This page describes what our process actually does. It is deliberately specific about its limits.
1. Primary sources only. When we publish a number attributed to an agency or a manufacturer, we open that source and link to it in the same sentence. We do not write “studies show” or “according to reports”. If we cannot find the primary source, the number does not go in.
2. Source text is stored, not just the value. For our reference datasets we keep the sentence a figure was read from, alongside the figure itself, and publish both in the downloadable file. A transcription error then has to be visible in the quoted sentence rather than hiding as a plausible-looking number.
3. Published totals are checked against their own columns. Where a source prints a total, we recompute it. Where the two disagree, we say so on the page and show our working rather than silently substituting our own figure.
4. Specifications are normalised before comparison. Capacity, output, surge, and weight are converted to consistent units so that two products can be compared on the same axis. Where a manufacturer publishes a figure under conditions that are not comparable, we note the condition instead of dropping it into a table as if it were equivalent.
5. Model status is tracked. Products move through current, superseded, and discontinued. A model that is no longer current does not appear as a recommendation.
6. Prices are given as ranges. Street prices in this category move by thirty to forty per cent around sales events. A precise dollar figure would be wrong within weeks, so we describe bands and say when a figure was checked.
7. Independent review before publication. Reference datasets are reviewed by someone other than the person who assembled them, against two separate questions: is every figure traceable, and would an outside editor be willing to cite this. Both have to pass.
What this process cannot tell you. We do not measure runtime, noise, inverter behaviour under load, or real-world charge cycles. Those require physical testing we do not perform. Where that evidence matters for a decision, we say that it is missing rather than filling the gap with an impression.