Review Methodology

Last updated: August 17, 2026

Every number on this site comes from a published manufacturer specification, and every one of them has to pass two checks before it appears. This page sets out both checks, what happens to the figures that fail, and the claims we will not make regardless of what the data says.

1. Where the figures come from

Specifications are collected from retail product listings across each category, at a scale intended to describe the category rather than a selection from it. Each extracted value is stored alongside the exact source text it was read from, so any published figure can be traced back to the words that produced it.

That is the only source. There is no test tank, no instrumented rig and no long-run reliability data behind any page. Where a manufacturer’s own documentation disagrees with a retail listing, the manufacturer is the better source and should be treated as authoritative over anything here.

2. The coverage gate

A specification is published for a category only where it appears on at least 40% of that category’s products and on at least eight products absolutely.

Both conditions are needed. Eight values out of twelve products is a description of a category; eight out of two hundred is a description of which brands fill in forms. Where a field misses the gate it is listed as withheld with its real coverage shown, so the page distinguishes between a figure we chose not to publish and one the data cannot support.

3. The unit gate

Retailers file different physical quantities under the same specification slot, and an extractor that keeps the number while discarding the unit will produce a figure that is confidently wrong.

So every extracted value is re-read against its source text, and the unit in that text is checked against the units the field can legitimately hold. Where fewer than 80% of a field’s values survive, the field is dropped for that category outright — not published with a caveat.

4. Quartiles, not averages

Comparison pages publish the 25th percentile, the median and the 75th percentile. They do not publish means.

The reason is that every contamination pattern found in retail specification data so far lands in the upper tail — a capacity stated in the smaller unit, an accessory listed as a device, a package dimension recorded as a product dimension. A mean is maximally sensitive to those values; a median is barely moved by them. Where a maximum in our data is plainly not a real product, we say so next to the table rather than trimming it silently.

5. What we refuse to publish

6. “Not stated” is a real value

Where a specification is absent from the source, the correct entry is “Not stated”. It is never a plausible default.

This is the most common way specification content goes wrong: a column exists, most rows fill it, and the remainder quietly receive whatever the category typically has. The output reads as data and is invention. Tables on this site carry “Not stated” wherever the record is silent, and a column with several of them is telling you something true about how the category documents itself.

7. Category premise checks

Before a category page is written we confirm the corpus can actually supply it. A category whose signature specification is not published at a usable rate gets no comparison page, and the finding is stated on the hub. Writing a thin table would imply coverage that does not exist.

8. Updates and corrections

Each page shows the date its figures were last refreshed, and a page is a snapshot rather than a live feed. Corrections are made in place with the date changed rather than silently. Report anything that looks wrong to editorial@aquariumtechdata.com, naming the page and the product.