Corrections

Six so far. Each one names the pages it affected so you can see what the wrong version said and where the corrected figure now lives. The standard behind this page is that a correction gets published with the old and new numbers side by side rather than quietly edited out, and the open items list carries the ones not resolved yet.

13 September 2026: a flip rate counted over more states than the state count beside it

What was wrong. The volatility page reported both a share of season pairs that changed answer and a count of states that flipped, in the same sentence, as though they came from the same set. They did not. The pair level figures were accumulated over every state with a usable price history, and the state level count then dropped every state without a current gas price before counting.

Effect. None on the pair level figures, which were right and are unchanged: 760 pairs, a 7.1% median move, 37% of moves over 10%, a 62% maximum. What was wrong was the denominator printed next to them, which made the flip rate look like it came from 45 states when it came from 51. Six states have sixteen years of history but no current gas price to compare against, so they were counted in the numerator and excluded from the denominator.

How it was found. By adding a chart. The chart draws one line per state and drew 51 of them, directly under a sentence saying 45, which is the kind of disagreement a picture makes impossible to leave alone and a table hides.

Fix. Every state with a usable history now counts toward the state level figures. A missing current price only removes a state from the table that has a column for today's margin, which is the one figure that genuinely cannot be filled in for it.

31 August 2026: the reason given for the 10% tie band was wrong

What was wrong. Every verdict here carries a tie band: inside 10% the site refuses to call a winner. The justification printed alongside it, on the about page, the homepage, the comparison pages and in the source comment on the constant itself, was that fuel prices move more than 10% between seasons.

They do not, typically. Across 760 state-season pairs over 16 heating seasons of EIA data, the median move in the electricity to gas price ratio is 7.1%.

Effect. None on any number. The 10% band itself is unchanged and no verdict, calculator or dataset figure moved. What was wrong was the reasoning offered for it, in seven places in the prose and once in a code comment.

The band survives on a better argument. It is earned by the tail rather than the median: 37% of season-to-season moves exceed 10% and the largest was 62%, so a margin of 8% is not safe merely because a typical year moves 7.1%. Roughly one season in three moves further than the band, and nobody knows in advance which.

How it was found. By finally checking. The claim had been on the site since the first build, was never tested, and sounded obviously right, which is why it survived. The full analysis is here, including the finding that 31 of 51 states changed answer at least once in 16 years.

Fix. All seven prose instances and the source comment now state the measured figures and link to the analysis.

31 August 2026: a backup heat figure that was typed rather than computed

What was wrong. The dual fuel page stated that fifteen percent of seasonal heat coming from electric resistance strips costs 27% of a heat pump's efficiency. The correct figure is 25%.

How it happened. The sentence around it computed the two coefficients of performance from the model, and then the percentage between them was typed in by hand. Two derived numbers and one asserted one, sitting in the same sentence, which is exactly the arrangement that lets an error survive.

Effect. Small, and confined to that one sentence. No calculator, verdict, state page or dataset figure used the wrong number; the model has always computed it correctly. What was wrong was the description of the model, not the model.

How it was found. Building the backup heat calculator, which computes the same quantity from the same two functions and disagreed with the page.

Fix. The percentage is now derived from the model like the two figures either side of it, so it cannot drift again. A check across the source found no other hardcoded figure of this kind.

31 August 2026: summer gas prices were inflating every verdict

What was wrong. The first build of the rate pipeline took each state's most recent monthly EIA figure. For natural gas that resolved to May or June 2026.

EIA's residential natural gas price is total revenue divided by total volume. In summer, residential volume collapses toward zero while the fixed monthly customer charge stays, so the apparent price per thousand cubic feet roughly doubles. The site was answering a heating question with summer prices.

Effect. Gas looked far more expensive than it is during heating season, which biased every verdict toward the heat pump.

A second and quieter bug sat behind the same line of code. Taking the most recent available month resolved to different months for different states, because states report on different lags. California came back as March while most others came back as May, so the states were not comparable to each other.

