Review scores cluster near the top, and the reason is structural

The theory of compression in review scoring is well covered elsewhere. This is about the version you meet in practice.

Why the bottom of the scale is empty

Publications review products that manufacturers send or that readers are likely to buy, which excludes most of the genuinely poor ones. A product bad enough to score three is usually not reviewed at all, since coverage is a scarce resource spent on plausible candidates.

The result is a sample already filtered for adequacy, so the observed range of scores is narrow before anyone judges anything. Readers then interpret a seven as poor, because they have learned that scores cluster high, which pushes the effective floor upward again. This feedback loop is well recognised inside publishing and is rarely explained to readers.

The costs of a low score

Manufacturers control access to review samples, and a publication that scores harshly may find samples arriving late or not at all. Advertising relationships create a further pressure that need not be explicit to be effective. Readers punish negative reviews of products they own, which affects engagement metrics that publishers watch.

On the bench, none of these forces requires a corrupt decision, and each shifts the distribution slightly upward. The honest reviewers acknowledge the pressure openly, and that acknowledgement is itself a useful signal.

What a score cannot encode

A single number collapses several incompatible dimensions, and the weighting between them is the reviewer's rather than yours. Value is usually folded into the score, which means a score changes meaning when the price changes and is rarely updated. Durability cannot be scored at launch, because it has not happened yet, so it is either omitted or guessed.

Judged against the category, suitability for a particular use is exactly what a general score cannot express. This is why the text of a review almost always contains more information than the number attached to it.

Reading the text instead

Look for specific observations with measurements, comparisons or described procedures, since these survive independent of the score. Look for the criticisms, which are usually accurate even in generous reviews, because reviewers rarely invent faults.

Note what the reviewer did not test, particularly anything requiring long-term use, which almost no launch review covers. Note whether the reviewer describes their own use case, since that determines which observations transfer to you.

A review that says clearly what it could not evaluate is more trustworthy than one that appears to have covered everything.

Aggregate scores and their own problems

Averaging scores across publications inherits every source's compression and adds a false impression of precision. Different publications score to different distributions, so averaging them mixes incompatible scales.

Aggregators weight sources in ways that are rarely published and that materially change the result. A narrow range across many sources tells you the product is unremarkable rather than that it is excellent. The dispersion between reviews is often more informative than the average, since disagreement points at a genuine trade-off.

Manufacturer figures are measured under conditions the manufacturer chose.

Using reviews without the numbers

Read three or four full reviews and extract the specific claims rather than the conclusions. Prefer publications that publish their test methodology, since a stated method can be criticised and an unstated one cannot. Prefer comparative testing done at one time by one team, which controls variables that separate reviews cannot.

Look for follow-up coverage months later, which is where durability information first appears. This site does not score products and does not test them, which is why it explains mechanisms and points you to the people who do.

The takeaway

Read the observations and ignore the number, because the number was compressed before anyone wrote it.

Buy for the failure you can live with, not the feature you will use twice.