Across most review outlets, the usable range of a ten-point scale is roughly the top third. The compression is systematic and has several independent causes that reinforce each other.
Genuinely bad products rarely reach review
Reviewers cover what an audience is interested in, and audience interest tracks products people are considering buying. Obscure and poor products are simply not covered.
Manufacturers also decline to supply units for products they expect to be criticised, which removes another band of the range before any scoring happens.
The pool being scored is therefore already filtered toward the competent. A scale applied to that pool cannot use its lower half.
Scores are read comparatively, not absolutely
Readers do not interpret a score against an abstract standard. They interpret it against other scores for similar products, most of which sit high.
An outlet that scores accurately on the full range produces numbers that look hostile in that context, and its scores get misread as condemnations.
Comparability pressure pushes every scale toward the same compressed band, because a number that cannot be compared is not doing the job a number is there to do.
The number is used as a summary it cannot support
A score aggregates performance, value, build and software into a single figure, and the weighting is rarely stated. Two reviewers can reach the same number by different routes.
Aggregation sites then average those figures across outlets with incompatible weightings, producing a precise-looking result built from incomparable inputs.
Because a low score travels further than the reasoning behind it, reviewers become cautious about issuing one, which tightens the band again.
Category inflation moves the whole scale
Scores are usually assigned relative to what is currently available rather than to a fixed standard. As a category improves, the same product would score lower today than at launch.
That makes scores non-comparable across years, though they are routinely compared that way. An old high score reflects an old field.
Outlets that re-score periodically avoid this and rarely do so, because retroactively lowering a score is unpopular with readers and manufacturers alike.
What replaces the number when it fails
Several outlets have dropped numerical scoring in favour of recommendation categories, or of naming the specific buyer a product suits and the one it does not.
That format resists compression because it is not a scale. A product can be strongly recommended for one use and dismissed for another without contradiction.
It also forces the reviewer to state the conditions of the verdict, which is the information a reader needed from the score and almost never received.