Fix. Both electricity and gas are now the mean of the most recent complete December to February heating season, using identical months for every state, with those months recorded in the data file and printed on every page. States without complete season data are excluded rather than estimated.

Effect of the fix. Under a cold climate heat pump against a 95% AFUE furnace, the number of states where the heat pump costs less to run fell from 19 to 4.

A second correction came out of this one. The fix above dropped any state without a complete gas season, which removed six of them from the site entirely: Florida, Louisiana, Maryland, Maine, Minnesota and New Mexico. That was the wrong rule. It is right for a gas comparison and wrong for every other comparison here, and it was worst for Maine, a state that heats overwhelmingly with oil, whose price EIA publishes, and which had four households in the dataset with no page to put them on. A missing fuel now removes a column rather than a state. All six have pages, the site went from 45 states to 51 including the District of Columbia, and the gas column on those six says not published rather than carrying a guess.

Where this now lives. The full write-up is in the note on the gas price most calculators get wrong, the fixed pipeline is described in how this site is made, and the corrected figures are on every state page and in the gas furnace comparison.

31 August 2026: the share of US homes heating with gas was wrong

What was wrong. Several pages stated that roughly 40% of American homes do not heat with natural gas. I had carried that figure from memory rather than checking it.

The correct figure. Natural gas is the main heating fuel in 47% of US homes, down from 49% in 2010, and 42% of households report electricity as their main space heating fuel. So slightly more than half of American homes do not heat with gas, not 40%.

Source. U.S. Energy Information Administration, Electricity use is becoming more common for residential heating, citing the U.S. Census Bureau American Community Survey and the Residential Energy Consumption Survey.

Effect. No calculation changed. The claim was context rather than an input, but it was stated as fact on several pages and it was wrong in the direction that understated how many households the heat pump case applies to. Corrected across the site, including the electric resistance comparison, where the size of that population is the whole reason the page exists.

31 August 2026: I called a homeowner's data impossible and I was wrong

What was wrong. I published an analysis of self-reported heat pump usage from a household in southwest Indiana: 2,160 sq ft, 499 kWh in January. Working backwards I got an implied heat loss coefficient of 1.68 BTU per square foot per degree day, called it "not possible as a whole house figure", and concluded the 2-ton unit could not be carrying the whole load.

Why it was wrong. Two reasons, and the first is mine. I used an estimated January degree day figure derived from an annual normal rather than the homeowner's actual 1,097 HDD, which alone moved my number from 1.68 to 1.82. Second, and more importantly, I reasoned from a typical house instead of reading what the owner said about theirs.

What the owner actually documented. Walls at R-21 and attic at R-83, upgraded from R-23. Ducts and air handler inside the conditioned space. A bi-level with the lower floor half below grade, which the owner estimates cuts the load on those 1,080 sq ft by around half. Simple rectangular plan, 8 foot ceilings, no dormers, replaced windows. A stated design load of 10,200 BTU/hr at 10F, against unit capacity of 15,200 BTU/hr, giving a balance point near 1F. That design load implies a coefficient of 1.89, and my corrected back-calculation from metered energy gives 1.82. Those agree to within 4%.

The house is real and it is unusually good. The owner also cross-checked their own coldest day against manufacturer rated power draw and landed within 1% of what the Daikin report said. Two independent routes, monthly and single day, imply a January COP of 2.63 and 2.59.

Effect. No calculation elsewhere on the site changed, because I had excluded the figure rather than used it. What changed is the conclusion I published about someone else's data, which was wrong and is now rewritten in the fieldwork note. The lesson I have taken is that "this number looks impossible" and "this number contradicts my assumptions" are different statements, and I published the second while believing the first.

How to report an error

Write to banerjeeashutosh01@gmail.com. Include the page, the number you think is wrong, and what you believe it should be. Confirmed errors are published here with both figures and the date they were fixed.