1. Never invent a number
Part numbers, dimensions, intervals, page yields and setup codes are researched from a citable source or they do not ship. Where no citable source exists, we say so on the page and narrow what the tool claims — we do not fill the gap with something plausible. Two tools on this site deliberately answer a smaller question for exactly this reason, described on the about page.
2. Separate what is cited from what is judgement
Some of what a tool does is a quoted standard; some is our own modelling on top of it. We keep those apart in the text. The dehumidifier matrix is ENERGY STAR's, quoted verbatim — the basement and climate multipliers applied to it are our judgement, and the page says so. Every data file carries a confidence note recording which is which.
3. Show the table
Each tool renders the reference data it used, in full, on the page. If our answer looks wrong to you, the material to check it against is right there. We consider a result you cannot audit to be worth less than one you can.
4. Report uncertainty as uncertainty
Where a dataset is sparse we publish the sample size next to the figure. Where a range is genuinely a range, we give the range instead of a falsely precise midpoint. Where an assumption drives the answer — an air-change rate, a usage pattern — we name it.
5. Prefer primary sources
Government datasets and manufacturer publications come before aggregators, and aggregators before commentary. When we compute from a raw dataset we say which dataset, which resource, how we filtered it, and when we pulled it, so the figure can be reproduced.
6. Date everything and re-check it
Every tool shows when its data was last reviewed. Reference data decays; an undated figure is an unverifiable one.
7. Monetisation never touches the answer
The result is computed from the reference data before any commercial link is generated. Affiliate links render after the answer, never above it, and always carry a disclosure. No retailer pays for placement, inclusion, or a recommendation, and no tool has ever been shaped to sell something.
8. Synthesise, never republish
Where we draw on a manufacturer's published material we compile and summarise it with attribution. We do not reproduce a source's text at length or mirror an entire published table.
9. Correct in the open
When we get something wrong we fix the data and move the review date. Reports from readers are the fastest way we find errors — see contact.