Every number on this site, and where it came from
The table below is generated from the same data module the calculators read, at build time. It is not a hand-written summary of our sources — it is the sources. If a figure changes in the module, this page changes with it, and if a page tries to display a figure that is not in the register the build fails rather than printing an unsourced number.
How to read the labels
Every figure carries exactly one of three labels, and the distinction between them is the most important thing on this page.
sourced means the number is quoted verbatim from the cited page. There is a stored snapshot of the quote, the URL and the retrieval date, and where the live page blocks automated readers the figure was verified against a named Internet Archive capture of that exact URL rather than from memory.
derived means we did arithmetic on sourced data. The arithmetic is written out further down this page. A derived figure inherits every uncertainty of its inputs and adds whatever the operation assumes.
assumed means it is ours, with no direct source behind it. There are 6 of these on the entire site and every one is named in its own section below. Wherever an assumed figure reaches a page, the page says so at the point of use, not in a footnote.
Register size: 108 figures, of which 80 are sourced, 22 derived and 6 assumed, drawn from 10 distinct named sources. Every figure was retrieved on 2026-08-06. A figure older than 183 days blocks a new deployment.
The complete figure register
Access mode, and the sources that block automated readers
Two of the cost publishers we cite return an HTTP 403 to anything that is not a browser. That is their right and it is common. It also means a figure taken from them cannot be re-verified by a simple fetch, so those figures were checked against a named Internet Archive capture of the exact cited URL and the capture timestamp is recorded beside the figure in the data module. The alternative — quoting a bot-walled page from memory and hoping — is how wrong numbers survive for years.
The complete list of named sources behind this site:
- 10 CFR 430.32(c)(5), Energy and water conservation standards
- City of San Diego Development Services, Information Bulletin 103, Fee Schedule for Mechanical, Electrical, Plumbing/Gas Permits (revised May 2026)
- ENERGY STAR (U.S. EPA), Air-Source Heat Pumps key product criteria
- ENERGY STAR (U.S. EPA), Room Air Conditioners
- Forbes Home, How Much Does It Cost To Install A Mini Split AC System? (audited and verified December 1, 2024)
- HomeGuide, Cost to install a dedicated circuit (October 15, 2025)
- HomeGuide, Ductless mini split installation cost (updated October 8, 2025)
- Internal Revenue Service, Energy Efficient Home Improvement Credit
- U.S. Energy Information Administration, Electric Power Monthly, Table 5.6.A (May 2026 data, released July 23, 2026)
- minisplitcostcalculator.com — our own labelled adjustment, disclosed on the methodology page
Note the concentration. Government and standards sources carry the sizing, the efficiency thresholds, the electricity prices, the degree days and the tax position. Installed cost has no government source anywhere, so it rests on two commercial cost aggregators, and that is the weakest part of this site by a distance. We cite both, print their disagreements rather than averaging them, and never blend a low end from one with a high end from the other.
The derived arithmetic, written out
Ceiling height scaling
The sizing table states a 8-foot baseline and the direction to move for taller ceilings, without giving a formula. We scale in proportion to volume: factor = max(1, ceiling feet / 8). The floor at 1 is deliberate — a low ceiling never becomes a discount the source did not offer.
Indoor unit style premium
premium = style band − wall band, computed inside one publisher’s own style table so both operands were gathered the same way, and applied once per zone.
Zone counts nobody published
Our zone source publishes rows for one, two, three, five and eight zones only. For four, six and seven we interpolate linearly between the adjacent published rows and relabel the result derived. Past the top published row we return nothing.
Where the point estimate sits inside a band
The midpoint is 0.5. Each named complexity factor — a long line-set run, awkward mounting, a ceiling-mounted indoor unit — moves the point estimate 0.125 of the band further toward the top, capped so it can never leave the sourced band. The weighting is ours; the band is not.
Low-ambient capacity and draw
Cold Climate certification minimum at 5 F = nameplate × 0.7 from the certification floor, and draw = capacity / (COP × 3.412142) using the watt-to-BTU identity. Neither step involves a judgement call.
Running cost
draw in kW = rated BTU/h / (SEER2 or HSPF2 × 1000), then cost per 1,000 hours = draw × 1000 × state price per kWh. Both inputs are published; there is no assumption in this pair, which is why the site leads with it.
