit conversions checked against independent derivations, including kWh per therm at 29.3071.
• The client side calculator and the build time model produce identical verdicts for the same inputs. New York returns 94% more expensive on both paths.
• The effective COP function behaves correctly at both limits: zero backup returns the heat pump COP, full backup returns 1.0.
• Every state page is generated from the same data file as the calculator, so the two cannot disagree.

31 August 2026: the heat loss coefficient was tested against a real load calculation

The weakest input on this site is the heat loss coefficient of 8 BTU per square foot per heating degree day, used to turn floor area into an annual heating load. It is standard practice rather than a measurement, and every page using it says so.

I commissioned a Manual J for a 2,000 square foot house in Cincinnati at a 5F design temperature. It returned a heating load of 42,000 BTU/hr, which converts to 7.75 BTU per square foot per degree day, within 3.1% of the assumed figure. Separately, a self-reported new build with open cell spray foam implied 4.80, close to the 5 the calculator offers for a tight envelope.

Two independent sources landing near the middle and tight ends of the range is not a validation study, and the wording on the calculator still describes the coefficient as a rule of thumb. It is more than the site had before. Written up in asking three contractors the same question.

Open items

• Propane and heating oil prices are national placeholders. EIA publishes both regionally and they should be wired in.
• Six states lack complete heating season gas data and are currently excluded.
• Utility level heat pump tariffs are not modelled. Several states now offer them and they can change the answer.
• Balance point and backup heat fraction are user inputs. A climate aware default would be better than the current flat 10%.