The assumptions, named individually
These are the 6 figures on this site that no source backs. Each one is disclosed at the point it is used, not only here.
| Figure id | Value | What it does and why we could not source it |
|---|---|---|
m1.envelope.tight | 0.85 | Envelope multiplier on the sizing table. The published table carries no insulation term at all, and we could not obtain a residential multiplier we were willing to cite. |
m1.envelope.average | 1.00 | Envelope multiplier on the sizing table. The published table carries no insulation term at all, and we could not obtain a residential multiplier we were willing to cite. |
m1.envelope.leaky | 1.20 | Envelope multiplier on the sizing table. The published table carries no insulation term at all, and we could not obtain a residential multiplier we were willing to cite. |
m10.cooling_hours_per_cdd | 0.80 | Converts a state cooling degree-day total into equivalent full-load run hours. Utility Technical Reference Manuals publish these; we could not obtain one at a stable public address. |
m10.heating_hours_per_hdd | 0.30 | The same conversion on the heating side, with the same missing source. |
m10.uncertainty_fraction | 0.35 | The band we place around any annual figure, chosen to reflect how little confidence the run-hour assumption deserves. |
Two further assumptions are structural rather than numeric and belong in the same list. The oversizing caution fires above 25 percent overshoot — a threshold we chose. The open-plan rule collapses exactly 1 head when you say the main space is open, which is the conservative version of a rule with no published form.
Our own labeled adjustment. The ENERGY STAR table has no insulation term; a room whose envelope is much better or worse than typical will sit outside these multipliers. If a contractor has run a Manual J on your house, that number beats this one.
Our own labeled assumption converting a state's degree days into equivalent full-load run hours. We could not obtain a citable utility Technical Reference Manual hours table. If your utility publishes one, its number beats ours. The cost-per-hour figures on the same page use no assumption of any kind.
What this site refuses to price, and why that is not an oversight
Line set beyond the standard run. The source states the direction only. We could not obtain any published dollars-per-foot price for line set beyond the standard run from a source we were willing to cite, so this site prints none. A long run moves the estimate toward the top of the sourced band and is listed as an itemise-it-in-writing question for the contractor.
A main panel or service upgrade. Unresolved source conflict. This site does not pick a winner and does not put a panel upgrade in any total. Both figures appear on the permits and electrical page so you can see the disagreement rather than take our word for it.
Permit fees outside one city. Permit schedules are municipal, published as PDFs, and structured differently everywhere. We publish the one we read. Adding a national average would be inventing a number that describes nowhere.
An annual run-cost figure for Alaska, Hawaii or the District of Columbia. Their electricity prices are published; their statewide degree days are not, in the source we use. The per-hour figures work; the annual estimate does not exist for them.
State and utility rebates. Accurately tracking fifty states of programme rules is a full-time job and a stale rebate figure does real financial damage. We track the federal position, which is a single document, and say plainly that we do not track the rest.
Limitations you should hold against every number here
The sizing table is a table. It knows your floor area and, through its adjustment rules, a little about sun and occupancy. It does not know your window areas, your orientation, your infiltration rate or your neighbours’ shade. A room-by-room load calculation run on your actual house beats it, and beats this whole site, every time.
The installed-cost figures are national survey data from commercial publishers, one of which describes its method as surveying cost databases and the other as tracking estimates homeowners report. Neither is your local market, neither is audited, and their underlying article dates are older than their retrieval dates — one is an article last verified in December 2024. That lag is real and is the reason the cost pages talk in bands rather than in prices.
The electricity prices are statewide averages of what residential customers actually paid, published as preliminary estimates and subject to revision. If you are on a time-of-use tariff your marginal evening rate can be far above the average, and a heat pump running through a cold evening sits squarely in that window.
The degree days are one calendar year, population-weighted, statewide. A single year is not a normal, and a statewide figure averages a mountain county with a coastal one.
Corrections are welcome and are the fastest way to improve this. If a figure here is wrong, or a source has moved, the contact page explains what to send and the editorial policy explains what happens next.
Generated from the data module at build time. Figures on this page: 108. Freshness horizon: 183 days from retrieval.